Issuu on Google+


Editorial Board ISSN: 2218-6026 (Print)

ISSN: 2304-3679 (Online)

http://www.ijesci.org

﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡ Editor-in-Chief Prof. José Antonio Orosa Garcia, the University of A Coruña, Spain Editorial Board Dr. Yinjie Tang, Washington University, USA Dr. Gang Quan, Florida International University, USA Dr. Hansong Tang, The City University of New York, USA Dr. Linxia Gu, University of Nebraska, USA Dr. Shun-Chung Lee, National Cheng Kung University, Taiwan Dr. Qais H. Alsafasfeh, Tafila Technical University, Jordan Dr. Lukumon O. Oyedele, Queen's University Belfast, UK Dr. Berhan Ahmed, Melbourne University, Australia Dr. Ji-Hyoung Ryu, Chonbuk National University, Korea Dr. Alireza Bahadori, Curtin University, Australia Prof. Jamal Mahmoud Nazzal, Jordan Cooperation Group, Jordan Prof. Qifeng Zhang, University of Washington, USA Prof. Yu Bo, China University of Petroleum, China Dr. Ali Z. Hamadani, Isfahan University of Technology, Iran Dr. Dragos Isvoranu, Polytechnic University of Bucharest, Romania Dr. Chao Xu, Chinese Academy of Sciences, China Prof. Pawan Tyagi, University of the District of Columbia, USA Prof. Zvonimir Glasnovic, University of Zagreb, Croatia Dr. Nakorn Tippayawong, Chiang Mai University, Thailand Dr. Karmen Margeta, University of Zagreb, Croatia Dr. Qingzhao Wang, University of Florida, USA Dr. Wojciech M. Budzianowski,Wrocław University of Technology, Poland Dr. Yongfu Huang, United Nations University World Institute, Finland Dr. Zuhdi Hamdi Salhab, Palestine Polytechnic University, Palestine Dr. Bindeshwar Singh, Kamla Nehru Institute of Technology, India Dr. Messaouda Azzouzi, "Ziane Achour" University of Djelfa, Algeria Dr. Vijay Kumar Thakur, Nanyang Technological University, Singapore Dr. Sanjeev Kumar Aggarwal, M.M. Engineering College, India Dr. Xiao-Sen Li, The Chinese Academy of Sciences, China


﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡﹡

TABLE OF CONTENTS Volume 2, Issue 6 December 2012 Solar Radiation Forecast Using Artificial Neural Networks Fernando Ramos Martins, En io Bueno Pereira, Ricardo André Guarn ieri………………………………………..217

Air Cells Using Negative Metal Electrodes Fabricated by Sintering Pastes with Base Metal Nanoparticles for Efficient Utilization of Solar Energy Taku Saiki, Takehiro Okada, Kazuhiro Nakamura, Tatsuya Karita, Yusuke Nishikawa, Yukio Iida…228

Blends of Diesel – used Vegetable Oil in a Four-Stroke Diesel Engine Charalampos Arapatsakos .……………………………………………………………………………………………………………….235

Catalytic Pyrolysis by Heat Transfer of Tube Furnace for Produce Bio-Oil Kittiphop Pro mdee, Tharapong Vitidsant.............................………………………………………………………………..241

CMOS Bandgap Reference and Current Reference with Simplified Start-Up Circuit Guo-M ing SUNG, Ying-Tzu LAI, Chien-Lin LU………..……………………………………………………….…………247

Transient Analysis of Three-Phase Self Excited Induction Generator Using New Approach Vivek Pah wa, K. S. Sandhu………………………………………………..……………………………………………………………255

On the Sensitivity of Principal Components Analysis Applied in Wound Rotor Induction Machines Faults Detection and Localization J. Ramahaleo miarantsoa, N. Heraud, E. J. R. Sambatra, J. M . Razafimahenina…………..……………………262

Evaluation of the Quality of Service Parameters for Routing Protocols in Ad-Hoc Networks Zeyad Ghaleb Al-Mekhlafi, Rosilah Hassan, Zurina Mohd Hanapi………………………………………...……….272

An Investigation of Power Performance of Small Grid Connected Wind Turbines under Variable Electrical Loads Md. Alimu zzaman, M.T.Iqbal, Gerald Girou x.………………………………………...……….…………………………….282


International Journal of Energy Science (IJES) Journal Information IJES is published bimonthly by Science and Engineering Publishing Company. This journal is a multi-disciplinary focus on activities relating to the development, assessment and management of energy-related programs. As a professional journal, it aims at becoming a first-class international journal. It is hoped that this publication will prove to be an important factor in raising the standards of discussions, analyses, and evaluations relating to energy programs. The benefits to publish your papers in this journal include: ※RAPID PUBLICATION: Manuscripts are peer-reviewed and published within 50 days, and accepted papers are immediately published online. ※INDEXING SERVICES: All the published papers will be sent to be indexed by any third parties, including Google Scholar, Scirus, INSPEC, etc. ※REFERENCE SERVICES: All the published papers will be sent to the related researchers for potential reference by email.

C COPYRIGHT ○ Copyright ©2012 Science and Engineering Publishing Company (website: www.seipub.org). Address: 7800 STATE ROAD 46 E, PO Box 551, Riley, Indiana, 47871 All Rights Reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise, except as described below, without the permission in writing form of the Publisher. Copying of articles is not permitted except for personal and internal use, to the extent permitted by national copyright law, or under the terms of a license issued by the national Reproduction Rights Organization. Request for permission for other kinds of copying, such as copying for general distribution, for advertising or promotional purposes, for creating new collective works or for resale, and other enquiries should be addressed to the Publisher. Statements and opinions expressed in the articles and communications are those of the individual contributors and not the statements and opinion of Science and Engineering Publishing Company. We assumes no responsibility or liability for any damage or injury to persons or property arising out of the use of any materials, instructions, methods or ideas contained herein. We expressly disclaim any implied warranties of merchantability or fitness for a particular purpose. If expert assistance is required, the services of a competent professional person should be sought.

Pricing: Personal subscription: $100


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Solar Radiation Forecast Using Artificial Neural Networks Fernando Ramos Martins, Enio Bueno Pereira, Ricardo André Guarnieri Center for Earth System Science Brazilian Institute for Space Research São José dos Campos, Brazil. 12227-010 fernando.martins@inpe.br Abstract The fast increase in importance of the solar e nergy resource as viable and promising source of rene wable ene rgy has booste d research in me thods to e valuate the short-term forecasts of the solar e nergy resource. There is an increase on de mand from the e nergy sector for accurate short-term forecasts of solar e nergy resources in orde r to support the planning and manageme nt of the e le ctricity gene ration and distribution systems. The Eta mode l is the mesoscale mode l running at CPTEC/INPE for weather forecasts and climate studies. It provide s outputs for solar radiation flux at the surface , but the se solar radiation forecasts are greatly ove restimate d. In orde r to achie ve more re liable information, Artificial Ne ural Ne tworks (ANN) we re use d to re fine shortterm forecast for the downward solar radiation flux at the surface provide d by Eta/CPTEC mode l. Ground measure me nts of downward solar radiation flux acquire d in two SONDA site s locate d in Southe rn region of Brazil (Florianópolis and São Martinho da Se rra) we re use d for ANN training and validation. The short-te rm forecasts produce d by ANN have prese nte d highe r corre lation coe fficie nts and lowe r de viations. The ANN re move d the bias observe d in solar radiation forecasts provide d by Eta/CPTEC mode l. The skill improve me nt in RMSE was highe r than 30%whe n ANN was use d to provide short-term forecasts of solar radiation at the surface in both measure me nt sites. Keywords Solar Energy Forecast; Short-Term Forecast; Artificial Neural Network; Energy Meteorology

Introduction The scientific community points out that the fossil fuel expenditure is the major reason of the observed growth of the greenhouse gases concentrations in atmosphere along the last century [1]. Developed countries and advanced economies have been charged for the environmental damages due to consumption of conventional energy sources to meet their energy demand. However, emerging economies such as Brazil,

India, China, and Russia are increasingly sharing this responsibility as a result of their growing demand for energy to support their fast growing economic development The commitment to reduce the emissions of carbon dioxide (and other greenhouse gases) established at the Kyoto Protocol and the perspectives of oil depletion in next decades are key factors to boost the research and development on alternatives and renewable energy sources such as solar and wind [2, 3]. Furthermore, the search for improvement on energy security has been driving the government policies and incentive programs to stimulate the employment of alternative renewable energy sources even in countries with large share of clean energy in their electricity generation matrix. For example, in Brazil, where hydroelectric energy is responsible for more than 70% of the electricity matrix, an energy shortage happened in 2001 due to very low precipitation during the wet season of the previous year [4]. After this event, Brazilian government created incentive programs for renewable energy sources like wind energy. The solar energy is one of the promising alternatives in Brazil since most of its territory is located in the intertropical region where solar energy resources are accessible all year round [5]. The main obstacles to the commercial exploitation of solar energy resources are the highest cost compared to the conventional electricity generation technologies, lack of information on resource assessment and variability, and the deep dependency on the weather and climate conditions [4]. The investment costs are expected to fall during the next decades due to technological advances and market demands [6]. The growing market for solar energy leads to an increase on the demand for more reliable information concerning to solar resources, including its spatial and temporal variability in short and long terms.

217


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

In addition, the management of electricity generation and distribution systems is also asking for more accurate short-term solar energy forecasts. Several methodologies were developed in order to provide solar radiation forecast in high temporal resolutions and short-term horizons [7, 8]. Some of them use numerical weather models (NWP). Such models have radiation parameterization codes to simulate the radiative atmospheric processes. Nevertheless, solar irradiation forecasts provided by NWP models for one or two days in advance have shown large deviations from solar irradiation data acquired at surface [9]. The major factors responsible for such deviations are related to the solar irradiation dependence on clouds and weather conditions which intrinsically involve non-linear physical processes [10]. Absorption and scattering interactions are the atmospheric radiative processes that attenuate the solar radiation flux. Therefore, the atmospheric optical properties should be known in order to correctly evaluate the solar irradiation at any specific site and time. Clouds are the main factor that modulates the solar radiation incidence at the surface [11, 12, 13, 14, 15]. Atmospheric aerosols also have an important role in atmospheric radiative processes, mainly in some regions where anthropogenic emissions from biomass or fossil fuel burning takes place. The Eta/CPTEC mesoscale model runs operationally in the Center of Weather Forecast and Climate Studies at Brazilian Institute for Space Research (CPTEC/INPE) and provides short-term forecasts for many meteorological variables, including surface solar irradiation. However, the references [11] and [12] showed that Eta/CPTEC model systematically overestimates the surface solar irradiation, as well as the sensible and latent heat fluxes at surface. A common issue in numerical atmospheric radiation codes is the excess of the incoming shortwave radiation at the surface as a result of the deficient parameterization of extinction interactions with water vapor, atmospheric aerosols and clouds. Several methodologies were published in order to improve solar forecasts provided by numerical weather models [9, 16, 17, 18]. This work aims to present a methodology to reduce deviations of solar irradiation forecasts provided by Eta/CPTEC model by using a statistical post-processing applied to the model outputs. This paper presents the results obtained when Artificial Neural Networks

218

(ANN’s) were used as statistical tool to refine the solar radiation forecast provided by Eta/CPTEC model. Artificial neural networks (ANN) are data-driven instead of model-driven techniques once the results provided by them depend on the available data used to feed the ANN. Relationships between predictors (input data) and predictions are developed after building a system which simulates the physical processes in atmosphere. Artificial neural networks have been applied in renewable energy research for modeling and design solar systems and to provide short-term forecasts for energy resources [19]. Reference [20] indicated that the ANN systems are able to predict the solar radiation time series more effectively than the conventional procedures based on the clearness index. The authors observed that the forecasting ability can be further enhanced with the use of additional meteorological parameters like temperature and wind direction. References [21] and [22] discussed different methodologies using ANN to provide short-term forecasts for solar radiation by extracting knowledge from a long ground data series. Reference [23] compared some statistical models and ANN systems using meteorological data as input data. The authors concluded that ANN systems were a promising alternative to the traditional approaches for estimating global solar radiation, especially in cases where solar radiation measurements are not readily available. This paper presents an attempt to get better predictability for the solar energy resources using operational Eta/CPTEC model and it constitutes an important application of the meteorology science to the energy planning and decision-making processes in energy sector. The target is to provide more precise and reliable information on future availability of solar resources in order to optimize electricity generation and distribution systems. Methodology Forecasting solar irradiation depends on prospecting the future atmospheric conditions. Despite the intrinsic uncertainties, NWP models provide information about many meteorological variables, including solar radiation data and atmospheric optical properties for several future timeframes. However, earlier studies demonstrated that solar radiation data provided by such models presents a large bias making its use inappropriate to electricity system management where several solar power plants are connected [10, 16, 17, 18].


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

This work employed the weather forecast outputs provided by the Eta/CPTEC model together with environmental data to feed Artificial Neural Network (ANN). The main goal was to achieve a short-term forecast for solar irradiation with lower deviations than the ones provided by the Eta/CPTEC model. The solar radiation data acquired in two SONDA ground sites located in the Southern region of Brazil was used as reference for training and performance evaluation of the ANN. Model Eta/CPTEC The Eta/CPTEC model is used for operational weather forecasting, climate investigation, regional climate change studies and research on several issues like pollutant transport [24]. The Eta model, which has been running at CPTEC since 1996, was set up and optimized to the South America atmospheric conditions. The Eta/CPTEC model runs routinely for South America continent and neighboring oceans: latitudes from 50.2ºS to 12.2ºN, and longitudes from 83ºW to 25.8ºW. The horizontal resolution equals to 40km and 38 vertical layers were used for this study. The Eta/CPTEC model employs the “finite difference” scheme to solve the equations system that describes the physical processes in atmosphere. The model uses the vertical coordinate “Eta”, η, defined as: (1) where pt is the pressure at the top of the model atmosphere, pref is the reference pressure to the vertical profile, and psfc and zsfc are the pressure and height of the lower boundary surface, respectively. The Eta coordinate was adopted to reduce the large errors observed in several numeric weather forecast models that use the sigma surfaces [12]. These deviations are related to the determination of the horizontal pressure gradient force, as well as the advection and the horizontal diffusion on a steeply sloped coordinate surface [25, 26]. The discretization of the space domain uses the SemiStaggered Arakawa E-grid on the horizontal and the Lorenz grid on the vertical. The radiation modeling uses the schemes described in [27] for shortwave radiation, and in [28] for long wave radiation. More detailed descriptions about the physical parameterizations adopted in Eta/CPTEC model can be found in [26, 29, 30, 31].

The Eta/CPTEC model was executed using initial conditions at 00UT provided by NCEP analyses. The CPTEC Atmospheric Global Circulation Model (AGCM) provided the lateral boundary conditions. The outputs provided by Eta/CPTEC model for 2001 till 2005 were used. The output file contains forecasts for 58 atmospheric variables at the synoptic timeframes (6, 12, 18 and 24UT) for 7 days in advance. The model provided the total amount in atmospheric column for forty-nine variables, and vertical profile values at 19 atmospheric pressure levels for the remaining nine variables. Only 33 out of the 58 atmospheric variables were used in this study. All vertical profile data were discarded together with 16 variables not representative of the atmospheric condition like topography, soil temperature and humidity for levels under surface. Table I presents a complete list of model output data used for this work with a short description of them. Instantaneous values at each synoptic time were recorded for most of the data. However, average values regarding to the 6-hour period before each synoptic time were stored for some of the meteorological output variables, such as “ocis”. SONDA network SONDA (Brazilian System for Environmental Data applied to the Energy Sector) is a network of ground measurement sites, operated and managed by INPE. The goal is to acquire reliable surface solar irradiation and wind data at different climate areas in Brazil in order to develop, improve and validate numerical models used for renewable energy resources assessment and environmental research. The SONDA database will provide valuable information applied to the research on the energy meteorology in Brazil. In this work, the SONDA ground data acquired at two SONDA sites was used for the ANN training and configuration as described later in this paper. Besides that, ground data were used to evaluate the deviations presented by short-term forecast provided by both methodologies: Eta/CPTEC model and ANN. Both measurement sites were located in the Brazilian Southern region: São Martinho da Serra (SMS) – 29.44ºS/53.82ºW. Florianópolis (FLN) – 27.60ºS/48.52ºW; Fig. 1 shows the location of measurement sites of SONDA network featuring SMS and FLN sites. These both sites were chosen in order to evaluate the performance of ANN and Eta/CPTEC model in two

219


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

different climate conditions. SMS is located in the continental area at 500m above the sea level. FLN is located at the coastal area of Brazilian Southern region presenting the largest total precipitation along the year in Brazilian territory. The SMS has been collecting data since June 2004 and FLN has been acquiring data since 1995.The other SONDA sites are more recent and have smaller databases. The SONDA website (http://sonda.ccst.inpe.br) presents all information about measurement sites and describes the data quality assurance program. For this work, data acquired from January/2001 to October/2005 in FLN and from July/2004 to October/2005 in SMS were used. The Kipp&Zonen CM21 pyranometers [32] were used to acquire global solar irradiation data. One-minute average solar irradiation data was stored and its quality was checked. Both sites take part in Baseline Solar Radiation Network (BSRN) and meet all the quality criteria established by World Meteorological Organization (WMO).

The training group was used for the ANN training. The validation group was employed to evaluate and establish the end of the training step. The investigation group was used to evaluate the reliability of ANN outputs. More details on each these three steps are described latter in this paper. Data Management As explained earlier, the solar and meteorological database used to feed ANN comprises the output data provided by the model Eta/CPTEC (Table I). In addition, other three variables were calculated in order to supply ancillary information for the ANN: solar radiation flux at TOA (STOA), mean air mass (airm), and mean solar zenith angle (szam). Altogether, 36 variables were used as ANN predictors. As described on Table I, the solar irradiation data provided by the Eta/CPTEC model, “ocis”, represents the 6-hour average solar irradiation. In order to achieve the same time-scale, the solar irradiation data acquired in FLN and SMS sites were averaged over the same 6hour intervals. In summary, ground and model data of solar irradiation represents the total energy in the 6hour period and they are expressed in MJ.m -2 (mega joules per squared meter). The 6-hour average solar radiation flux at the top of the Earth’s atmosphere (STOA) was calculated taking into consideration local latitude, solar zenith angle, eccentricity and solar declination [13, 14]. As the ground solar irradiation data and “ocis”, the STOA solar radiation flux was also expressed in MJ.m -2.

FIGURE 1 LOCATION OF GROUND S ITES OF SONDA NETWORK. FLORIANÓPOLIS AND SÃO MARTINHO DA SERRA WERE USED FOREVALUATION OFS HORT-TERM FORECAS TS.

After data-quality verification, 1150 days for FLN and 472 days for SMS were available for this work. The ground database was divided into 3groups as follows: Training group: with 575 days for FLN and 236 days for SMS; Validation group: with 288 days for FLN and 118 days for SMS; Investigation group: with 287 days for FLN and 118 days for SMS.

220

Relative humidity, atmospheric pressure, air temperature, wind velocities and all other instantaneous data, provided by Eta/CPTEC model for synoptic time (Table I),were averaged by taking the two consecutive values. The averages were assigned to the second synoptic time in order to set up the database in a similar way used for ground data. This procedure aims to better represent the atmospheric and meteorological variability in the 6-hour interval. In addition, the solar zenith angle (szam) and the air mass (airm) were obtained and stored for the same 6hour intervals. Thus, the “ocis” data and all 36 variables used to feed ANN have the same temporal resolution and represent the equivalent timeframes. The 36 predictors and the ground data are disposed into four timeframes each day: 6:00, 12:00, 18:00 and 24:00UT. Each timeframe represents the corresponding


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

time interval: 0-6UT (Rad06UT), 6-12UT (Rad12UT), 1218UT (Rad18UT), and 18-24UT (Rad24UT). This paper only presents results for the Rad18UT timeframe. The

Rad18UT was chosen because the highest fraction (63% – 80%) of solar radiation flux occurs during the 12-18UT intervals throughout the year at both ground sites [35].

TABLE 1 THE METEOROLOGIC DATAUSED AS PREDICTORS IN ANN. ALL DATA WAS PROVIDED BY MODELETA/CPTEC VARIABLE

DESCRIPTION (UNITS)

KEY FEATURES

rh2m

Relative humidity at 2m-height (0 to 1 – adimensional)

Instantaneous values

pslc

Pressure at surface (hPa)

Instantaneous values

tp2m

Temperature at 2m-height above the surface (K)

Instantaneous values

dp2m

Dew Point Temperature at 2m above the surface (K)

Instantaneous values

u10m

Zonal wind at 10m-height above the surface (m s -1)

Instantaneous values

v10m

Meridional wind at 10m-height above the surface (m s -1)

Instantaneous values

wnds

Wind velocity at 10m-height above the surface (m s -1 )

Instantaneous values

prec

Total rainfall (kg m-2 dia-1)

Total in the 6h period

prcv

Convec tive rainfall (kg m-2 dia-1 )

Total in the 6h period

prge

Large sc ale rainfall (kg m-2 dia-1)

Total in the 6h period

c lsf

Latent Heat Flux at the surface (MJ m-2 ) m-2 )

Average value in the 6h period

cssf

Sensible Heat Flux at the surface (MJ

Average value in the 6h period

ghfl

Heat Flux in the soil (W m-2)

Average value in the 6h period

tsfc

S urfac e Temperature (K)

Instantaneous values

qsfc

S pec ific humidity at the surface (kg(H2O) kg(air) -1)

Instantaneous values

lwnv

Cloud cover Index for low c louds (0 a 1 - adimensional)

Instantaneous values

mdnv

Cloud cover Index for average c louds (0 a 1 - adimensional)

Instantaneous values

hinv

Cloud cover Index for high c louds (0 a 1 - adimensional)

Instantaneous values

c bnt

Mean Cloud cover Index (0 a 1 - adimensional)

Instantaneous values

ocis

Downward shortwave radiation flux at the surface (MJ m-2)

Average value in the 6h period

olis

Downward longwave radiation flux at the surface (MJ m-2)

Average value in the 6h period

oces

Upward shortwave radiation flux at the surfac e (MJ m-2)

Average value in the 6h period

oles

Upward longwave radiation flux at the surface (MJ m-2 )

Average value in the 6h period

roce

Upward shortwave radiation flux at the TOA (MJ m-2 )

Average value in the 6h period

role

Upward longwave radiation flux at the TOA (MJ m-2)

Average value in the 6h period

albe

Albedo (0 a 1 - adimensional)

Instantaneous values

c ape

Available potential convec tive energy (m2 s -2 )

Instantaneous values

c ine

Energy to avoid convec tion (m2 s -2)

Instantaneous values

agpl

Instantaneous precipitable water amount (kg m-2)

Instantaneous values

pc bs

Pressure at the bottom of the c louds (hPa)

Instantaneous values

pc tp

Pressure at the top of the c louds (hPa)

Instantaneous values

tgsc

Soil temperature at the surface layer (K)

Instantaneous values

ussl

Soil humidity at the surface (0 a 1 - adimensional)

Instantaneous values

221


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Artificial Neural Networks (ANNs) Artificial Neural Networks (ANN) is computing systems, which attempt to simulate the structure and function of biological neurons. Generally, the ANN consists of a number of interconnected processing elements, called neurons. Fig. 2 presents an artificial neuron. The ANN usually consists of an input layer, some hidden layers and an output layer. Signals flow from the input layer through to the output layer via unidirectional connections (synapses). Synapses connect neurons of neighboring layers. The input data (x i) is weighted by values associated with each synapse (w ij), called synaptic weights. Knowledge is usually stored as a set of connection weights (presumably corresponding to synapse efficacy in biological neural systems). The activity level of a neuron (υj) is determined by summing up all its weighted values together with its bias (bj). The neuron output is a result from an activation function (φ(υj )). Generally, the activation function is a linear or hyperbolic-tangent function. The non-linear activation functions allow ANNs to simulate non-linearity behaviors and complex patterns [19]. The ANN architecture depends on the physical process, the training method and the kind of data that the neural network will simulate. The multi-layer perceptron (or feed forward ANN) is the most widely ANN architecture used in meteorological topics [23]. A schematic diagram of typical multilayer neural network architecture is shown in Fig. 3. The input layer consists on one neuron for each input data (called predictor), and the output layer consists of one neuron for each simulated data (called predictant). The number of hidden layers and their total amount of neurons are not a priori established. There is no standard procedure to identify the best combination of neurons and layers. The most widespread training algorithm used for multilayer perceptrons is the back propagation algorithm [33]. In this work, we use a modified version of back propagation, called Resilient Back propagation or Rprop [34]. The validation dataset was employed to verify the performance of the ANN with an independent data sample – data not used in training process. This procedure allowed to check the generalization capacity achieved by the ANN along the training and to find out the appropriate moment to stop the training step in order to avoid overlearning. After

222

training, the weights and bias are fixed and the ANN is ready to be used in simulations. For this study, preliminary experiments revealed that better ANN performances were achieved using two hidden layers of neurons. These experiments were developed in two different situations. First, the 36 variables described earlier were used as input to the ANN; and, in the second situation, only a set of 8 out of the 36 input variables were used. Table II shows the best neurons distributions verified for each ANNmodel. On both cases, only one neuron is the output layer to provide information on solar radiation flux at surface. The number of neurons in the input layer is equal to the number of predictors used to feed ANN. The investigation dataset was used to evaluate the performance of ANN to provide reliable solar irradiation forecast. The next topic discusses the statistical parameters used to evaluate deviations of the ANN and Eta/CPTEC outputs and the skills of each model to provide reliable forecasts. TABLE 2NUMBER OF ARTIFICIAL NEURONS IN EACH ANN LAYER ANN-36p

ANN-8p

Input layer

36

8

First hidden layer

36

16

Second hidden layer

18

8

Output layer

1

1

ANN-36p – ANN using 36 variables a s predictors ANN-8p – ANN using 8 variables a s predictors

FIGURE 2 S YMBOLIC REPRESENTATION OF AN ARTIFICIAL NEURON AND ITS PARAMETERS.


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

N

R=

∑ ( F − F )(O − O) i =1

i

i

N

∑ (F − F ) i =1

i

2

(4)

N

∑ (O − O) i =1

2

i

In order to compare the performance of ANN and Eta/CPTEC model, the skill-score index was used as defined in eq. (5):

Skill ( Score, ref ) =

FIGURE 3SCHEMATIC DIAGRAM OF A FEEDFORWARD ANN USED IN THIS STUDY.

Statistical analysis of ANN and Eta/CPTEC outputs The outputs (forecasts – F) were compared with measured values (observations – O), and deviations between them (F - O) were calculated. The performance of the Eta/CPTEC and ANN models was checked with two statistical indices: mean error (ME) or bias, and root mean squared error (RMSE). ME values provide information about the systematic deviations of the forecasts indicating if the models overestimateor underestimate the actual solar irradiation at the two measurement sites. RMSE is a measure of how effectively the models predict ground observations. Since the deviations are squared, large deviations have greater contribution to RMSE. For this study, both ME and RMSE indices were normalized and expressed as percentage of the average solar irradiation in the two measurement sites, as shown in eq. (2) and (3).

Results and Discussion Initially, the Eta/CPTEC forecast and ground data for solar radiation flux were compared. As demonstrated in previous studies [10, 11], a significant positive bias (overestimation) was observed in the solar radiation flux provided by Eta/CPTEC model. Table III shows the performance scores obtained for Eta/CPTEC estimates using only the investigation dataset (N = 287 for FLN; N = 118 for SMS). Similar scores were obtained when complete dataset was used for comparison between model estimates and ground data. Based on these results, it was assumed that the investigation dataset are representative of the complete dataset. Since ANN performance must be evaluated using the investigation dataset, only the Eta/CPTEC performance scores using this dataset were considered from this point on. TABLE 3PERFORMANCE SCORESOBTAINED BY MODEL ETA/CPTEC

∑ (F − O ) i

i =1

∑ (O ) i =1

RMSE % = 100 ⋅

i

N

(5)

Score perf − Scoreref

where Score can be the ME% or the RMSE% values obtained for a particular model (Eta/CPTEC or ANN) in evaluation, Scoreref is the score calculated for a reference method and Scoreper f is the score value expected for perfect-forecast.

N

ME % = 100 ⋅

Score − Scoreref

%

(2)

Scores

(3)

where N is the number of data pairs (forecast and observation) used in the evaluation – 287for FLN and 118 for SMS. In addition, the Pearson’s correlation coefficient (R) was computed as described in eq. (4):

São Martinho da Serra

N =1150

N =287*

N =472

N =118*

R

0.747

0.720

0.790

0.775

R2

0.558

0.519

0.624

0.600

ME%

24.7%

24.6%

27.8%

28.0%

RMSE%

39.7%

40.0%

41.9%

43.2%

i

1 N ∑ ( Fi − Oi ) 2 N i =1 % 1 N ∑ (Oi ) N i =1

Florianópolis

* - results obtained using only the investigation dataset.

As previously mentioned, various statistical analysis and simulations were performed using different subsets of the predictors listed in Table I in order to find a reduced dataset of predictors which produces a performance similar to that obtained when all 36 predictors are used. These analysis point out a set of 8

223


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

predictors: solar radiation flux at TOA (STOA), relative humidity (rh2m), surface temperature (tsfc), precipitable water amount (agpl), zonal wind speed at 10 m height (u10m), and predictors for cloud fractions (cbnt, hinv and mdnv). Hereafter, the ANNs using 36 and 8 predictors will be called ANN-36p and ANN-8p, respectively. Table IV presents the performance scores obtained for ANN-36p and ANN-8p using the investigation dataset for both ground sites. As noticed, there is a very similar performance in terms of correlation (R) and RMSE deviations. However, the ANN-8p provided solar irradiation forecasts for both sites with 50% less ME than the ANN-36p. As noticed by comparing Tables III and IV, the ANN36p and ANN-8p provided solar irradiation forecasts presenting larger correlation with ground observations in both sites. The ANN-8p outputs presented the lowest systematic deviation while Eta/CPTEC forecasts showed the largest deviations (ME and RMSE) for both ground sites.

Meanwhile, the scatter-plots for ANNs showed better agreement between forecasts and observations – most of the data points are located near the perfect-forecast line (diagonal line). Small difference was observed when ANN-8p is used instead ANN-36p, indicating that the 8 selected predictors was able to provide solar irradiation forecast as reliable as the forecast obtained by using the 36 predictors. TABLE 5SKILL-S CORE CALCULATED WITH RMS E% VALUES FOR ANN TAKING MODEL ETA/CPTEC AND PERS IS TENCE AS REFERENCE METHODS Florianópolis Scores

São Martinho da Serra

ANN36p

ANN8p

ANN-36p

ANN-8p

Skill(RMSE%, persistence)

0.429

0.414

0.464

0.487

Skill(RMSE%, Eta)

0.344

0.328

0.333

0.361

* - results obtained using investigation dataset.

Fig. 4 and 5 present four scatter-plots comparing forecast values and observations. Besides the scatterplots for Eta model, ANN-36p and ANN-8p, it is also showed a plot for a forecast method called persistence. The persistence forecast is the simplest method to predict meteorological data and it consists in taking the value observed in a previous day as the forecast for the current day. Any forecast method is useful if it can lead to better results than the persistence forecast. According to Fig. 4 and 5, the solar radiation flux outputs provided by Eta/CPTEC model are better than persistence forecasts, in general. However, it can be observed the positive bias mentioned before. The Eta/CPTEC model overestimated the observations, especially for cloudy days when solar radiation flux at the surface is lower. TABLE 4 PERFORMANCE SCORESOBTAINED BY ANN-36P AND ANN-8P Scores

Florianópolis

São Martinho da Serra

ANN-36p

ANN-8p

ANN-36p

ANN-8p

R

0.804

0.790

0.839

0.848

R2

0.646

0.625

0.704

0.720

ME%

-2.1%

-0.8%

-1.7%

-0.7%

RMSE%

26.2%

26.9%

28.8%

27.6%

All results obtained using investigation dataset.

224

FIGURE 4 SCATTER-PLOTS OF FORECASTS VERS US GROUND DATAFOR FLN: (A) PERS IS TENCE METHOD, (B) MODEL ETA/CPTEC, (C) ANN-36P, AND (D) ANN-8P

Fig. 6 shows a short temporal series taken from the investigation dataset prepared for FLN and SMS sites. Outputs from Eta/CPTEC model and ANN were put together with observations acquired in Winter/2005 and Summer/2004-2005. Fig. 6 demonstrates the best agreement between the ANN forecasts and ground data. The deviations for each day are presented in Fig. 7. It is clear that an important improvement in short-term forecast for solar radiation flux is achieved when ANN is used to refine solar irradiation outputs provided by


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

model Eta/CPTEC. However, no significant differences were observed between ANN-36p and ANN-8p. Again, the analysis of Fig. 7 demonstrates that the eight selected predictors provide enough information to ANN simulate the atmospheric processes with good performance. To quantify the improvement acquired by the use of ANNs, the skill-score values were calculated using RMSE% score, and the results are presented in Table V. In general, the ANNs lead to skill-scores in RMSE% 30% higher if compared to model Eta/CPTEC.

indices showed that both ANNs have improved the confidence and reliability on the solar radiation forecasts in more than 30% for both sites: Florian贸polis in coastal area and S茫o Martinho da Serra in continental region. The improvements in predictability were also observed as indicated by the correlation coefficients: from 0.72 to 0.80 in FLN, and from 0.78 to 0.85 in SMS.

FIGURE 6 S HORT TIME SERIES COMPARING FORECASTS AND GROUND DATA FOR SOLAR RADIATION FLUX AT S URFACE IN FLN AND SMS

FIGURE 5SCATTER-PLOTS OF FORECAS TS VERS US GROUND DATAFORSMS: (A) PERS ISTENCE METHOD, (B) MODEL ETA/CPTEC, (C) ANN-36P, AND (D) ANN-8P

Conclusions Currently, the renewable sources of energy are getting more importance into electricity generation systems. Therefore, there is an increasing demand from the energy sector for accurate forecasts of solar energy resources in order to support and manage electricity generation and distribution systems. The forecasts provided by numerical weather models could supply this demand but, in general, these forecasts present large deviations reducing their confidence and reliability. In Brazil, the Eta/CPTEC model provided solar irradiation forecasts with bias around 25%. Lower deviations were observed when ANN was used to refine the forecasts provided by the Eta/CPTEC model. The comparison between solar irradiation forecasts and ground data showed a bias reduction from 25%for Eta/CPTEC forecasts till -1% for the ANN outputs. Both ANNs, ANN-36predictors and ANN-8predictors, have presented very similar performances. The skill-score

FIGURE 7 DEVIATIONS BETWEENFORECASTS AND GROUND DATA FOR SOLAR RADIATION FLUX AT S URFACE IN FLN AND SMS. THE MODEL ETA/CPTEC PROVIDED ESTIMATES WITH LARGER DEVIATIONS. ACKNOWLEDGMENTS

The authors would like to thank FINEP (project22.01.0569.00) and PETROBRAS/CENPES (project 0050.0019.104.06.2) for their finance support to

225


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

the SONDA project. Thanks are also due to the following colleagues: Silvia V. Pereira, Sheila A. B. Silva, Rafael Chagas and to the technologists of Laboratório de Instrumentação Meteorológica (LIM/CPTEC). In addition, acknowledgments are due to CPTEC/INPE and CNPq (grants 151700/2005-2, 141844/2006-0, 132148/2004-8, 555764/2010-9). REFERENCES [1]

Che n, Z., Marquis, M., Averyt,K.B., Tignor, M., Mille r, (Eds.),

Climate

2007:

for Weather Forecast and Climate Studies – Brazilian

Change .

Institute for Space Research (CPTEC/INPE), São José

Pane l

on

of

Climate

Be ntle y, R.W., 2002: Global Oil and Gas De ple tion: An Ge ller, H. S., 2003: Re volução Energé tica: políticas para Martins and Pe re ira, 2011.Enhancing information for Energy Policy., 39, 4378-4390.

Brazil.

Available

online

at

<http://e tamode l.cpte c.inpe .br/re fer.shtml>. [13] Kidde r, S. Q., and Vonder Haar, T. H., 1995: Satellite

Meteorology: an introduction. Acade mic Press, 466pp. [14] Iqbal, M., 1983: An Introduction to Solar Radiation.

Academic Press Canada, 390pp. [15] Liou, K.

N., 1980: An Introduction to Atmospheric

Radiation. Acade mic Press Inc., 392pp.

2006. Brazilian Atlas for Solar Energy. Brazilian Institute

Shortwave Radiation with Data Obtaine d from the

for Space Research, 70pp.

Atmosphe ric Radiation Measure ment Program. Master

Viana

et

al,

2011.

Assessing

the

potential

of

of Science Scholarly Paper, Unive rsity of Maryland, USA.

conce ntrating solar photovoltaic ge ne rationin Brazil

[17] Zamora, R. J.; Solomon, S.; Dutton, E. G.; Bao, J. W.;

with sate llite -de rive d direct normal irradiation. Solar

Traine r, M.; Portmann, R. W.; White , A. B.; Ne lson, D.

Energy, 85, 486-495.

W.; Mc Nide r, R.T., 2003: Comparing MM5 Radiative

Re ikard, G., 2008: Pre dicting solar radiationat high

Hamme r, A.; He ine mann, D.; Hoye r, C. and Lore nz, E., 2001: Sate llite Base d Short-Term Forecasting of Solar Irradiance - Comparison of Me thods and Error Analysis, Proceedings of the EUMETSAT Meteorological Satellite Data Users' Conference, Antalya, Turkey, 01.-05.10.2001. He ine mann, D.;Lore nz, E.; Girodo, M., 2004: Forecasting

Fluxes with Observations Gathere d During the 1995 and 1999 Nashville Southern Oxidants Studies. Journal of Geophysical Research,108(D2), 4050-4060. [18] Barbosa, H. M. J.; Tarasova, T. A.; Cavalcanti, I. F. A.,

2008:

Impacts

of

a

Ne w

Solar

Radiation

Paramete rization on the CPTEC AGCMClimatological Features. Journal of Applied Meteorology and Climatology, 47(5), 1377-1392.

of Solar Radiation. Proc. International Workshop on

[19] Kalogirou, S. A., 2001: Artificial ne ural networks in

Solar Resource from the Local Le ve l to Global Scale in

re ne wable e ne rgy syste ms applications: a re vie w.

Support to the Resource Manage ment of Rene wable

Renewable and Sustainable Energy Reviews, 5, 373-401.

Ele ctricity Ge neration. Institute for Environme nt and

[20] Sfe tsos, A.; Coonick, A. H., 2000. Univariate and

Sustainability, Joint Re search Ce nter, Ispra, Italy.

226

Campos,

[16] Armstrong, M. A., 2000: Comparison of MM5 Forecast

Energy, 83, 342-349.

[9]

dos

Pe re ira, E. B., Martins, F. R., Abre u, S. L. and Ruther R.,

re solutions: a comparison of time series forecasts. Solar [8]

[12] Mesinge r, F., 2005: The Eta: brie f history, summary,

the

Assessme ntRe port

solar and wind e ne rgy technology de ployme ntin Brazil.

[7]

South América. J. Geophys. Res., 107(D20), LBA 56, 1-20.

to

um futuro suste ntáve l. Re lume Dumará, 299pp.

[6]

[11] Chou, S. C., Tanajura C. A. S., Xue , Y., and Nobre , C. A.,

Weather Pre dictabilityand Mode ling Workshop. Ce ntre

Ove rvie w. Energy Policy., 30, 189-205.

[5]

104(D16), 19535-19549.

physicalscie nce basis. Contribution of Working Group I

York, p. 996 Online : <htpp://www.ipcc.ch>.

[4]

comparison with measure d ARM data. J. Geophys. Res.,

re sults, discussion. Procee dings of ICTP Regional

Fourth

Change

Cambridge Unive rsity Press, Cambridge , UK and Ne w

[3]

of surface e ne rgy budge t and cloud prope rties by

The

the

IPCC

Intergove rnmental

[2]

T., 1999: An e valuation of NCEP Eta mode l pre dictions

2002: Validation of the couple d Eta/SsiB mode l ove r

IPCC, 2007. In: Solomon, S., Quin, D., Manning, M., H.L.

[10] Hinke lman, L. M., Ackerman, T. P., and Marchand, R.

multivariate forecasting of hourly solarradiation with


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

artificial inte llige nce te chniques. Solar Energy, 68(2), 169-

[28] Fe ls, S. B., and Schwarzkopf, M. D., 1975: The simplifie d

e xchange approximation: a ne w me thod for radiative

178. [21] Mihalakakou, G.; Santamouris, M.; Asimakopoulos, D.

N., 2000. The total solar radiation time series simulation

transfer calculations. J. Atmosph. Sci., 32(7), 1475-1488. [29] Black, T. L., 1994: The ne w NMC mesoscale Eta mode l:

in Athe ns,using ne ural ne tworks. Theoretical and Applied

Description

Climatology, 66, 185-197.

Forecasting, 9(2), 265-278.

[22] Hocaoglu, F. O.; Gere k, O. N.; Kurban, M., 2008: Hourly

solar radiation forecasting using optimal coe fficie nt 2Dline ar filters and fee d-forward ne ural ne tworks. Solar Energy, 82, 714-726. [23] Tymvios, F.S.; Jacovides, C.P.; Michae lides,S.C.; Scoute li,

C., 2005: Comparative study of Angström’s and artificial ne uralne tworks me thodologies in estimating global solar radiation. Solar Energy, 78, 752-762. [24] Chou, S. C.; Bustamante , J. F.; Gomes, J. L., 2005:

Evaluation of seasonal pre cipitation forecasts over South Ame rica using Eta mode l. Nonlin. Proc. Geophys., 12, 537-555. [25] Staude nmaie r Jr., M., 1996: A description of the Meso

Eta mode l. Western Region Technical Attachment no. 96-06, 5pp. Available online at <http://www.wrh.noaa.gov/wrh/96TAs/TA9606/ta9606.html>. [26] Ničković, S., Mihailović, D., and Rajković, B., 1988:

Scientifc

documentation

of

the

Eta

model.

Word

Me teorological Organization. the

absorption

of solar

forecast

e xamples.

Weather and

[30] Janjić, Z. L., 1994: The ste p-mountain Eta coordinate

mode l: Furthe r de ve lopme nts of the convection, viscous sublaye r, and turbule nce closure sche mes. Monthly Weather Review, 122(5), 927-945. [31] Mesinge r, F.; Janjić, Z. L.; Ničković, S., and Deaven, D.

G.,

1988:

The

ste p-mountain

coordinate :

Mode l

description and performance for cases of Alpine lee cicloge nesis

and

for

a

case

of

Appalachian

re deve lopme nt. Monthly Weather Review,116, 1493-1518. [32] Kipp & Zone n, 2006: Kipp & Zone n Co. Webpage :

manufacture r of mete orological and industrial se nsors. Available online at <http://www.kippzone n.com/>. [33] Rume lhart, D. E., Hinton G. E., and Williams, R. J., 1986:

Learning re prese ntations by back-propagation e rrors. Nature, 323, 533-536. [34] Rie dmille r, M., and Braun H., 1993: A dire ct adaptive

me thod for faster backpropagation learning: The RPROP algorithm.

Proceedings, IEEE International

Conference on Neural Networks. [35] Guarnie ri, R. A., 2006: Empre go de re des Ne urais

Artificiais e Regressão Linear Múltipla no re finame nto

[27] Lacis, A. A., and Hanse n, J. E., 1974: A parame te rization

of

and

radiation

in

earth's

das pre visões de radiação solar do mode lo Eta.Insituto Nacional de Pesquisas Espaciais, 171pp.

atmosphe re . J. Atmosph. Sci., 31, 118-133.

227


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Air Cells Using Negative Metal Electrodes Fabricated by Sintering Pastes with Base Metal Nanoparticles for Efficient Utilization of Solar Energy Taku Saiki, Takehiro Okada, Kazuhiro Nakamura, Tatsuya Karita, Yusuke Nishikawa, Yukio Iida Department of Electrical and Electronic Engineering, Faculty of Engineering Science, Kansai University 3-3-35 Yamate, Suita, Osaka. 564-8680, Japan tsaiki@kansai-u.ac.jp Abstract Research on the produce of re ne wable e ne rgy as a source of solar powe r has continuously advance d. We have propose d an e nergy cycle that uses solar-pumpe d pulse lase rs and base me tal nanoparticles. He re , Fe and Al nanoparticles we re pre pare d by lase r ablation in liquid for the e nergy cycle . Solar powe r was confine d in base me tal nanoparticles. Me tal plates we re fabricate d by sinte ring metal paste with base me tal nanoparticles. Electricity was ge ne rate d by air ce lls using the sintere d me tal plate . A highly re pe titive laser pulse , which was an alte rnative to lasers drive n by solar power, was use d for laser ablation in liquid, and me tal oxides (Fe3 O4 or Al2 O 3 ) we re re duce d and metal nanoparticles were fabricate d. Me tal plates with a low e lectrical resistance were fabricate d by sinte ring them at a low tempe rature of 520 K. The e lectrical properties of the air ce lls fabricate d using sinte re d paste with nanoparticles as negative e lectrical cathodes we re the same as those of the air ce lls fabricate d in a blast furnace . It was found that the sinte re d me tal nanopaste could be use d for air ce lls. Keywords Solar Power; Laser; Renewable Energy; Air Cells; Nanoparticles

Introduction Recently, concerning the desire to develop methods for the reduction in the amounts of gases that causes the global earth warming, low-cost and energy-saving method are encouraged. Also, research on the production of renewable energies has continuously advanced [1-3]. High temperatures are generated by focusing solar light on metal oxides, thereby reducing the metal oxides by the generated high temperatures. As a result

228

of this process, chemical energy is stored as the difference of chemical potential between metal and metal oxide, and hydrogen is generated [1]. In previous research, CW lasers have been generated using solar energy is used for the reduction of magnesium oxides. Here, an energy cycle in which reduced magnesium is used as a renewable energy is proposed [3, 4]. Our group is also developing solarpumped lasers [3, 5-7] that employ solar light for pumping laser materials. We propose an energy cycle using base metallic nanoparticles and solar-pumped pulsed lasers. A Nd/Cr:YAG ceramic laser has been used as a solarpumped laser [6, 7]. Its lasing wavelength is 1064 nm, which is the same as that of the Nd:YAG laser. The ceramic laser has a special lasing mechanism because its photon energy includes thermal energy due to the phonon-assisted cross-relaxation effect [8]. The maximum theoretical optical-optical (from solar light to laser) conversion efficiency reaches close to 80%. Very high opt.-opt. conversion efficiency close to 60% has been achieved in experiments. The generation of efficient high peak-power and highly repetitive laser pulses from solar-pumped lasers, whose conversion efficiency is close to that of CW laser [6], has been realized. The reduction and production of metal nanoparticles using laser ablation in liquid and solar-pumped pulsed lasers has been performed here. The method is different from that adopted by Yabe et. al [4]. Laser ablation in liquid with a high fluence and a high intensity of laser pulses can produce nanoparticles and


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

reduce metal oxides [9-13]. The physics of thermal ablation and coulomb explosion [14-16] have been proposed for the ablation of metal oxides in liquid, which is remarkably different from that using a CW laser.

The laser ablation method in liquid is described in the paragraph below. When laser pulses are irradiated onto metal oxides in liquid, the metal oxides melt and

It has been considered that reduced metal oxides can be used as a negative electrode of air cells. Electricity can be obtained by reacting metals with oxygen in air. Research on air cells has been carried out for 40 years since Ferro developed Zn air cells [17, 18]. Air cells are expected as a future electrical power source because their energy storage density per unit weight is higher than that of the Li ion battery. Metal oxides, such as Fe3O4 and Al2 O3, exist in large quantities in the ground.

nanoparticles rapidly. The merits of the use of this method are as follows: 1) We do not need to use specific materials for reduction. 2) A high reduction

Our aim in this paper is to develop primary air cells [17, 18]. The electrical property of air cells using sintered metal paste was investigated and compared with that of air cells using conventional metals. No such researches have been performed previously. The possibility of using sintered paste with reduced metal oxide particles fabricated by laser ablation in liquid as a negative electrode was investigated. Experimental Reduction of metal oxide powder and production of nanoparticles By using pulse laser ablation in liquid, metal oxides were reduced, and metal nanoparticles such as Fe and Al were fabricated. The chemical formulas for the reduction are shown. We first show the chemical formula for Fe3O4 and Fe:

resolve and the melted oxides are set outside the metal nanoparticles. The surrounding liquid cools the metal

rate of metal oxides is obtained because the recombination between oxygen and metals is prevented. 3) The re-collection of nanoparticles is easier than in other methods. 4) It has a very low cost. We obtained reduced Fe nanoparticles with 20 nm diameters by this method [13]. The experimental setup for laser pulse ablation is shown in Fig. 1. A microchip Nd:YAG laser was used in this experiment. The maximum output averaged laser power was 250 mW, the laser wavelength was 1064 nm, the repetitive rate of the laser pulses was 18 kHz, and the pulse duration was 8 ns. A beam with a diameter of 6 mm (1/e2) was focused using a lens with a focal length of 50 mm. Thus, the diameter of the focused beam was 20 µm at the front of each glass bottle. Glass bottles with a size of 20 mmΦ x 5 mm were used in the experiment. Fe3O4 andα-Al2O3 powders (Koujyundo Chemical Laboratory) were mixed with water in each glass bottle for the experiment. The mean size of each Fe3O4 andα-Al2O3 was 1 µm. Each glass bottle was set after the focused laser beam. The weight of the Fe3O4 andαAl2O3 powders was measured using an electronic force

(1)

balance. Their measured weight was 200 mg. 4mL of pure water was placed in each glass bottle. Laser

The chemical formula for Al2O3 and Al is shown next:

pulses were irradiated to the water with the metal oxide in glass bottle for 20 minutes. Here, the fluence

1 2 Fe 3 O4 → Fe + O2 , ΔH=373 kcal/mol. 3 3 1 3 Al2O3 → Al + O2 , ΔH=838 kcal/mol. 2 4

(2)

Fe oxides are resolved and vaporized at a temperature of above 1600 K. Furthermore, Al oxides are resolved and vaporized at a temperature of above 3300 K.

of the irradiated laser pulse was estimated to be 4.5 J/cm 2. Ketones, such as acetone, are used as liquids for laser ablation in liquid. Here, we neglect the oxidation at the surface of metal nanoparticles. A magnetic stirrer was used to mix the liquid. After the irradiation of the laser pulses for 20 minutes in the case of Fe3 O4, the surface of the nanoparticles was black. Thus, it has been presumed that the surfaces of metal nanoparticles were surrounded by Fe3O4. In the case of α-Al2O3 powder, its color changed to gray, which is

FIG. 1 LASER SYS TEM FOR LASER ABLATION IN LIQUID US ING MICROCHIP LASER

close to the color of Al powder. The powder after irradiating laser pulses in the water was dried to not change chemically.

229


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Fabrication of metal plates by sintering paste

(a)

(b) FIG. 2 SINTERED METAL PAST E. (A) FE AND (B) AL TABLE I MEAS URED ELECTRICAL RES ISTANCE

Al

Large (unmeasurable) Large (unmeasurable)

Large (unmeasurable) Large (unmeasurable)

Paste after sintering[Ω] 0.0 0.0

The dried Fe and Al nanopowders were mixed with 3mg of Ag nanopastes (NAG-10 Daiken Chemical); the viscosity of the paste was high. Fe paste was sintered using an electrical hot plate for 5 minutes (1 minute at 520 K, 4 minutes at 570 K). In the case of the Al paste, the paste was also sintered for total 5 minutes (1 minute at 510K, and 4 minutes at 550K) to generate less gas per unit time. The sintered metal plates are shown in Figs. 2(a) and 2(b). The sintered Fe plate is shown in Fig. 2(a), and the sintered Al plate is shown in Fig. 2(b). The opposite surface of either plate was not metalized by oxidation. A tester measures the electrical resistance of the metal oxide, and the metal paste, and sintered metal paste. The Fe3O4 andα-Al2O3 powders, and Fe and Al pastes were set on a glass sheet with 1 mm thickness and sintered. The results are shown in Table II. The distance between the needles of the tester was 8 mm. The electrical resistances of the Fe3O4 andα-Al2O3 powders were very high; they could not be measured

230

600 Al Al 2 O3

500 Intensity (arb. unit)

Fe

Paste[Ω]

400 300 200 100 0

20

30

40

50 2θ (degree)

60

70

80

(a) 350 Fe Fe3 O4

300

Al

Intensity (arb. unit)

Original powder[Ω]

because both powders are insulators. The electrical resistances of the metal pastes mixed with reduced Fe and Al nanoparticles by irradiating laser pulses were also measured. However, they were also very high and thus could also not be measured. Finally, after sintering the metal pastes, the measured electrical resistance is 0.0Ω. It was prospected that the Fe and Al pastes were both metalized. For the sintered Al paste prepared by this method, a weak ferromagneticity was observed, and thus it was presumed that the metal structure of Al is markedly different from the structure of common metals. Crystal structure analysis using XRD was performed to check the quantities of Fe3O4 andα-Al2O3 in the sintered metal paste samples. The results are shown in Figs. 3(a) and 3(b). An XR D instrument (MAXima_X XRD-7000 Shimazu Japan) was used for the experiment. K-α X ray radiation of Cu was irradiated onto the samples. The results for analyzing Al and Fe plates are shown in Fig. 3(a) and Fig. 3(b), respectively. In the case of Al, a large spectrum of Al and a small of spectrum of α-Al2O3 were slightly observed. The Al spectrum contained components of a little Ag nanoparticles and the Al stage. Thus, the quantity was not determined.

250 200 150 100 50 0

20

30

40

50 2θ (degree)

60

70

80

(b) FIG. 3 RESULT S OF XRD ANALYSIS: (A) AL AND (B) FE


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Moreover, determining the quantity of Fe was difficult because, normally, the peak spectrum of the Fe3O4 is larger than that of Fe at the same quantity [19]. However, we can see the two components at angles of 45 and 65 degrees in Fig. 3(b), but the Fe component was clearly recognized. The spectrum peak of Fe at 44.7 degrees evaluated by extracting the Al background component is nearly three times higher than that of Fe3O4 at 43 degrees, and it was found that the sintered Fe paste contains negligible quantity of Fe3O4 [19]. Additionally, the sintered structure was observed by a microscopy, and it was found that the porous structure consisted of small metal particles in the metal plates.

from the negative electrodes per chemical reaction, the electrical power density per unit weight is high. Metal nanopaste could help to connect different metal tightly. It is expected that the output voltage of the air cells will be improved by using different metal junctions. The air cells using different species of metal junctions utilize the phenomenon of galvanic corrosion. For example, new local cells are constructed between Fe nanopaste and an Al plate, and a new electromotive force is generated. Here, Fe corrosion occurred before Al corrosion occurred resulting in the generation of a high output voltage depending on the use of Al as a total air cell.

Air cell An experimental setup for testing air cells is shown in Fig. 4. Fe (JIS G3141) and Al (JIS 1050) plates were used as the metal plates in the negative electrodes. The dimensions of the metal plates were set to be 20 mm x 15 mm x 0.5 mm. However, the size of the metal paste was set to be 8 x15 mm 2, which is almost half of the metal plate, because of the difference in duration of the output voltage between the use of the metal pastes and that of the conventional metal plates. The electrical property of the sintered metal pastes on the metal plates was compared with that of the conventional metal plates. Here, a Pt-doped carbon electrode with a layer for diffusing oxygen was used as the positive electrode. A separator made of 8 pieces of papers piled up was set between the positive electrode and the negative metal electrode. The thickness of each piece of paper was 0.1 mm. Saturated salt water was injected into each piece of papers at intervals of 5 minutes. Th e chemical formulas of Fe air cells are shown below:

1 O2 + H 2O + 2e − → 2OH − , 2 Fe + 2OH − → Fe(OH) 2 + 2e − .

(3 )

The chemical formulas of Al air cells are also shown:

3 3 O2 + H 2O + 3e − → 3OH − , 4 2 Al + 3OH − → Al(OH) 3 + 3e − .

FIG. 4 EXPERIMENTAL SET UP FOR METAL AIR CELLS

Electrical Property of Air Cells Previously, as shown in section 2, metal pastes with Fe and Al were sintered on metal plates, and air cells were fabricated. The electrical properties of the air cells were then investigated. The open output voltage of air cells using a negative Mg electrode was 1.8 V, and that of air cells using negative Al electrode was 1.2 V. That of air cells using a negative Fe electrode was 0.6V. Fe paste on a Fe plate was sintered. Al paste on an Al plate was also sintered. The measured shortcut currents of the air cells using Fe and Al plates were both 70 mA. In the case of Fe and Al pastes, the extracted shortcut current was 40 mA because the area was half of that using Fe and Al plates. Small connecting load

(4)

The theoretical electromotive forces of Fe and Al air cells are 1.2 and 2.7 V, respectively. However, using NaCl dissolved in water, the output voltages of the cells were 0.4 and 0.7V, respectively. Because the valence of the Al ion is 3 and 3 electrons are emitted

The temporal property of the output voltage when the electrical load connected to the air cells is small is shown in Fig. 5(a). A 1.0 kΩ resistor was connected to the constructed air cell, and their output voltages were measured. The obtained output voltages of the Fe, Fe paste–Fe and Fe paste–Al, air cells were all 0.4V, one of the Al and Al paste–Al air cells were all 0.7V, and

231


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

that of the Mg and Al paste–Mg air cell were all 1.4V. Current can be evaluated by dividing the output voltage by the 1.0 kΩ resistance. The evaluated output current of the Mg air cell was 1.4mA, that of one of the Al air cells was 0.7 mA, and that of one of the Fe air cells was 0.4 mA. Large connecting load A large load, which was applied using an electrical motor (Solar motor, Mabuchi RF-500TB Tamiya, Japan) with a propeller, was connected to the air cells to induce the flow of large currents. The temporal property of the output voltage is shown in Fig. 5(b). The motor rotated normally. However, when connected to the Fe air cells, the motor did not rotate owing to low output voltage. The temporal property of the output voltages was measured when the motor was connected to the air cells. In the case of the Fe paste –Al air cell, the initial output voltage was 0.7 V. In the case of the Al paste –Al air cell, the initial output voltage was also 0.7 V. The output current was 26 mA. The duration of the output voltage was 60 minutes for the Al air cell, and that for the Al paste –Al air cell was Mg Al Fe Fe paste -Al Al paste -Mg Fe paste -Fe Al paste -Al

2

(V)

1.5

V

out

1

0.5

0 -10

0

10

20 30 40 Time (Min.)

50

60

70

(a) 2 Mg Al

30 minutes, half of that for the Al air cell. Moreover, that for the Fe paste –Al air cell was 30 minutes, which was also half that for the Al air cell. The initial output voltage in the case of the Mg air cell was 1.4 V, and that in the case of the Al paste –Mg air cell was also 1.4 V. The output currents were 28 and 27 mA, respectively. The duration of the output voltage was 80 minutes for the Mg air cell, and that for the Al paste –Mg air cell was 40 minutes, which was also half of that for the Mg air cell. DISCUSSION We conducted experiments to produce nanoparticles and reduce metal oxides by laser ablation in liquid. It has been proved in the experiments that metal plates prepared by sintering metal pastes can be obtained, and that the electrical resistances of the metal plates are close to those of conventional metal plates. The Clarke numbers of Al and Fe are 3 and 4, respectively. They exist abundantly in the ground of the earth. The proposed method of reducing and metalizing metal oxides is an alternative to the conventional electrolysis method in aluminium refining. Reduction is performed in two steps; laser ablation and sintering. Ag nanopaste has less ability to reduce metal oxides. Ag paste with only Fe3O4 powder could not be sintered, and no perfect metal plate was fabricated in fact. The inner surface of the paste was not sintered into the metal. Some heat sources or the solar light can sinter the metal paste. When sintering metal paste, the degradation of surface energy induces heat generation during the chemical reaction. When the paste reaches a given temperature, sintering starts automatically with the generated heat. No heat is required to melt a common metal plate. Required energy to start sintering is adequately lower than the stored energy in the sintered metals owing to the vanishing surface energy.

Fe paste -Al

1.5

Al paste -Mg

28 mA

Al paste -Al

(V)

27 mA

V

out

1

23 mA

0.5

0

0

20

40

60 Time (Min.)

80

100

120

(b) FIG. 5 MEASURED OUTPUT VOLTAGE OF METAL AIR CELLS. THE LOAD WAS (A) 1KΩ AND (B) A MOTOR

232

Here, we consider the input-output energy balance. The injected total laser energy determined by calculation was 290 J when the averaged laser power was 250 mW and the irradiation time was 20 minutes, considering surface loss of each glass bottle. The minimum chemical energy per mol to reduce Fe3O4 to Fe is 373 kJ, that to reduce Al2 O3 to Al is 838 kJ. In the case of Fe, when its weight is 200mg, the required minimum energy to reduce it is evaluated to be 960 J. This energy is 3.3 times as large as the injected total laser energy in liquid. In the case of Al, when its


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

weight is 200 mg, the required minimum energy to reduce it is evaluated to be 3300J. This energy is 11 times as large as the injected total laser energy. Moreover, more energy for reduction is needed because the generated metal and oxide atoms must be removed far away from metal atoms. These results strongly show that the physics of the reduction process does not depend on conventional thermal ablation. Because the estimated irradiated laser fluence was 4.5J/cm 2 in this case, the ablation of the metal oxides occurred by the coulomb explosion. Also, it has been thought that reduction of 200 mg of Fe3O4 or Al2O3 powder can be almost perfectly performed when the averaged laser power is 250 mW, and the irradiation time is 10 minutes. In coulomb explosion, avalanche ionization occurs in metal oxides. Some of the electrons of Fe3O4 are ejected into water. Fe and O atoms are ionized and exploded by the coulomb repulsion between ions. Finally, local plasma is generated. Air cells using sintered metal nanopaste were constructed and their electrical property was investigated. It was clarified that the fabricated air cells using sintered metal nanopaste can be used as primary air cells. If the weight of oxygen is neglected, the potential electrical energy density per unit weight of Fe used in air cells is estimated to be 1160 Wh/kg, and that of Al is estimated to be 8100 Wh/kg. The potential energy of Al is 7 times higher than that of Fe. The atomic weight of Fe is 55.6, and that of Al is 27. Thus, the potential current of Al is higher than that of Fe, indicating that Al has an advantage for generating electricity. The electrical resistance of m etal plates must be adequately low because currents are extracted from such plates. A large plate made from sintered paste may have a high electrical resistance, resulting in the degradation of the output voltage. Because we did not fabricate large-scale metal plates, the sintered metal plates had lower electrical resistances, and their resistance could not be estimated. However, the resistance is as low as that of common metal plates. A dissimilar metal joint is important for improving the electromotive force of air cells. By sintering metal paste on common metal plate, a more rigid connection between them is obtained, and contact resistance is reduced. From the experimental result, it has been thought that a dissimilar metal joint between the Al and Mg plates for the negative metal electrode has a low contact resistance. When using a Mg plate, the output voltage of air cells decreased 80

minutes after connecting electric codes. However, when using Al paste and a Mg plate as an air cell, the output voltage was sustained 40 minutes after connecting the codes. Because the output voltages of the Al paste -Al air cells are the same as that of the Al air cells when the load is low, the output voltage will be maintained near 1.4 V for 80 minutes if the area of the metal paste is twofold. After using Fe and Al in air cells, metal oxides are generated. These metal oxides return to Fe and Al by laser ablation in liquid. The negative metal electrode used is exchanged to new ones. In this experiment, to improve the electromotive force of air cells, we had used a Mg plate as a base metal plate. Using Li plates will improve the electromotive force when sintered Al paste is used, and a higher electrical stored energy per unit weight of the air cells will be obtained. Solar energy or other natural energies are considered as the sources of laser power for laser ablation in liquid. However, the most suitable lasers for energy cycles are considered to be solar-pumped lasers because common lasers have a low electro-optical conversion efficiency and their generated energy gains are vanished. Repetitive usage of metal oxide by laser ablation and researching the maximum stored energy gain of metals are the future objects. Conclusions As an example of renewable energies for utilizing solar energy efficiently, a renewable energy obtained using metallic nanoparticles fabricated by laser ablation in liquid was proposed in this study. Laser pulses are generated using solar power. It has been clarified that metal nanoparticles can be used in metal air cells to generate electricity. By using high repetitive laser pulses, an alternative to pulsed solar-pumped lasers, Fe3O4 and Al2O3 were reduced, and Fe and Al nanoparticles were fabricated. Metal plates were obtained by sintering paste with Fe and Al nanoparticles at around 520 K. It was found that Fe and Al metal plates have low electrical resistances. The electrical properties of the air cells using Fe and Al plates fabricated by sintering Fe and Al nanopaste had been compared with those using Fe and Al plates fabricated in a blast furnace. Very close output voltages and currents were obtained. Different species of junction metal plates were used as negative electrodes

233


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

for the air cells. A higher output voltage of the Al paste–Mg air cells than that of the Al air cells has been obtained. Al was connected to Mg tightly by sintering Al nanopaste. The measured out put voltage of Al paste-Mg air cell was 1.4 V when an electrical motor was connected, which is the same as that of the Mg air cell. Also, a high output voltage of the Fe paste-Al air cell than that of the Fe air cell was obtained.

[3]

[4]

Production

by

Three -Ste p

Solar

[7]

[10] M. S. Sibbald, G. humanov, and T. M. Cotton,

“Re duction of cytochrome c by halide -modifie d, lase r-

[11] M.

Kawasaki

fragme ntative

and

N.

Nishimura,”Laser-induce d

decomposition of ke tone -suspe nde d

nanoparticle s”, J. Phys. Chem. C, vol. 112 pp.15647-

July 2007.

15655, Se p. 2008.

D. G. Rowe , “Solar-powe re d lase rs”, Nature Photonics,

[12] H. Q. Wang,, A. Pyatenko, K. Kawaguchi, X. Y. Li , Z.

vol. 4 pp.64-65, Feb. 2010.

Swiatkowska-Wackocka, and N. Koshizaki, ”Se le ctive

T. Yabe , T. Okubo, S. Uchida, K. Yoshida, M.

pulse d heating for the synthesis of se miconductor and

Nakatuska, T. Funatsu, A. Mabuti, A. Oyama, K.

me tal submicromete r sphe res”, Angew. Chem. Int. Ed.,

Nakagawa, T. Oishi, K. Daito, “High-e fficie ncy and

vol. 49 pp.6361-6364, Aug. 2010.

economical solar-e nergy-pumpe d lase r with Fresne l

[13] S. Taniguchi, T. Saiki, T. Okada, T. Furu, “Synthesis of

le ns and chromium codope d laser me dium”, Appl. Phys.

reactive metal nanoparticles by lase r ablation in liquids

Lett., vol. 90 pp.261120-261122, June 2007.

and its applications”, Re port on The 421th Topical

M. S. Mohame d, T. Yabe , C. Baasandash, Y. Sato, Y.

Mee ting of

Mori, L. Shi-Hua, H. Sato, and S. Uchida, ”Lase r-

Technology for 21st Ce ntury, No. RTM-11-56, (2011)

induce d magnesium production from magnesium

pp.25-30 [in Japanese ].

The

Laser Socie ty of Japan,

Laser

[14] J. Purne ll, E. M. Snyde r, S. We i, and A. W. Castleman,

pp.113110-113116, De c. 2008.

Jr, “Nuclear fusion induce d by coulomb e xprosion of

M. We ksle r and J. Shwartz”, Solar-pumpe d solid-state

he teronuclear cluste rs”, Chem. Phys. Lett., vol. 229

lasers”, IEEE J. Quantum. Electron, vol. 24 pp.1222-1228,

pp.333-339, Nov. 1994. [15] K. Yamada, K. Miyajima, and F. Mafune , “Ionization of

T. Saiki, S. Motokoshi, K. Imasaki, K. Fujioka, H.

gold nanoparticle s in solution by pulse laser e xcitation

Yoshida,

C.

as studie d by mass spectrome tric de tection of gold

Yamanaka, ”Highly-re peate d lase r pulses amplifie d by

cluste r ions”, J. Phys. Che m. C, vol. 111, pp.033401-1-4,

Nd/Cr:YAG ce ramic amplifie r unde r CW arc-lamp

July 2007.

H.

Fujita,

M.

Nakatsuka,

and

light pumping”, Opt. Comm., vol. 282, pp.2556-2559,

[16] M. Shoji, K. Miyajima, and F. Mafune ,”Ionization of

May 2009. T. Saiki, S. Motokoshi, K. Imasaki, K. Fujioka, H.

gold nanoparticle s in solution by pulse laser excitation

Yoshida, H. Fujita, M. Nakatsuka, and C. Yamanaka, “Nd3+- and Cr3+-dope d yttrium aluminum garnet

cluste r ions”, J. Phys. Chem. C, vol. 112 pp.1929-1932,

pulse

lase r

using

Cr4+-dope d

yttrium

aluminum garne t crystal passive Q-Switch”, Jpn. J. Appl. Phys., vol. 48 pp.122501-1-7, Jan. 2009.

234

pp.5457-5471, May 1993.

Oxide Systems”, Energy & Fuels, vol. 21, pp.2919-2928,

ce ramic

[8]

atom-to-me tal transition”, J. Phys. Chem., vol. 97

Ag2O micropowde rs to nove l se lf-stabilize d Ag

July 1988. [6]

che misorption, composite me tal particles, and the

The rmoche mical Cycles Using Hydroxides and Me tal

oxide using re ducing age nts”, J. Appl. Phys., vol. 104 [5]

me tal particles in solution: "microe lectrode " reactions,

pp.4672-4678, Mar. 1996.

P. Charvin, S. Abanades, F. Lemort, and G. Flamant, “Hydroge n

[2]

A. He ngle in, “Physicoche mical prope rties of small

ablate d silve r colloids”, J. Phys. Chem., vol. 100

REFERENCES [1]

[9]

T. Saiki, M. Nakatsuka, K. Imasaki, “Highly e fficie nt lasing action of Nd3+- and Cr3+-dope d yttrium aluminum garnet ce ramics base d on phonon-assiste d cross-re laxation using solar light source ”, Jpn. J. App. Phys., vol. 49 pp.082702-1-8, Aug. 2010.

as studie d by mass spectrome tric de tection of gold Jan. 2008. [17] K. Ikawa, T. Horiba, Air Ce lls, Japan Pate nt 258782, Oct.

8, 1993. [18] Y. Kudo, M. Suzuki, Al slid-stage Air Ce lls, Japan

Pate nt 147442, July 20, 2006. [19] Y. Kataoka, and S. Itushiki, For Analysis by X radiation,

Production Research vol. 12 (1960) p.311 [in Japanese].


International Journal of Energy Scie nce Vol. 2 Iss. 6, November 2012

Blends of Diesel â&#x20AC;&#x201C; used Vegetable Oil in a Four-Stroke Diesel Engine Charalampos Arapatsakos1 Department of Production and Management Engineering, Democritus University of Thrace V. Sofias Street, 67100, Xanthi, Greece xarapat@agro.duth.gr

1*

Abstract In the days before the proliferation of large cities and industry, natureâ&#x20AC;&#x2122;s own systems kept the air fairly clean. Wind mixed and dispersed the gases, rain washed the dust and other easily dissolved substances to the ground and plants absorbed carbon dioxide and replaced it with oxygen. With increasing urbanization and industrialization humans started to release more wastes into the atmosphere than nature could cope with. Since then, more pollution has been added to the air by industrial, commercial and domestic sources. There are several many types of air pollutant. These include smog, acid rain, the greenhouse effect and holes in the ozone layer. The atmospheric conditions such as the wind, rain, stability affect the transportation of the air pollutant. This paper examines the use of diesel-used vegetable oil mixtures in a four-stroke diesel engine. The mixtures that have been used are the following: diesel-5% used vegetable oil, diesel-10% used vegetable oil, diesel-20% used vegetable oil, diesel-30% used vegetable oil, diesel40% used vegetable oil, diesel-50% used vegetable oil. For those mixtures the gas emissions of carbon monoxide (CO), hydrocarbons (HC), nitrogen monoxide (NO), smoke are being measured. Also the gas emissions temperatures are being measured and the consumption for any fuel mixture is examined. The fuel temperatures were 30 oC and 40 oC. Keywords Gas Emissions; Vegetable Oil; Biofuels; Fuel Temperature Introduction Air pollution is one of the most serious environmental problems confronting our civilization today. Air pollution is the presence of toxic chemicals or compounds in the air. Thes e compounds may be found into the air in two major forms, in a gaseous and

in a solid form. The most common causes of air pollution are various human activities, including industry, construction, transport agriculture etc. However, there are some natural processes such as volcanic eruptions and wildfires too [1, 2, 3]. The effects of air pollution vary from simply coughing or skin problems to serious diseases, such as cancer, chronic respiratory disease, heart disease etc. People of all ages can be affected from air pollution and particularly from sources such as vehicle exhausts and residential heating, but mainly those with existing heart and respiratory problems are in an extra risk. Air pollutants are also responsible for the acidification of forests and water ecosystems and eutrophication of soils and waters and corrode buildings and materials [4, 5, 6]. One of the main causes of air pollution is transportation and particularly the increased emissions from the road traffic. In order to improve air quality scientists are focusing in the use of alternative fuels that can give energy without harming the environment. Biomass offers a physical way to produce energy without damaging the environment. Biofuels are alcohols, ethers, esters, and other chemicals made from cellulosic biomass such as herbaceous and woody plants, agricultural and forestry residues, and a large portion of municipal solid and industrial waste. The term biofuels can refer to fuels for electricity and fuels for transportation. Unlike petroleum, which is a non-renewable natural resource, biofuels are renewable and inexhaustible source of fuel. Biofuel is used to produce power, heat and steam and fuel through a number of different processes. Consequently, it can be used to power vehicles, heat homes and for cooking. Vegetable oil is an alternative renewable fuel for diesel engines [7, 8, 9]. There are two main types of vegetable oil fuels, the straight vegetable oil and the waste vegetable oil. Straight vegetable oil is the relatively unprocessed or unadulterated oil pressed from a variety of vegetables

235


International Journal of Energy Scie nce Vol. 2 Iss. 6, December 2012

and plants. These oils can be used for cooking and power vehicles too. Some examples of vegetable oil are palm oil, cottonseed oil and corn oil. Waste vegetable oil is the oil that has already been used for cooking and can no longer be used for that purpose. Both types of oil can be used just as they are or they can be mixed with diesel fuel in engines modified to use them. The use of vegetable oils has many benefits. First of all it is better for your engine as it provides additional lubrication and reduces engine deposits. It is less likely to cause a fire or explosion in the case of an accident. It also results in lower emissions, as the carbon dioxide produced by burning vegetable oil is less than the amount absorbed by the plants from which the oil is obtained, vehicles running on vegetable oil produce no net increase in atmospheric carbon dioxide. Finally, vegetable oil fuel is indefinitely renewable. However, in order to use vegetable oil either straight or waste, it requires engine modification, which is inconvenient and expensive [10]. The major issue is how a four-stroke diesel engine behaves on the side of pollutants and operation, when it uses directly mixed fuel of diesel â&#x20AC;&#x201C; used vegetable oil [11]. Instrumentation and Experimental Results In the experiment stage has been used directly used

vegetable oil (used sunflower oil that emanated from cooking) in the mixture of diesel in to a four â&#x20AC;&#x201C; stroke diesel engine. Specifically it has been used diesel, mixture diesel-5% used vegetable oil (tig5), diesel-10 used vegetable oil (tig10), diesel-20% used vegetable oil (tig20), diesel-30% used vegetable oil (tig30), diesel40% used vegetable oil (tig40), diesel-50% used vegetable oil (tig50) in a four-stroke diesel air-cooled engine named Ruggerini type RD-80, volume 377cc, and power 8.2hp/3000rpm, who was connected with a pump of water centrifugal. Measurements were made when the engine was functioned on 1000, 1500, 2000 and 2500rpm. The fuel temperatures were firstly 30 oC and secondly 40 oC. During the experiments, it has been counted: The percent of CO, the ppm of HC, the ppm of NO, the percent of smoke, the gas emissions temperature and the fuel consumption. The measurement of rounds/min of the engine was made by a portable tachometer (Digital photo/contact tachometer) named LTLutron DT-2236. Smoke was measured by a specifically measurement device named SMOKE MODULE EX HAUST GAS ANALYSER MOD 9010/M, which it has been connected to a PC unit. Th e CO and HC emissions have been measured by HORIBA Analyzer MEXA-324 GE. The NO emissions have been measured by a Single GAS Analyser SGA92-NO. The experimental results are shown at the following figures:

PIC. 1 EXPERIMENTAL LAYOUT

236


International Journal of Energy Scie nce Vol. 2 Iss. 6, December 2012

o

30 C diesel

0.08

tig5 tig10

0.07

tig20 0.06

tig30 tig40

CO%

0.05

tig50

0.04 0.03 0.02 0.01 0 1000

1500

2000

2500

rpm

FIG. 1 THE CO VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS 30O C

40o C 0.08 diesel 0.07

tig5 tig10

0.06

tig20 tig30

CO%

0.05

tig40 tig50

0.04 0.03 0.02 0.01 0 1000

1500

2000

2500

rpm

FIG. 2 THE CO VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS 40O C

30oC 60

diesel tig5

50

tig10

40

tig30

HC(ppm)

tig20 tig40 30

tig50

20

10

0 1000

1500

2000

2500

rpm

FIG. 3 THE HC VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS 30 OC

237


International Journal of Energy Scie nce Vol. 2 Iss. 6, December 2012

o 40 C

60

50 diesel

HC(ppm)

40

tig5 tig10

30

tig20 tig30

20

tig40 tig50

10

0 1000

1500

2000

2500

rpm

FIG. 4 THE HC VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS 40 OC

30oC 1800

diesel tig5

1600

tig10 1400

tig20 tig30

NO(ppm)

1200

tig40 1000

tig50

800 600 400 200 0 1000

1500

2000

2500

rpm

FIG. 5 THE NO VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS 30O C

40oC diesel

1800

tig5 1600

tig10 tig20

1400

tig30

NO(ppm)

1200

tig40 tig50

1000 800 600 400 200 0 1000

1500

2000

2500

rpm

FIG. 6 THE NO VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS 40O C

238


International Journal of Energy Scie nce Vol. 2 Iss. 6, November 2012

Îż

30 C diesel tig5

40

tig10 tig20 tig30 30

tig40

smoke%

tig50

20

10

0 1000

1500

2000

2500

rpm

FIG. 7 THE SMOKE VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS 30OC o

40 C

40 diesel tig5 tig10 tig20

30

smoke%

tig30 tig40 tig50 20

10

0 1000

1500

2000

2500

rpm

FIG. 8 THE SMOKE VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE, WHEN THE FUEL TEMPERATURE IS 40OC 300 diesel tig5 250

tig10 tig20

o

gas temperature ( C)

tig30 200

tig40 tig50

150

100

50

0 1000

1500

2000

2500

rpm

FIG. 9 THE GAS TEMPERATURE VARIATION ON DIFFERENT ENGINE RPM REGARDING TO THE MIXTURE

In the case of 30 oC as fuel temperature:

1500rpm.

From figure 1 it can be noticed that the most constant behaviour appeared in the mixture of tig40, while the best behaviour appeared in the case of diesel at

From figure 3 it can be noticed that the biggest reduction of HC emissions regarded to diesel presented in the mixture of tig 40. Figure 5 show that

239


International Journal of Energy Scie nce Vol. 2 Iss. 6, December 2012

the biggest reduction of NO emissions regarding to diesel appeared in the mixture of tig40. Finally, from figure 7 it can be said that the biggest reduction of smoke emissions regarding to diesel appeared in the mixtures of tig30 and tig40. In the case of 40 oC as fuel temperature: From figure 2 it is clear that mixtures tig5, tig10, tig20, tig30, tig40 and tig50 presented lower CO emissions regarding to diesel. From figure 4, it can be seen a reduction of HC emissions when using different mixtures than diesel. In figure 6 it is also presented a reduction of NO emissions regarding to diesel with the exception of the engine functioned on 2000 rpm, in where the diesel presented lower NO emissions than the mixtures. Finally, from figure 8, it can be seen that mixtures tig10, tig20, tig30, and tig50 presented lower smoke emissions than diesel. However, when the engine functioned on 1000, 1500 and 2000 rpm, the mixture tig40 presented higher smoke emissions than diesel. On the other hand, the mixture tig5 presented lower smoke emissions than diesel with the exception of the engine functioned on 2500 rpm, in where the smoke emissions were higher than diesel. From the above figures it can be concluded that the use of different mixtures can constitute changes to CO, HC, and NO and smoke too. It is also important to mention that there were no changes in the rounds of the engine, as well as in the supply of water during the use of mixtures. As far as the gas emissions temperature (fig. 9) and the fuel consumption is concerned, did n ot observed any changes with the use of different mixtures on the different fuel temperatures.

References [1]

stroke outboard e ngine using gasoline - e thanol mixtures”. Transaction of SAE, Book SP-1565, 2000. [2]

It is also important to mention, that during the combustion of the mixtures there was not presented any reduction in the power of the engine. Finally, it has not been presented engine malfunction from the directly use of fuel mixtures diesel - used vegetable oil.

240

C. Arapatsakos, “Testing the tractor engine using die se l – e thanol mixtures unde r full load conditions”. International Journal of Heat & Technology, Vol. 19, n.1, 2001.

[3]

C. Arapatsakos, A. Karkanis, P. Sparis, “Gas emissions and e ngine be havior whe n gasoline alcohols

mixtures

are

use d”

Journal

of

Environme ntal Technology, Vol. 24, pp. 1069-1077. [4]

C.

Arapatsakos,

A.

Karkanis,

P.

Sparis,

“Environme ntal Contribution of Gasoline - Ethanol Mixture s” WSEAS Transactions on Environme nt and De ve lopme nt, Issue 7, Volume 2, July 2006. [5]

S. Siddharth. “Green Ene rgy-Anaerobic Dige stion. Converting

Waste

to

Electricity”

WSEAS

Transactions on Environme nt and De ve lopme nt, Issue 7, Volume 2, July 2006. [6]

William Ernest Schene we rk “Automatic DRAC LMFBR to Spee d Lice nsing and Mitigate CO2” WSEAS

Transactions

on

Environme nt

and

De ve lopme nt, Issue 7, Volume 2, July 2006. [7]

Timothy T. Maxwe ll and Jesse C. Jones “Alte rnative fue ls: Emissions,

Economics and Pe rformance”

Publishe d by SAE, 1995. [8]

C.

Arapatsakos,

A.

Karkanis,

P.

Sparis,

Environme ntal pollution from the use of alte rnative fue ls in a four-stroke e ngine , International journal of

Conclusion The use of mixtures diesel-used vegetable oil has as result the gas emissions variation. Better behaviour presented in the mixtures of tig30 and tig40. The density and viscosity of those mixtures did not create any problems in the spraying of fuel. As it has already been mentioned above the different fuel temperatures (30 oC, 40 oC) differentiate the gas emissions.

C. Arapatsakos, “Air and wate r influe nce of two

e nvironme nt and pollution 21 (2004) 593-602. [9]

C. Arapatsakos, A. Karkanis, P. Sparis, Tests on a small four e ngine using gasoline -ethanol mixtures as fue l, Advances in air pollution 13 (2003) 551-560.

[10]

C.

Arapatsakos,

A.

Karkanis,

P.

Sparis,

Gas

emissions and e ngine be haviour whe n gasoline alcohol

mixtures

are

use d,

Environme ntal

technology 24 (2003) 1069-1077. [11]

C. Arapatsakos, D. Christoforidis, A. Karkanis, The use of ve ge table oils us fue l on diese l e ngine International journal of heat and technology. Vol 29, No 1, pp. 25-31, 2011.


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Catalytic Pyrolysis by Heat Transfer of Tube Furnace for Produce Bio-Oil Kittiphop Promdee1,2* , Tharapong Vitidsant3 1

Inter-Department of Environmental Science, Chulalongkorn University, Bangkok, Thailand

2

Department of Environmental Science, Chulachomklao Royal Military Academy, Nakorn Nayork, Thailand

3

Department of Chemical Technology, Chulalongkorn University, Bangkok, Thailand kp19_89@msn.com; 2 kp19_89@msn.com ; 3 tharapong.v@chula.ac.th

1

Abstract Catalytic pyrolysis by heat transfe r mode l can be solve d the control tempe rature in tube furnace to produce bio-oil by continuous pyrolysis process and this study conce rn the products yie ld of bio-oil from mixe d biomass consist of Cogongrass, Manilagrass and the leaf of trees, which conducte d te mperatures in the range of ~ 400-550°C, considering the feed rate of 150, 350, and 550 rpm (r·min−1 )]. Pre liminary result of proximate analysis was founde d that the high volatile matter, low ash and moisture . The products yie ld calculation showe d that the liquid yie ld of bio-oil was highest of 55.60 %, and 45.45%., at 350 rpm and 550 rpm., re spective ly, the solid yie ld of bio-oil was highest of 27.35 %, at 350 rpm, and the gas yie ld of bio-oil was highest of 43.60 %, at 150 rpm. Indicate d that biomass from mixe d biomass had good rece ive d yie lds because of low solid yie ld and gas yie ld and high liquid yie ld. The compounds de tecte d in bio-oil from mixe d biomass showe d that the functional groups, especially; phe nols. For the purpose that; in this research not only conce rn the fee d rate and the heat transfer for contact biomass but also conce rn the control gas flow and tempe rature balance d. Keywords Catalytic Pyrolysis; Heat Transfer; Continuous Pyrolysis Reactor; Received Oil Yield

Introduction This research was conducted by using mixed biomass transformed to bio-oil by continuous pyrolysis reactor on standard criteria and analysis the properties of material and products. In present, the fuel is being concerned in every country [1-3]. Now we are looking at the fuel which synthesized from natural matter, especially; residual plant[5-6], by using the pyrolysis method combined with the theory of heat transfer for control the temperature balance in the continuous pyrolysis reactor (tube furnace)(Fig 1). The fuels from natural matter have a good solve and can reduce a

waste in widespread areas of central part of Thailand. Continuous pyrolysis reactor is a one excellent of technology for synthesized bio-oil [7-9]. In this case want to produce bio-oil in high potential performance of yield and properties by applied the heat transfer model for control some criteria of reactor to an generate the alternative energy source.

FIG. 1 CONDUCTION IN A CONTINUOUS PYROLYS IS CYLINDER WITH UNIFORM HEAT GENERATION

To determine the temperature distribution in the cylinder reactor, the appropriate from of the heat equation. For constant thermal conductivity k, Equation (1) reduces to

1 d  dT  q • . = r =0 + r dr  dr  k

(1)

Separating variables and assuming uniform generation, this expression maybe integrated to obtain

r⋅

q• 2 dT r + C1 = 2k dr

(2)

Repeating the procedure, the general solution for the temperature distribution becomes

T (r ) = −

q• 2 r + C1 ln r + C 2 4k

(3)

To obtain the constants of integration C1 and C2, we apply the boundary conditions

241


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

dT dr

r =0

= 0 and T (ro ) = Ts

(4)

From the foregoing symmetry condition at r = 0 and Equation (2), it is evident that C1 = 0. Using the surface boundary condition at r = r o with Equation (3), we then obtain

C 2 = Ts +

q• 2 ro 4k

(r·min −1). The bio-oil product were analyzed by Ultimate analyzer, Proximate analyzer, calculate the received oil yields and analyze the chemical compound by Gas Chromatography with Mass Spectrometer.

(5)

The temperature distribution is therefore

 q•r 2  r 2  T (r ) =  o  1 − 2  + Ts  4k   ro 

(6)

Evaluation Equation (6) at the centerline and dividing the result into Equation (6), we obtain the temperature distribution in nondimensional form

 T (r ) − Ts  r   = 1−    To − Ts   ro 

2

(7)

Where To is the centerline temperature. The heat rate at any radius in the pyrolysis cylinder may, of course, be evaluated by using Equation (6) with Fourier’s law [10,11]. To relate the surface temperature, Ts, to the temperature, T∞, of the cold fluid, either a surface energy balance or an overall energy balance may be used. Choosing the second approach, we obtain

q o (πro L = h(2πro L)(Ts − T∞ ) 2

Proximate Analysis Proximate analysis is the most used analysis for characterizing biomass in connection with their utilization, this experiment was analysis by ASTM D 3173-3175. The process are determined the distribution of products obtained when the sample is heated under specified conditions. Proximate analysis separates the products into 4 groups: (1) moisture, (2) volatile matter, (3) fixed carbon, the nonvolatile fraction of char, and (4) ash. Ultimate Analysis

Or

 q•r  Ts = T∞ +    2h 

FIG. 2 SCHEMATIC DIAGRAM OF EXPERIMENT SETUP: 1. PYROLYS IS REACTOR (TUBE FURNACE) 2. NITROGEN TANK 3. ROTAMETER 4. HOPPER 1,2 5. SEPARATOR 6. CONDENSER 7. FLAS K IN ICE BUCKET 8. ELECTRICAL COIL HEATER WITH TEMPERATURE CONTROLLER 9. ENCLOSED DEIONIZER WATER TANK 10. VACUUM PUMP

(8)

The foregoing approach may also be used to obtain the temperature distribution in cylindrical and in solid spheres for a variety of boundary conditions [11]. Experimental

In the experiment was analysis form of element components of bio-oil concerned with determination of only Carbon (C), Hydrogen (H) and Nitrogen (N) in a sample, these analyzed by Ultimate analyser [12-13]. Received Oil Yield

Feedstock and Experimental Set-up Preparation of mixed biomass, crust and bring to oven at 100 oC for ~ 2 hr until it is completed dry or less than 5 percent moisture. The samples were separated through a sieve to the approximate 450-1,000 microns. The samples were fed to continuous reactor (Fig. 2), for pyrolysis process at ~ 400-550 oC and control the N2 flow rate around 0, 50, 100, 150, 200, 400 ml/hr and feed samples averaging of 150, 350, and 550 rpm

242

WLiq  = 100 ×    Wini 

% Liquid yield

% Solid yield

W  = 100 ×  R  Wini 

= 100− % Liquid yield

% Gas yield yield

- % Solid

W ini

=

Initial weight

WR

=

Residual solid weight

W Liq

=

Liquid product weight


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Chemical analysis Gas Chromatography with Mass Spectrometer, GCMS was used to analyse the light components in biooil and investigating the molecular compositions qualitatively [14-15]. The analyses detective and identify organic compounds both aliphatic hydrocarbon and aromatic hydrocarbon. Results and Discussion Preliminary, proximate analysis of mixed biomass used in the three species (Cogongrass, Manilagrass and the leaf of trees) was founded that the fixed carbon of mixed biomass was 17.28 wt.%, which will have a major effect on the quality of bio-oil as well. The other three proximate analysis as following; The moisture, ash and volatiles of mixed biomass were 2.51, 17.00 and 63.21 wt.%, respectively (Table. 1). The results showed that the stability for the range of material compound in mixed biomass can be synthesized bio-oil in high efficiency, because consist of the high volatile matter and low ash and moisture. TABLE 1 PROXIMATE ANALYS IS AND ULTIMATE ANALYS IS OF MIXED BIOMASS P roximate analysis (wt.%)

mixed biomass

Moisture

2.51

Ash

17.00

Volatiles Fixed c arbon

FIG. 3 GAS YIELD OF BIO-OIL OBTAINED FROM MIXED BIOMASS

Liquid yield of bio-oil obtained from mixed biomass was highest of 55.6 %, at 350 rpm. And the another of liquid yield obtained from mixed biomass were 29.55 and 45.45 %, at 150 and 550 rpm., respectively (Fig. 4). Indicated that the liquid yield of bio-oil obtained from mixed biomass was high volume (> 50 %) by the heat control in continuous pyrolysis reactor and can be improving to high efficiency of bio-oil production on next step.

mixed

Ultimate analysis (wt.%)

biomass

C

38.23

H

5.27

63.21

N

1.00

17.28

O

55.16

The ultimate analysis of mixed biomass was found that the element contents as following; carbon, hydrogen, nitrogen and oxygen were 38.23, 5.27, 1.00 and 55.16 %, respectively (Table. 1)., according to the result of safflower seed [12,13] showed the carbon, hydrogen, nitrogen, and oxygen of 49.5, 6.9, 3.0, and 40.6, respectively. The result of products yield (3-phase; gas, liquid and solid) of bio-oil by during pyrolysis, which takes place at temperatures in the range of ~ 400-550°C, to compare the received oil yield from mixed biomass at a feed rate of feed rate of 150, 350 and 550 rpm; revolutions per minute (r·min −1)]. Preliminary calculate of the product oil yield of mixed biomass, the result showed that the gas yield of bio-oil obtained mixed biomass were 43.6, 19.05 and 29.25 %, at 150, 350 and 550 rpm., respectively., (Fig. 3).

FIG. 4 LIQUID YIELD OF BIO-OIL OBTAINED FROM MIXED BIOMASS

FIG. 5 SOLID YIELD OF BIO-OIL OBTAINED FROM MIXED BIOMASS

243


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Solid yield of bio-oil obtained from mixed biomass was highest of 27.35 %, , at 350 rpm. And the another of solid yield obtained from mixed biomass were 25.85and 24.3 %, at 150 and 550 rpm., respectively (Fig. 5). TABLE 2 COMPOUNDS DETECTED IN BIO-OIL FROM MIXED BIOMASS Compound

1,2-Benzenediol

%

0.82

2-Cyc lopenten-1-one, 2,3-dimethyl-

1.08

Phenol, 2,3-dimethylPhenol, 2,4-dimethylPhenol, 2,5-dimethylPhenol, 2,6dimethoxyPhenol, 2,6-dimethylPhenol, 2-ethylPhenol, 2-methoxy-

×

2.23

C8 H10 O

122.16

×

1.41 0.42

*c an not determined

244

× C8 H10 O

122.16

12.47 C8 H10 O

122.16

0.85

C8 H10 O

122.16

3.81

2.87

× ×

*

Phenol, 2-methoxy-4propyl-

Phenol, 4-ethyl-2methoxy-

× 94.11

0.72

Phenol, 4-ethyl-

×

C6 H6 O

Phenol, 2-methoxy-4methyl-

Phenol, 3-methyl-

112.12

19.78

*

Phenol, 3,4-dimethyl-

Detection ×

C6 H8 O2

Phenol,2-methoxy-4(1-propenyl)-, (E)-

Phenol, 2-methyl-

MW

0.5

1,2Cyclopentanedione, 3-methyl-

Phenol

formula

× ×

C10 H12O 2

164.19

C10 H12O 2

164.19

×

C10 H12O 2

164.19

×

C7 H8 O

108.13

×

0.62

Con clusions The continuous pyrolysis reactor for produce bio-oil from mixed biomass showed the proximate analysis of mixed biomass presented a high volatiles content and showed a moderate level of fixed carbon. The amount of the elemental composition of mixed biomass, can found the concentration of carbon was relatively high volume. The products oil yield showed that liquid yield of bio-oil obtained from mixed biomass was a good result was 55.6 %, at 350 rpm. Also the result of solid yield bio-oil obtained from mixed biomass as same a high volume at 350 rpm. The compounds detected in bio-oil from mixed biomass can found the phenols, alcohols, and ketones, especially; phenols group. Thus, in this research, the process of continuous pyrolysis depended on the mechanism of heat transfer with cylinder shape. If control the N2 flow, control temperature system balance, according to as good as the heat transfer model, the yields and qualities of bio-oil should to be high efficiency and the another concern that to the overall performance system of the continuous pyrolysis reactor. ACKNOWLEDGMENT

×

3.86

C7 H8 O

108.13

×

1.54

C8 H10 O

122.16

×

C9 H12 O2

152.18

×

2.23

The compounds detected in bio-oil from mixed biomass showed that the hydrocarbon compounds compose of hydroxyl and carboxyl groups, especially; phenols (Phenol, 2,3-dimethyl-, Phenol, 2,6dimethoxy-4-(2-propenyl)-, Phenol, 2-ethyl-4-methyl-, Phenol, 2-methoxy-, Phenol, 3-methyl-, Phenol, 4ethyl-2-methoxy-), alcohols, and ketones (Table. 2) as same the result of pyrolysis two energy crops [15] and the other result of pyrolysis biomass [3-516,17,18,19,20].

This work was supported by the Higher Education Research Promotion and National Research University Project f Thailand, Office of the Higher Education Commission (Project Code : EN272A), and thank you for Department of Chemical Technology, Faculty of Science, Chulalongkorn University, were advised and supported the laboratory for experiment and analysis in this research. REFER ENCES [1]

M. Pate l, M. Ne e lis, D. Gie le n, J. Olivie r, T. Simmons, and J., The unis, “Carbon dioxide e missions from none ne rgy use of fossil fue ls: Summary of key issues and conclusions from the country analyse s,” Resources,


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Conservation and Recycling , Vol 45, Issue 3, pp.195-209, [2]

M. Garcia-Pe re z, J. She n, X. S., Wang, and C-Z Li,

re actor: Part 2. Structural characte rization of pyrolysis

“Production and fue l prope rtie s of fast pyrolysis

bio-oils,” Bioresource Technology, 99, pp.5498-5504, Se p.

oil/bio-die se l ble nds,” Fuel Processing Technology, Vol

2008. [14] P. Duan, and P. E., Savage , “Upgrading of crude algal

G. Duman, C. Okutucu, S. Ucar, R. Stahl, and J., Yanik,

bio-oil in supe rcritical wate r,” Bioresource Technology,

“The slow and

Vol 102, pp.1899-1906, Jan. 2011.

fast

pyrolysis of

che rry se e d,”

Bioresource Technology, Vol 102, Issue 2, pp.1869-1878, [4]

Composition of Bio-oils Produce d by Fast Pyrolysis of

D. Wu, S. Zhang, J. Xu and T., Zhu, “The CO 2

Two Ene rgy Crops,” Energy & Fuels, pp.2104-2109, Apr.

Re duction Effe cts and Climate Be ne fit of Be ijing 2008

2008. Knudse n, and A.D., Je nse n,”A re vie w of catalytic

Q. Lu, W-Z. Li, and X-F., Zhu, “Ove rvie w of fue l

upgrading of bio-oil to e ngine fue ls,” Applied Catalysis,

biomass

fast pyrolysis oils,” Energy

Conversion and Management, 50, pp.1376–1383, May

Sharifi, and J., Swithe nbank. Combustion of oil palm

H. Li, Q. Xu, H. Xue , and Y., Yan. 2009, “Catalytic

stone in a pilot-scale fluidize d be d re actor. Fue l

re forming of the aque ous phase de rive d from fast-

Proce ssing Te chnology, 92 (2011), p. 2219-2225.

H.S. He o, H.J. Park, J-H. Yim, J.M. Sohn, J.H. Park, S-S. Kim, C.K. Ryu, J-K. Je on, and Y-K., Park. “Influe nce of ope ration variable s on fast pyrolysis of Miscanthus sine nsis var. purpurasce ns,” Bioresource Technology, Vol 101, Issue 10, 2010, pp.3672-3677. K.

Promde e ,

T.

Vitidsant,

and

S.

Vanpe tch.

“Comparative study of some physical and che mical prope rtie s of bio-oil from Manila grass and Wate r hyacinth

transforme d

International Journal

of

by

pyrolysis

Chemical

proce ss”,

Engineering

and

Applications, Vol.3, No.1, pp.72-75. Fe b. 2012. [9]

[18] L. Me i-kue i, T. We m-tie n; S. Yi-lin and L. She au-horng.

Pyrolysis of Napie r Grass in an Induction-he ating

2009.

[8]

Vol 407, Issue s 1-2, pp.1-19, Aug. 2011. [17] R. Razuan, Q. Che n, N.K.Finne y, V.N. Russe ll, N.V.

2009.

pyrolysis of biomass,” Renewable Energy, pp.1-6, Apr. [7]

[16] P.M. Morte nse n, J-D. Grunwaldt, P.A. Je nse n, K.G.

Volume 5, pp. 280-296, Apr.2011. prope rtie s of

[6]

[15] C.A. Mulle n, A. Charle s, and A., Akwasi, “Che mical

Jan. 2011.

Summe r Olympics Gre e n Practice ,” Energy Procedia, [5]

safflowe r

(Charthamus tintorius L.) Se e d pre ss cake in a fixed-bed

91, Issue 3, pp.296-305, Mar. 2010. [3]

[13] S. Se vgi, and D., Angin, “Pyrolysis of

Jul. 2005.

K. Promdee , and T. Vitidsant, and S. Vanpetch: J. Chem Eng and App. 3, 1, Fe b 2012.

[10] F.P. Incrope ra, and D. P. De Witt. Fund of Heat and Mass

Transfer. 2 1985. [11] T.G. Hicks. Mechanical engineering formulas. 1, 2003. [12] S. Se vgi, and D., Angin, “Pyrolysis of

safflowe r

(Charthamus tintorius L.) See d press cake : Part 1. The e ffect of pyrolysis paramete rs on the product yie lds,” Bioresource Technology, 99, pp.5492-5497. Se p. 2008.

Re actor. Analytical and Applied Pyrolysis, (2010),Vol 88, Issue s 2, pp. 110-116, Jul. 2010. [19] H. Kaze mi Esfe h, B. Ghanavati, and T. Ghale Golabi.

“Prope rtie s of modifie d bitume n obtaine d from natural bitume n by adding pyrolysis fue l oil”, International Journal of Chemical Engineering and Applications, Vol.2, No.3, pp.168-172. Jun. 2011. [20] U. Je na, and K. C. Das. “Comparative e valuation of

the rmoche mical lique faction and pyrolysis for bio-oil production from microalgae . Ene rgy฿fue ls”, 25 (2011), p.5472-5482, Se p. 2011. Kittiphop Promdee was born in Ubonrachathani Province, Thailand, in 1975. He rece ive d the M.S. degree in e nvironme ntal scie nce from the school of e nvironme ntal scie nce , Kasetsart Unive rsity, Bangkok, Thailand, in 2004. He is currently a le cturer with De partment of Environme ntal Scie nce , Chulachomklao Royal Military Acade my, Nakorn Nayork, Thailand. His research inte rests include agricultural re siduals, re ne wable e nergy, fue l, e nvironme ntal e ne rgy, gree n and biomass technology and catalytic pyrolysis processes. He is curre ntly pursuing the Ph.D. degree with Inter-De partme nt of Environme ntal Scie nce , Chulalongkorn Unive rsity, and Bangkok, Thailand.

245


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Tharapong Vitidsant re ce ive d the B.S. degree and the M.S. de gre e in c he m ica l e ng ine e r ing fr om C hula lo ng kor n Unive rsity, Bangkok, Thailand. He re ce ive d the Ph.D. de gree from. Institut National Polytechnique de Toulouse (INPT), France, in 1999. He is curre ntly an associate professor with

246

De partment of Chemical Technology, Chulalongkorn Unive rsity. His research inte rests include re ne wable ene rgy, fue l e nergy, catalytic pyrolysis processes and reaction e ngineering.


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

CMOS Bandgap Reference and Current Reference with Simplified Start-Up Circuit Guo-Ming SUNG, Ying-Tzu LAI, Chien-Lin LU Department of Electrical Engineering, National Taipei University of Technology 1, Sec. 3, Chung- Hsiao E. Rd. Taipei 10608 Taiwan gmsung@ntut.edu.tw; 2t5319013@ntut.edu.tw; 3lu8169@yahoo.com.tw

1

Abstract This pape r prese nts a CMOS bandgap re fere nce and current re fe re nce base d on the resistor compe nsation. The propose d archite cture consists of various high positive te mpe rature coe fficie nt (TC) resistors, a two-stage ope rational transconductance amplifie r (OTA) and a simplifie d start-up circuit in 0.35-µm CMOS process. In the propose d bandgap re fe re nce and curre nt refe re nce , nume rous compe nsate d re sistors, which have a high positive te mperature coefficie nt (TC), are adde d to the parasitic n-p-n and p-n-p bipo lar junction transistor devices, to ge nerate a te mpe ratureinde pe nde nt voltage re fere nce and curre nt re fere nce . With the simplifie d start-up circuit, the propose d resistorcompe nsation bandgap re fe rence and curre nt re fere nce can be starte d at 43 ns at a minimum supply voltage of 1.35 V. The measureme nts verify the current re fe rence of 735.6 nA, the voltage re fere nce of 888.1 mV, and the power consumption of 91.28 µW at a supply voltage of 3.3 V. The voltage TC is 49 ppm/℃ in the tempe rature range from 0oC to 100oC and 12.8 ppm/℃ from 30oC to 100oC. The current TC is 119.2 ppm/℃ at tempe ratures of 0oC to 100oC. Measure ment results also demonstrate a stable voltage re fe re nce at high te mpe rature (> 30oC), and a constant curre nt re fe rence at low te mperature (< 70oC). Keywords Bandgap Reference; Current Reference; Resistor Compensation; Temperature Coefficient; Start-up Circuit

Introduction A bandgap reference is extensively adopted in several integrated circuits, including analog, mixed-mode and memory circuits. In CMOS bandgap reference design, the sub-1-V curvature-compensated CMOS bandgap reference is favored for use with resistive subdivision methods [1-3], parasitic n-p-n and p-n-p bipolar junction transistor devices [4], and the compensation approaches associated with layout, operational amplifiers, and V EB linearization [5-7]. However, the low-voltage bandgap reference frequently functions with a higher temperature coefficient than the conventional bandgap reference [4]. To reduce the

temperature coefficient further, many curvature compensations have been introduced. Guan et al. developed a current mode curvature-compensated BGR (bandgap reference) using the trimming approach [8]. Leung et al. introduced second-order curvature compensation based on resistors with opposing temperature coefficients [9]. Audy introduced a third- order curvature compensation with a low-TCR resistor in parallel with a high-TCR resistor and in series with low-TCR tail resistors [10]. Malcovati had designed a high-order curvature temperature compensation method with Malcovati topology [7]. Chen also presented a programmable and high-precision temperature independent current reference by adding a positive TC current to a negative TC circuit [11]. However, the simulation result had not been verified yet. Start-up circuits are required to prevent the bandgap reference from operating at the zero point. The components of a start-up circuit must be limited to three or less. Xing et al. developed a start-up circuit that comprised three NMOSs, M S1-MS3 [5]. Xiaokang Guan et al. also introduced a start-up circuit with a resistor R6 and two PMOSs, M7-M8 [8]. Ker further developed two start-up circuits in the proposed bandgap reference. One consists of one PMOS and two NMOSs, MSN1-MSN3. The other comprises a NMOS and two PMOSs, M SP1-MSP3 [4]. Unfortunately, as is well known, the rise times of the proposed start-up circuits are long. More effort must be made to design a highspeed start-up circuit. Additionally, the current trend in industry is toward new bandgap references with simultaneous voltage reference and current reference [3], [12]. If a temperature-independent current reference is required, then the bandgap reference must generally be divided by a resistance. However, the resistance depends on temperature. Accordingly, a current reference also depends on a curvaturecompensation method. This work presents a resistor-

247


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

compensation CMOS bandgap and current reference with a current reference of 735.6 nA and a voltage reference of 888.1 mV at a supply voltage of 3.3 V in a standard 0.35-µm CMOS process. The proposed bandgap and current reference, which comprises a two-stage operational transconductance amplifier (OTA) with differential NMOS input stage, is validated. Notably, the voltage TC is 49 ppm/℃ in the temperature range from 0 oC to 100 oC, and 12.8 ppm/℃ from 30 oC to 100 oC; the power consumption is 91.28 µW. Hence, section II describes the basic principles and circuit designs associated with the proposed bandgap and current reference. Section III presents and discusses experimental results. Conclusions are finally drawn in Section IV, along with recommendations for future research. Basic Princip les and Circuit Design Basic Principles of CMOS Bandgap Reference The temperature-independent bandgap reference is the conventionally adopted voltage reference because temperature commonly skews the operating point and affects noise in the semiconductor. A bandgap reference working with zero TC, which has both positive TC and negative TC, is therefore required. Figure 1 depicts the basic framework of a bandgap reference [13-15]. The output reference voltage, V ref, is

Vref = VEB + K ⋅ Vt

(1)

where K is a constant which is normally equal to 17.2 [13], V EB is the voltage difference between the emitter and the base in bipolar junction transistors (BJT) and V t is the thermal voltage. V EB typically exhibits a negative-TC characteristic, while thermal voltage V t usually has a positive TC. V t is realized from the difference between two emitter-base voltages, ∆V EB, which is proportional to the absolute temperature (PTAT). However, the output voltage of the bandgap reference suffers from variation with temperature, even when curvature-compensation is considered. To solve this problem, an improved cascading CMOS bandgap reference (BGR) with second-order curvature-compensated circuit has been presented [16]. However, it does not plot a temperature-independent current reference Iref. To have both voltage reference and current reference in a temperature-independent bandgap reference, the curvature-compensated method must be further investigated.

248

FIG. 1 BAS IC FRAMEWORK FOR BANDGAP REFERENCE

Proposed Resistor Compensation Circuit Figure 2 presents the schematic of the proposed temperature-independent bandgap reference and current reference where a two-stage operational transconductance amplifier (OTA) is used to replace a traditional cascade current mirror. Notably, Resistors, R3 and Rb3, and BJT, Q3 , compensate for the output voltage reference, V ref, in what is called second-order curvature compensation [9]. Resistors Ra1 and Ra2 have two purposes. The first is to increase the input voltages, V in+ and V in-, of the N-type stage OTA to ensure that the OTA works properly. The second is to compensate for the temperature-dependent variation of V in+ and V in-. The OTA is employed to equality V in+ and V in-. Transistors Q2 and Q1 are vertical PNPs with a base-emitter area ratio of 5:2, passing the same current, such that the current though R 2 is PTAT. Transistors Q1 and Q3 are identical. Rb1, R b2 and Rb3 are added to produce the second-order curvature compensation circuit [16]. Notably, the base-emitter area ratio of Q2 and Q1 is smaller than the traditional ratio, because current compensation is performed using several resistors. Therefore, the positive TCs of Rb1, Rb2 and Rb3 compensate for the negative temperature coefficients of Q1, Q2 and Q3, respectively. MOSFETs M1-M4, are also identical to each other. They equalize the currents I1, I2, I3 and Iref, and provide a temperature-independent current reference Iref. Restated, the proposed bandgap and current reference simultaneously provides both a temperatureindependent voltage reference V ref and temperatureindependent current reference Iref. Next, the resistor-compensation circuit is described in deta il. Th e emitt er-base volta ge V E B of BJT has a n ega t iv e TC w h er ea s t h e em itt er -b a s e v olta ge difference ∆V BE and all resistances have a positive TC.


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

FIG. 2 SCHEMATIC OF PROPOSED RES ISTOR-COMPENSATION BANDGAP AND CURRENT REFERENCE WITH VARIOUS COMPENSATED RES IS TORS AND A TWO-S TAGE OPERATIONAL TRANSCONDUCTANCE AMPLIFIER (OTA)

As the temperature (T) increases, the voltages, V C, V D and V E, of nodes C, D and E, will drop because of the negative TC of V EB; meanwhile, the values of compensated resistors, Rb1, Rb2 and Rb3, are increased. Three compensated currents flow into the three BJTs, Q1, Q2 , and Q3, respectively, compensating for the emitter currents, which increase according to a positive TC such that IR2 = (V EB1-V EB2)/R2 = ∆V EB/R2. This increase yields three temperature-independent currents, I1, I2 and I3 . Simultaneously, voltages V in-, V in+ and V ref, will be compensated for to maintain constant with resistances, Ra1, Ra2 and R3, because their TCs are positive. Hence, both voltage reference and current reference will be independent of temperature. We here need to emphasize that the compensated resistances, Rb1, R b2 and Rb3 , must be very large and be fabricated with n-wells. However, the resistance R2, which is fabricated with n +-diffusion, is low and a TC of around one-third of that of n-well. Figure 3 presents the fivecorner simulations of current reference Iref as a function of temperature for the proposed bandgap and current reference. The proposed resistor-compensation circuit is analyzed mathematically. For simplicity, consider components Ra2, Rb2, R2 and Q2 in Fig. 2. Suppos e that OTA is ideal and that the counterpart resistors are equal, such that V in+ = V in-, Ra1 = Ra2 and Rb1 = Rb2; now, I1 = I2 . Therefore,

VEB1 + VRa1 = VEB 2 + VRa 2 + VR 2

(2)

VRa1 = VRa 2 = I Ra1 × Ra1 = I Ra 2 × Ra 2

(3 )

I Ra 2 = I Rb 2 + I R 2 = I Rb1 + I R 2

(4)

FIG. 3 OUTPUT CURRENT REFERENCE VERS US TEMPERATURE OBTAINED BY FIVE-CORNER S IMULATIONS OF THE PROPOSED BANDGAP AND CURRENT REFERENCE

where VEB1 and VEB2 are the emitter to base voltages of the bipolar transistors, Q1 and Q2; V Ra1, V Ra2 and V R2 are the voltage declines across Ra1, Ra2 and R2, and IRa1, IRa2 and IR2 are the currents through Ra1, Ra2 and R2, respectively. Differentiating Eqn . (4) with respect to temperature (T) yields

∂I Ra 2 ∂I Rb 2 ∂I R 2 = + ∂T ∂T ∂T

(5)

If a temperature-independent current reference IRa2 is required, ∂IRa2/∂T = 0 is set; thus,

∂I Rb 2 ∂I R 2 + =0 ∂T ∂T

(6)

where IRb2 is a negative-TC current because IRb2 = IRb1 = V EB1/Rb1, whereas IR2 is a positive-TC current because IR2 = (V EB1-V EB2)/R2 = ∆V EB/R2. Combining the first differential item, ∂I Rb2/∂T with the second differential item, ∂IR2/∂T, yields a temperature-independent current IRa2. Passing through the current mirror, which has MOSFETs M1-M4, a temperature-independent current reference Iref is generated in the proposed bandgap and current reference. The voltage reference V ref associated with the bandgap and current reference can be expressed as,

Vref = I R 3 × R3 + V EB 3

(7)

where VEB3 is the emitter to base voltage of the BJT Q3 and IR3 is the current through resistor R3. Notably, IR3 equals I3 , which is a temperature-independent current. Differentiating the above equation with respect to temperature (T) yields

∂Vref ∂T

= R3 ×

∂I R 3 ∂R ∂V + I R 3 × 3 + EB 3 ∂T ∂T ∂T

(8)

249


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

A temperature-independent bandgap reference is obtained by setting ∂V ref/∂T ≈ 0 and ∂IR3/∂T ≈ 0. Therefore,

I R3 ×

∂R3 ∂VEB 3 + =0 ∂T ∂T

(9)

where R3 is a positive-TC resistor, whereas V EB3 is a negative-TC voltage. Properly setting the value of R3 yields a temperature-independent voltage reference, ∂V ref/∂T ≈ 0. Notably, R3 not only compensates for the temperature variation of V EB3, but also adjusts the output voltage as required. Moreover, resistors R 3 and Rb3 are fabricated using an n-well. The resistance R3 exceeds Ra2, while Rb3 is less than Rb2. Figure 4 schematically depicts the complete circuit of the proposed bandgap and current reference, which simultaneously provides temperature-independent voltage reference V ref and current reference Iref. The left-hand side of Fig 4 presents an OTA circuit with Ntype input [14], which is used to ensure that the positive input V in+ equals the n egative input V in- of OTA. Notably, the MOSFET, Mr , must be operated in the triode region as a resistor. The simulations of the two-stage telescopic OTA indicate those dc gain, bandwidth and phase margins are 61.35 dB, 9.52 MHz and 64 o, respectively. If an input offset voltage of OTA, owing to asymmetries, is considered, it will introduce error in the voltage reference V ref . Thus,

R  Vref = VEB 3 +  3  (VT × ln m − VOS ) + I Rb 2 R3 (10)  R2  where V OS is the input offset voltage of OTA, V T is the thermal voltage, and m is the base-emitter area ratio of Q2 to Q1, producing m ≈ 5/2. In this work, two methods are employed to lower the effect of V OS. One is that the OTA is a telescopic topology to minimize the offset because of symmetry and the other is that the layout of OTA incorporates common centroid and dummy in a large device. Proposed Start-up Circuit As shown in Fig. 2, the start-up circuit comprises three MOSFETs, Mn, Mp and MS, where the gate-source voltage V GS, of Mn is shorted to turn off Mn with a huge resistance. Restated, the gate voltage of M S, V MS,G, is half of the supply voltage V DD in the initial stage because both Mp and Mn are cut-off. As the supply voltage V DD increases slowly, V OTA,out , which is connected to the gate terminal of Mp, traces V DD because the OTA is off and the voltage difference

250

between source (V DD ) and gate of Mp is about zero due to the Cgs of Mp, M1 and M2. When MS is turned on under the condition V DD-VMS,G ≥ |V thp| with a threshold voltage of PMOS, |V thp|, a small conduction current IMS flows. Then the conduction resistance between drain and source, r DS, of Mp is reduced. By comparing with the huge resistance of Mn, the gate voltage of Ms increases. Mp flows current until V DS is reduced to nearly 0 V. The voltage of V MS,G keeps V DD due to parasitic capacitance. Finally, the operation mode of Mp is changed from saturation to the linear region and Ms will be turned off. Note that the Mn is used not only to speed-up the rise time, but also to save power without driving current. Figure 5 plots the simulated results concerning the start-up circuit over time, where the symbols ▲, ■, ★ and ◆ represent the power-supply voltage (V DD), the gate voltage of M S (V MS,G), the output voltage of OTA (V OTA, out ) and the source current of Ms (IMS), respectively. Importantly, the minimum power supply, V DD,min, is approximately 1.35 V, and the difference between the power-supply voltage (V DD) and the gate voltage of M S (V MS,G), V DD – V MS,G, exceeds 0.7 V; the start time is around 43 ns, and the source current of Ms , IMS, falls to zero. Experimental Results The proposed bandgap and current reference was fabricated with 0.35-µm 2P4M CMOS process. The layout is carefully considered to minimize the mismatches of the resistor and that of the transistor. Additionally, resistors Ra1 and Ra2 are implemented using n-well because of its large temperature coefficient, while n+-diffusion occurs in resistor R2 with a small temperature coefficient. The temperature-dependent performance was measured over operating temperatures from 0 oC to 100 oC. Figures 6 and 7 plot the measured voltage reference V ref and current reference Iref, respectively, of the proposed bandgap and current reference against temperature. Figure 6 indicates that the measured voltage reference is proportional to the temperature in the range 0 oC to 30 oC, and is roughly constant from 30 oC to 100 oC. Th e temperature coefficient of the voltage reference is approximately 49 ppm/oC and the maximum variation of V ref is approximately 4.35 mV at a supply voltage of 3.3 V and temperatures from 0 oC to 100 oC. The corresponding values are 12.8 ppm/oC and 0.8 mV from 30 oC to 100 oC. Figure 7 reveals that the measured current reference is almost constant from 0 oC to 70 oC, but increases rapidly in


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

FIG. 4 COMPLETE CIRCUIT OF THE PROPOSED BANDGAP AND CURRENT REFERENCE WHICH S IMULTANEOUS LY PROVIDES TEMPERATURE-INDEPENDENT VOLTAGE REFERENCE AND CURRENT REFERENCE

temperature range 70 oC to 100 oC. Th e temperature coefficient of the current reference is about 119.3 ppm/oC and the maximum variation of Iref is about 8.77 nA at temperatures from 0 oC to 100 oC. Additionally, the variations in voltage reference are measured, and plotted in Fig. 8 versus the power supply voltage from 0 V to 3.6 V. The measurements also demonstrate that output voltages from 0.871 V to 0.888 V are roughly proportional to the power supplied from 1.4 V to 3.4 V. Notably, the proposed bandgap and current reference can be started up at a supply voltage of 1.35 V, and so is suitable for operating with battery cell.

(a)

Table I pres ents the measurements of the proposed bandgap and current reference. Table II compares the proposed resistror-compensation bandgap and current r efer ence pr es ent ed h er ein w ith ot h er pr ior- art curvature-compensation bandgap references. In table II, the best voltage TC of 12.8 ppm/℃ is superior to that of other bandgap references, except [5], and the best current TC of 119 .2 ppm /℃ is a cceptab le by com pa r in g w it h r ef er en ce [ 1 9 ] . Ba s ed on t h e com par is on , th e pr opos ed ban dga p an d curr ent reference is suitable for use at temperatures of over 30 oC. However, the averaged current reference is lower. Note that large compensated resistors, Rb1, Rb2 and Rb3, were selected herein to reduce power consumption of

(b) FIG. 5 S IMULATED RES ULTS CONCERNING START-UP CIRCUIT OVER TIME IN NS. (A) VARIATIONS OF VDD (▲), VMS,G (■) AND VOPA,OUT(★) IN VOLT (V). (B) SOURCE CURRENT OF MS (◆) IN MICRO AMPERES (µA)

251


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

the chip and high positive-TC resistors, Ra1 and Ra2, were adopted to compensate for the temperaturedependent variation of V in+ and V in- in OTA. Figure 9 presents a die microphotograph of the proposed bandgap and current reference fabricated in a 0.35-µm CMOS process. In this chip, two capacitors, C1 and C2, are connected to power supply and voltage reference, respectively, to alleviate the unstableness.

FIG. 9 DIE MICROPHOTOGRAPH OF THE PROPOS ED BANDGAP REFERENCE AND CURRENT REFERENCE FABRICATED IN A 0.35-µM CMOS PROCESS TABLE I MEAS UREMENTS OF PROPOSED BANDGAP REFERENCE AND CURRENT REFERENCE Parameters FIG. 6 MEAS URED VOLTAGE REFERENCE (V) AS A FUNCTION OF TEMPERATURE (O C)

Typic al power supply (V)

3.3

Minimum power supply (V)

1.35

Averaged voltage reference (mV) (0℃~100℃)

888.1

Maximum variation of voltage reference (mV) (0℃~100℃)

4.35

Voltage temperature coeffic ient (ppm/℃) (0℃~100℃) Averaged voltage reference (mV) (30℃~100℃) Maximum variation of voltage reference (mV)(30℃~100℃) Voltage temperature coeffic ient (ppm/℃) (30℃~100℃) FIG. 7 MEAS URED CURRENT REFERENCE (NA) AS A FUNCTION OF TEMPERATURE (O C)

Measurements

49 888.7 0.8 12.8

Voltage referenc e settling time (V/µs)

38.1

Averaged c urrent reference (nA) (0℃~100℃)

735.6

Maximum variation of c urrent reference (nA) (0℃~100℃)

8.77

Current temperature coeffic ient (ppm/℃) (0℃~100℃) Power dissipation (µW) Chip area (µmµm)

119.2 91.28 237256

Conclusions

FIG. 8 VARIATION OF OUTPUT REFERENCE VOLTAGE (V) AGAINST S UPPLY VOLTAGE (V) FOR THE PROPOSED BANDGAP AND CURRENT REFERENCE

252

A resistor-compensation CMOS bandgap reference and current reference with a current reference of 735.6 nA and a voltage reference of 888.1 mV at a supply voltage of 3.3 V was presented. It consumes a maximum power of 91.28 µW. The voltage TC was 49


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

TABLE 2 COMPARIS ON AMONG THE CURVATURE-COMPENSATED BANDGAP REFERENCES This work

[17]

[18]

[19]

[5]

[3]

0.35 µm

0.35 µm

0.25 µm

0.18 µm

0.18 µm

0.13 µm

Typic al V DD (V)

3.3

3.3

NA

1.0

1.1

1.2

Minimum V DD (V)

1.35

NA

0.85

NA

0.90

NA

Averaged voltage reference (0℃~100℃)

888.1 mV

1173.2 mV

238.2 mV

598.5 mV

657 mV

630 mV

Voltage TC (0℃~100℃)

49 ppm/℃

47 ppm/℃

Voltage TC (30℃~100℃)

12.8 ppm/℃

(trimming)

58 ppm/℃

125 ppm/℃

10 ~ 40 ppm/℃

29 ppm/℃

735.6 nA

NA

NA

144 µA

NA

50.2 µA

119 ppm/℃

NA

NA

185 ppm/℃

NA

18 ppm/℃

0~100℃

-75~75℃

-10~120℃

0~100℃

0~150℃

-10~100℃

Tec hnology

Averaged c urrent reference(nA) (0℃~100℃) Current TC (0℃~100℃) Temperature Range

ppm/℃ at temperatures from 0 oC to 100 oC, and 12.8 ppm/℃ from 30 oC to 100 oC. The current TC was 119.2 ppm/℃ from 0 oC to 100 oC. The measurements also reveal that a good temperature-independent voltage reference V ref is realized at high temperature and a fine temperature-independent current reference Iref is performed at low temperature. With a simplified startup circuit, the proposed bandgap and current reference was verified to be effective in a standard 0.35-µm CMOS process. Restated, the proposed resistor-compensation CMOS bandgap and current reference, which is compensated with various high positive TC resistors, simultaneously provides both a temperature-independent voltage reference V ref and temperature-independent current reference Iref.

Furthermore, this work verifies that both n-well and n+-diffusion are suitable for developing a new resistorcompensation technique in bandgap reference or current reference. To further improve the performance of bandgap reference or current reference, the resistorcompensation technique can be utilized except highorder curvature compensation [5]. ACKNOWLEDGMENT

The authors would like to thank the National Science Council of the Republic of China, Taiwan, for financially supporting this research under Contract No. NSC 95-2221-E-027-138-MY3. The CIC is appreciated for fabricating the test chip and Ted Knoy is appreciated for his editorial assistance. re fe re nce with sub-1-V o pe ration,” IEEE Trans. Circuits

REFERENCES [1]

Syst. II, Express Briefs, vol. 53, no. 8, pp. 667–671, August

H. Banba, H. Shiga, A. Ume zawa, T. Miyaba, T. Tanzawa, S. Atsumi and K. Sakui, “A CMOS bandgap

2006. [5]

voltage re fere nce circuit with sub-1-V ope ration,” IEEE J.

precision CMOS bandgap re fere nce ,” Norchip, pp. l-4,

Solid-State Circuits, vol. 34, no. 5, pp. 670–674, May 1999. [2]

[3]

G.

Giustolisi,

“A low-voltage

low-powe r voltage

Nov. 2007. [6]

K. N. Le ung and P. K. T. Mok, “A sub-1-V 15-ppm/C

re fe re nce base d on subthreshold MOSFETs,” IEEE J.

CMOS bandgap voltage re fere nce without re quiring

Solid-State Circuits, vol. 38, no. 1, pp. 151–154, Jan. 2003.

low thre shold voltage device ,” IEEE J. Solid-State

D. O. Han, J. H. Kim and N. H. Kim, “De sign of

Circuits, vol. 37, pp. 526-530, April 2002.

bandgap re fere nce and curre nt re fe rence ge ne rator with

[4]

X. Xing, Z. Wang and D. Li, “A lo w voltage high

[7]

P. Malcovati and F. Malobe rti, “Curvature-compe nsate d

low supply voltage ,” in Proc. ICSICT’08, Oct. 2008, pp.

BiCMOS bandgap with 1 V supply voltage ,” IEEE J.

1733–1736.

Solid-State Circuits, vo1. 36, no. 7, pp. l076-1081, July

M.

D.

2001.

Ke r and J. S. Che n, “Ne w c urvature -

compe nsation

te chnique

for

CMOS

bandgap

[8]

X. Guan, A. Wang, A. Ishikawa, S. Tamura, Z. Wang

253


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

and C. Zhang, “A 3V 110uW 3.1ppm/℃ curvature -

[9]

[19] A. Be ndali and Y. Aude t, “A 1-V CMOS current

compe nsate d CMOS bandgap re fe re nce ,” IEEE Int.

re fe re nce

Symp. Circuits Sys. pp. 2861-2864, 2006.

compe nsation,” IEEE Trans. on Circuits and Systems I,

K. N. Le ung, P. K. T. Mok and C. Y. Le ung, “A 2-V 23-

vol. 54, pp. 1424-1429, 2007.

with

te mpe rature

and

process

uA 5.3-ppm/oC curvature -compe nsate d CMOS bandgap voltage re fe rence,” IEEE J. Solid-State Circuits, vol. 38, no. 3, pp. 561-564, March 2003. [10] J. M. Audy, “Bandgap voltage re fere nce circuit and

me thod with lo w TCR resistor in paralle l with high TCR and in se ries with lo w TCR portions of tail resistor,” U.S. Pate nt 5291122, Mar. 1, 1994. [11] J. Chen and B. Shi, “1 V CMOS current re fere nce with 50

ppm/oC te mperature coe fficie nt,” Electronics Letters, vol. 39, issue 2, pp. 209-210, Jan. 2003. [12] T. V. Cao, D. T. Wisland, T. S. Lande , F. Moradi and Y.

H. Kim, “Nove l start-up circuit with e nhance d powe rup characteristic for bandgap re fere nces,” in Proc. IEEE Int. SOC Conference, Se pt. 2008, pp. 123-126. [13] B. Razavi, Design of Analog CMOS Integrated Circuit,

McGraw Hill, 2001. [14] P. E. Alle n and D. R. Holbe rg, CMOS Analog Circuit

Design, 2nd Edition, Oxford University Press, 2002. [15] D. A. Johns and K. Martin, Analog Integrated Circuit

Design, Ne w York, John Wile y, 1997. [16] W. Wu, W. Zhiping and Z. Yongxue , “An improve d

CMOS bandgap refe re nce with se lf-biase d cascode d curre nt mirrors,” in Proc. IEEE Conf. on Electron Devices and Solid-State Circuits, Dec. 2007, pp. 945-948. [17] J. P. M. Brito, H. Klimach and S. Bampi, “A de sign

me thodology for matching improveme nt in bandgap re fe re nces,” in Proc. 8th Int. Symp. on Quality Electronic Design (ISQED’07), March 2007, pp. 586-594. [18] M. D. Ke r, J. S. Chen and C. Y. Chu, “A CMOS bandgap

re fe re nce circuit for sub-1-V ope ration without using e xtra

low-threshold-voltage

de vice ,”

IEICE Trans.

Electronic, vol. E88-C, no. 11, pp. 2150-2155, Nov. 2005.

254

Guo-Ming Sung rece ive d the B.S. and M.S. de grees in biome dical Engineering from the Chung-Yuan Unive rsity in 1987 and 1989, respective ly, and the PH.D. de gree in e lectrical e nginee ring from the National Taiwan Unive rsity, Taipe i, in 2001. In 1992, he joine d the Division of Engineering and Applie d Scie nces, National Scie nce Council, Taiwan, whe re he became an Associate Researcher in 1996. Since 2001, he has bee n with the Electrical Enginee ring De partment, National Taipe i Unive rsity of Technology, whe re he is an Associate Professor. His research interests include magne tic se nsors, integrate d circuits and syste ms for analog and digital circuits, motor control ICs, and mixe dmode ICs for XDSL. Ying-Tzu Lai rece ive d the M.S. degree in e lectrical e ngineering from Lunghwa Unive rsity of Scie nce and Technology, Taoyuan, Taiwan, R.O.C., in 2005, and now studying Ph.D. de grees in e lectrical e ngineering from National Taipe i Unive rsity of Technology since 2006. He r re search inte rests include mixe d-mode inte grate d circuit design, analog-todigital conve rters, and switche d-curre nt de lta-sigma modulator. Chien-L in Lu re ce ive d the B.S. degree from the De partme nt of Communications Engineering, Fe ng Chia Unive rsity in 2004, and now studying M.S. degrees in Ele ctronic Engineering from National Taipe i Unive rsity of Technology since 2008. He has bee n a me mber with the National Chip Imple mentation Ce nter (CIC), Taiwan, R.O.C. His research inte rests include analog circuit design, RF circuit design, and analog to digital conve rter (ADC).


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Transient Analysis of Three-Phase Self Excited Induction Generator Using New Approach Vivek Pahwa1, K. S. Sandhu2 Department of Electrical and Electronics Engineering, Haryana College of Technology and Management, Kaithal, India 1

Department of Electrical Engineering, National Institute of technology, Kurukshetra, Kurukshetra, India

2

pahwa1974@yahoo.com 1, kjssandhu@rediffmail.com 2 Abstract In this pape r, Matlab/Simulink base d ne w saturation mode l is propose d to investigate the transie nt pe rformance of a three -phase induction machine . The mode l as propose d is use d to pre dict the transie nt pe rformance of three -phase induction ge ne rator unde r diffe rent ope rating conditions. Closeness of simulate d results with e xpe rime ntal results on a test machine proves the effective ness of the propose d mode l. Keywords Modeling; Self Excited Induction Generator; Simulation; Transient Analysis

Introduction Global environmental concerns and growing demand of isolated power plants are some of the major issues due to which self excited induction generators are getting more and more popular. Further operational and constructional advantages are some other factors which are responsible for rapid establishment of suitable self excited induction generators in contrast to conventional synchronous generators. Further in the absence of grid a wind turbine generator in self excited mode is found to be very useful for isolated and remote locations. The self excited induction generator is essentially a three phase induction machine in which the magnetizing current is furnished by the static capacitors connected across the stator terminals [1, 2]. Whenever driven by a suitable prime-mover under favourable conditions, voltage build up occurs and power is transferred to connected load. The types of loads experienced on such isolated power plants consisting of self excited induction generators may be static/dynamic in nature. Sudden switching of such loads cause transients in the system, which are of

immense interest. Therefore transient analysis of a machine is must for design consideration and some researchers tried to investigate the dynamic performance of such generators [3-8]. [9, 10] used the well tested d-q axis based conventional model to investigate the transient behaviour of three phase induction machine. [11-14] describes the basic concept of transient modeling of the machine. Matlab / Simulink is found to be very useful tool for modeling electrical machine and it may be used to predict the dynamic behavior of the machines. In this paper Matlab / Simulink based new saturation model is proposed to study the dynamic behaviour of three phases self excited induction generator. Effects of ‘capacitor switching’, ‘load variation’, ‘input variation’ and ‘variation in moment of inertia’ on the transient performance of self excited induction generator have been taken in the present work. Mathematical Modeling The voltage equation of the induction machine model in rotor reference frame is given by [11-14];

[vG ] = [Z

] [i ]

(1)

[ ] represents

where [vG ] and i

matrices and are given as

i i i i   qs ds qr dr 

T

voltage and current

v v v v   qs ds qr dr 

T

and

respectively.

Impedance matrices as defined in equation (1) may be given as,

255


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

[Z ]

Im = Per phase magnetizing current (rms).

ω r Ls ω r Lm  Lm p  R s + Ls p  −ω L R s + Ls p − ω r Ls p Lm p  r s =  Lm p 0 R r + Lr p −0    0 0 Lm p R r + Lr p  

This curve for test machine [Appendix- A] may be used to develop the polynomial relationships between Xm and Im [Table-1] and is used to account the saturation in the simulated model. This is the conventional way to account the effect of saturation in electrical machines and is generally adopted by most of the researchers [2, 15-17] for the transient and steady state analysis of induction machine. 2) Proposed Saturation model

FIGURE 1 TWO-AXIS MODEL OF THREE-PHASE SELF EXCITED INDUCTION GENERATOR

And, the electromagnetic torque is

Te =

3P Lm (i 2 2

qs i dr

−i

ds i qr )

ψ = f (im) (2)

Equation of motion used to relate the electromagnetic torque developed and load torque may be defined as; (3)

Capacitor side equations are p[vG ] = (1 / c ) [ic ]

And

[ic ] = [i] + [i L ]

(4) (5)

[iL ] = [iLq iLd ]T [ic ] = [icq icd ]T

where

im = Instantaneous value of magnetizing current. Such relationships have been used by [13, 14, and 17] in case of transformer and reactors. However none of the research persons used such representations for the analysis of induction generators. Conventional magnetization curve as shown in figure A.2 may be modified as in figures B.1 [Appendix-B], which show the variation of instantaneous values of flux linkages and magnetizing current. This is used to develop the relationship between xm and im [Table-1]. For the first time, such relationship is proposed to account the saturation in three phase induction generator. TABLE1 RELATION BETWEEN MAGNETIZING REACTANCE AND MAGNETIZING CURRENT

Load side equation

[vG ] = LL p[i L ] + RL [i L ]

(6)

Saturation model Saturation curve of the machine may be represented [14] in two ways as discussed below.

Relationship between magnetizing reactance and magnetizing current

Due to conventional saturation c urve

This curve is a graphical representation between rms values of air gap voltage and magnetization current. Therefore mathematically air gap voltage is a function of magnetizing current and is represented as:

where E = Per phase air gap voltage (rms).

256

Xm = 0.3372 I3 m- 1.8650I2 m+ 9.1425 Im + 108.2482Ω

1) Conventional Saturation model

E = f ( Im )

(8 )

ψ = Instantaneous value of flux linkage.

Equation of motion

Te − TL = (2 J / P) pω r

This curve is a graphical representation between instantaneous values of air-gap flux linkage and magnetizing current. Therefore air-gap flux linkage is a function of magnetizing current and is represented as;

(7)

Due to proposed saturation c urve

xm = 0.1205 i3 m- 0.7154i2 m+ 8.6888 im + 100.4563Ω

Results And Discussions Figure 2 shows the MATLAB/SIMULINK based conventional and proposed models of a three-phase self excited induction generator used for simulations.


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Capacitor switching Out1

Va_proposed Ia_proposed

Out2

Vb_proposed Out3

Scope1

In1

Mux

Ib_proposed Out4

Tmech

y3i Vc_proposed

To Workspace1

Out5

Ic_proposed Out6

proposed

Mux1 Va_conventional

Out1

Ia_conventional

Out2

Vb_conventional

Out3

Figure 4 shows the effect of capacitor switching on the voltage build of induction generator under no load operation. The machine under consideration runs as a self excited induction generator with change of capacitance at 2, 3 and 4 seconds. At these instants capacitance is varied from 40 to 30, 30 to 20 and 20 to 10 microfarads respectively. From this figure following observations may be drawn:

Scope3

In1

Mux

Ib_conventional Out4

Proposed model results into high value of voltage in contrast to conventional model. In addition initial build up from zero to final value is dependent upon the type of modeling adopted for simulation.

y3r To Workspace2

Vc_conventional Out5

Ic_conventional Out6

Mux2

conventional

FIGURE 2 MATLAB/S IMULINK MODEL OF THREE-PHASE SELF EXCITED INDUCTION GENERATOR WITH PROPOSED AND CONVENTIONAL METHODOLOGY

300

Va(V)

Figure 3 shows the comparison of simulated and experimental results on test machine [Appendix-A]. Proposed saturation model yields better simulation results in contrast to conventional saturation model. This proves the effectiveness of proposed model in contrast to conventional model. Therefore it is recommended to use this model to predict the transient behavior of a self excited induction generator.

With decrease in capacitance the voltage decreased and ultimately it leads to voltage collapse, irrespective of the model used for simulation purpose.

200 100 0

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

300 simulated results with proposed model simulated results with conv. model

0

Ia(A)

4

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

2 0

400

Vb(V)

Vb(V)

200

Ib(A)

4

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

2

0 0 400

Vc(V)

100

0.5

1

1.5

2

2.5

3

3.5

4

4.5

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

0

0.5

1

1.5

2

2.5 time(sec)

3

3.5

4

4.5

5

300

simulated results with proposed model simulated results with conv. model experimental results

200 0

200

0

5

Vc(V)

Va(V)

400

200 100 0

200 0

Ic(A)

4

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

0

0.5

1

1.5

2

2.5 time(sec)

3

3.5

4

4.5

5

2 0

FIGURE 3 GENERATED VOLTAGES AND LOAD CURRENTS

Figure 4 to figure 7 show the comparison of simulated results with conventional and proposed modeling on test machine in self excited generating mode to analyze the effects of following: •

Capacitor switching

Load switching

Change in input power

Change in moment of inertia

FIGURE 4 EFFECT OF CAPACITANCE SWITCHING UNDER NO LOAD OPERATION, SPEED=1500 RPM

Load switching Figure 5 shows the effect of load switching on the transient behaviour of stator current of the test machine when load resistance is changed at 2 and 4 second. The load resistance is changed from 100 ohms to 200 ohms and 200 ohms to 400 ohms at the respective instants. Both models give the same transient response during the load switching. However initial response from zero value to final value of current found to be dependent upon the type of model used for simulation purpose.

257


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Change in moment of inertia

Ia(A)

3 2

It is well known that moment of inertia greatly affects the transient performance of three-phase induction machine in motoring mode. Figure 7.a to figure 7.c show the simulated results to look the effects of moment of inertia on the transient performance in self excited generating mode. It is observed that:

1 0

0

0.5

2

1.5

1

2.5

3

3.5

4

4.5

5

4.5

5

3

Ib(A)

simulated results with proposed model simulated results with conv. model 2 1 0

0

0.5

1

1.5

2

2.5

3

3.5

4

Any change in the moment of inertia of the machine affects the voltage build up of generator, irrespective of the type of model adopted for analysis. However this effect is more pronounced in case of proposed model in contrast to conventional model.

Ic(A)

3 2 1 0

0

0.5

1

1.5

2

2.5 time(sec)

3

3.5

4

4.5

5

FIGURE 5 EFFECT OF LOAD SWITCHING, C = 40 MICROFARADS, SPEED = 1500 RPM

Change in input power

Initial build up from zero to final value is dependent upon the type of m odeling adopted for simulation. Proposed model results into high value of voltage in contrast to conventional model.

0

0

0.1

0.2

0.3

0.4

0.5

0

0.1

0.2

0.3

0.4

0.5

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0 400

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0

0.1

0.2

0.3

0.4

0.5 time(sec)

0.6

0.7

0.8

0.9

1

Ia(A)

4

0

Vb(V)

0

Vc(V)

Ib(A)

4

0 4

0

200 0.1

0.2

0.3

0.4

0.5

2.5

3

3.5

4

0

0.1

0.2

0.3

0.4

0

0.1

0.2

0.3

0 0 400

0.1

0.2

0

0.1

0

0.1

0

0.5

1

1.5

2

2.5

3

3.5

4

simulated results with proposed model simulated results with conv. model

0

1

1.5

2

2.5

3

3.5

4

0 0 400

0.5

1

1.5

2

2.5

3

3.5

4

2

200

Vc(V)

Ib(A)

Ib(A)

0.5

Ic(A)

4

0

0.5

1

1.5

2

2.5

3

3.5

0

4

0.5

1

1.5

2 time(sec)

2.5

3

3.5

4

FIGURE 6EFFECT OF CHANGE IN INPUT POWER, C = 40 MICROFARADS, SPEED = 1500 RPM

1

simulated results with proposed0.8 model 0.5 0.6 0.7 simulated results with conv. model

0.9

1

0.4

0.5

0.6

0.7

0.8

0.9

1

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0.2

0.3

0.4

0.5 time(sec)

0.6

0.7

0.8

0.9

1

2 0

0

0.9

200

4

2

0.8

2

0 0

0.7

200

4 0

4

Vb(V)

0

0.6

2

400

200 0

258

0

Ia(A)

2

1

400

Ic(A)

Va(V)

Ia(A)

400 Vb(V)

1.5

simulated results model 0.6 0.7 with proposed 0.8 0.9 simulated results with conv. model

FIGURE 7 a EFFECT OF CHANGE IN MOMENT OF INERTIA, C = 40 MICROFARADS, SPEED = 1500 RPM WITH J = 0.913 KGM2

2 0

Vc(V)

1

1

2

0

0.5

0.9

2

4 0

0.8

200

200

4

0.7

200

400

0

0.6

2

400

Ic(A)

With decrease in input power the voltage as well as the current is decreasing simultaneously.

200

Va(V)

Figure 6 shows the effect of change of input mechanical power applied by the prime mover on the voltage build of induction generator. The machine under consideration runs as a self excited induction generator with change of input mechanical power at 1 and 3 seconds. At these instants input mechanical power is varied from 1 pu to 0.5 pu and 0.5 pu to 0.25 pu respectively. From this figure following observations may be drawn:

Va(V)

400

FIGURE 7 b EFFECT OF CHANGE IN MOMENT OF INERTIA, C = 40 MICROFARADS, SPEED = 1500 RPM WITH J = 0.95 KGM 2


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

200 0

0

0.1

0.2

0.3

0.4

0.5

0

0.1

0.2

0.3

0.4

0.5

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0 0 400

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

0

0.1

0.2

0.3

0.4

0.5 time(sec)

0.6

0.7

0.8

0.9

1

Ia(A)

4

0

Vb(V)

0.7

0.8

0.9

1

simulated model 0.6 results 0.7with proposed 0.8 0.9 simulated results with conv. model

1

2

400 200 0

Ib(A)

4

Vc(V)

0.6

2

0

Ic(A)

2 0

Appendix-A 3-hp, 3-phase, 50 Hz, 220 volts Induction Motor; Stator Resistance, Rs= 3.35 ohms

200

4

From above observations it may be concluded that the proposed model, which is found to be superior to conventional model, results into different but reliable simulations. Therefore, it is strongly recommended to use this model for investigating the transient performance of self excited induction generator.

FIGURE-7 c EFFECT OF CHANGE IN MOMENT OF INERTIA, C = 40 MICROFARADS, SPEED = 1500 RPM WITH J = 1.0 KGM2

Conclusion Due to global acceptability of self excited induction generators in wind power conversion, in this paper an attempt is made to analyze the transient behaviour of such machines under capacitor and load switching. In addition simulated results were also taken to include the effects of ‘input mechanical power’ and ‘moment of inertia’ on the performance of the machine. Matlab/Simulink based new model is proposed to investigate the transient performance of a self excited induction generator. Simulated results as obtained with new proposed model are found to be closer to experimental results. This proves the effectiveness and superiority of proposed model in contrast to conventional model. Simulated results as shown in figures 4 to figure 7 may be used to draw the following observations.

Rotor Resistance, Rr = 1.7 ohms Stator & Rotor Inductance,Ls =Lr = 15.44 mH Moment of Inertia of test machine set up With coupling,J = 0.913 kgm 2 Two machines as shown in figure A.1 must run in the same direction in case fed individually. After that test machine is driven at synchronous speed with prime mover. Input current, power is recorded for different values of input voltage. Data as obtained is used to draw the magnetization curve of test machine as shown in figure A.2.

FIGURE A 1 SET UP FOR S YNCHRONOUS RUN TEST

Proposed model results into a delayed voltage build up in case of self excited mode for any given value of excitation capacitance, load, mechanical input and moment of inertia. Delay in voltage build-up further increases with an increase in moment of inertia i.e. especially for large rated machines. Nature of effects of ‘capacitor switching’, ‘load variation’, ‘input variation’ and ‘variation in moment of inertia’ is found to be same, irrespective of model used for simulation purpose. However simulated results (using proposed model) for such effects are found to be slightly different than those with conventional model.

250 EN EN-1 200

En 150 En-1 E(V)

Va(V)

400

E2 100

E1 50

Im1 0

0

0.5

Im2 1

Imn-1

Imn

1.5 Im(A)

ImN-1 2

ImN 2.5

3

FIGURE A 2 CONVENTIONAL MAGNETIZATION CURVE

259


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Appendix-B

ωs

Conventional magnetization curve as shown in figure A.2 may be converted to the proposed saturation curve with the procedure laid down by [13, 17] and it is shown in figure B.1:

rotating reference frame

350 psiN psiN-1 300

250

=Angular speed (radian/sec.) in synchronously

J

= Inertia of Motor

Te

= Electrical Torque

TL

= Load Torque

p

= Operator for differentiation

Subscripts:

psin

psi

200 psin-1

q

= Quadrature axis

d

= Direct axis

s

= Stator quantities

r

= Rotor quantities

psi2 150

100 psi1 50

im1 0

im2 1

0

imN-1

imn-1imn 2

3 im(A)

4

imN 5

6

REFERENCES [1]

FIGURE B 1 PROPOSED SATURATION CURVE.

For a sinusoidal input voltage of frequency, ω, the corresponding flux linkage is given by

[2]

im ,

k = 1,-----n-1, n,-------, N

[3]

= Mutual inductance/phase

Rr

= Rotor Phase Resistance/phase

Lr

260

= Rotor self inductance/phase

R.

C.

Prasad and B.

M.

Karan,

Mode ling and

Se lf-Excite d

Induction

Ge nerator

supplying

Dynamic Load”, Ele ctric Machine s and Powe r Systems, vol. 27, pp. 941-954, 1999. [4]

L. Wang and R.Y. De ng, “Transie nt Pe rformance of an Isolate d

induction

Ge nerator

under

Unbalance d

Excitation Capacitors”, IEEE Transactions on Ene rgy Conversion, vol. 14, no. 4, pp. 887-893, De cember 1999. [5]

S.K. Jain, J.D. Sharma and S.P. Singh, “Transie nt Pe rformance of Three -phase Se lf-e xcite d Induction

= Stator Phase Resistance/phase

Lm

Kishore ,

B. Singh, L. Shridhar and C.S. Jha, “Transie nt Analysis of

Gj is the slope of line joining points (j-1) and j, as seen from vertical axis [13, 17].

= Stator Self inductance/phase

A.

March 26-29, 2006.

j =1

Ls

of

Research Symposium, Cambridge , USA, pp. 312-316,

imk = ∑ G j (Ψ j − Ψ j −1 )

Rs

and Design

Induction Ge ne rator”, Progress in Ele ctromagnetics

k

Nomenclature

Pe rformance

Dynamic Characte ristics of Three Phase Se lf Excite d

Ψ 0, Ψ 1, -----------, Ψ N can be obtained.

For calculation of proposed value of current

“The

“MATLAB/SIMULINK base d D-Q

k = 0, 1,-----n-1, n,-------, N

2.

Say,

Distributors, Third e dition, 2002.

Ψk = 2 E k / ω So,

G.

Alte rnating Curre nt Machines”, CBS Publishe rs and

Conversion guidelines: 1.

M.

Ge nerator during Balance d and Unbalance d Faults”, Proc. Inst. Ele ct. Eng., Ge n., Transm., Distrib., vol. 149, pp. 50-57, January 2002. [6]

Y. S. Wang and L. Wang, “Unbalance d Switche d Effects on Dynamic Pe rformance of an Isolate d Three phase Se lf-e xcite d Ge ne rator”, Electric Machines Powe r Systems, vol. 29, no. 4, pp. 375-387, April 2001.


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

[7]

D. Se youm, C. Grantham, and M.F. Rahman, “The Dynamic Characte ristics of an Isolate d Induction Ge nerator

Drive n

by

a

Wind

Turbine ”,

IEEE

Transactions on Industry Applications, vol. 39, no. 4, [8]

Ong,

“Dynamic

Simulation

of

Ele ctric

Machine ry”, Prantice Hall PTR, 1998. [14] S. N. Talukdar, J. K. Dickson, R. C. Dugan, M. J.

pp. 936-944, July/August 2003.

Sprinze n and C. J. Le nda, “On Mode ling Transformer

F. Khate r, R.D. Lore nz and D.W. Novotny, “Se lection

and Reactor Saturation Characteristics for Digital and

of Flux Le ve l in Fie ld-Orie nte d Induction Machine

Analog

Controlle rs with Conside ration of Magne tic Saturation

Apparatus and Systems, vol. PAS-94, no. 2, pp. 612-622,

Effects”, IEEE Transactions on Industry Applications,

1975.

vol. 23, no. 2, pp. 276-282, March/April 1984. [9]

Analysis and Control”, Pearson Prentice Hall, 2007. [13] C.M.

studie s”,

IEEE

Transactions

on

Powe r

[15] Nuh Erdogan, Humberto He nao and Richard Grise l,

Julio C. More ira and Thomas A. Lipo, “Mode ling of

“The Analysis of Saturation Effects on Transie nt

Saturate d AC Machines including Air Gap Flux

Be havior of Induction Machine Direct Starting”, IEEE,

Harmonic

pp. 975-979, 2004.

Compone nts”,

IEEE

Transactions

on

Industry Applications, vol. 28, no. 2, pp. 343-349, March/April 1992. [10] Paul C. Krause, O. Wasynczuk and S. D. Sudhoff,

“Analysis of Ele ctric Machinery and Drive Systems”, IEEE Pre ss Series on Power Enginee ring, John Wiley & Sons Inc. Publication, 2004. [11] B. K. Bose , “Powe r Electronics and AC Drives”,

Pearson Pre ntice Hall, 2007. [12] R. Kr ishna n, “Ele c tr ic Mo to r Dr ive s. Mo de ling ,

[16] O. I. Okoro, “MATLAB Simulation of Induction

Machine with Saturable Leakage and Magne tizing Inductances”, The Pacific Journal of Scie nce and Technology, vo1. 5, no. 1, pp. 5-15, April 2003. [17] S. Prusty and M.V.S. Rao, “A Direct Piece wise

Line arize d Approach to Convert rms Saturation Characte ristics to Instantaneous Saturation Curve ”, IEEE Transactions on Magnetics, vol. 16, no. 1, pp.156160, January 1980.

261


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

On the Sensitivity of Principal Components Analysis Applied in Wound Rotor Induction Machines Faults Detection and Localization J. Ramahaleomiarantsoa1, N. Heraud2 , E. J. R. Sambatra3 , J. M. Razafimahenina4 Université de Corse, U.M.R. CNRS 6134 SPE, BP 52, 20250 Corte, France

1, 2

Institut Supérieur de Technologie, BP 509, 201 Antsiranana, Madagascar

1, 3

Ecole Supérieure Polytechnique, BP O, 201 Antsiranana, Madagascar

1, 4

ramahaleojacques@yahoo.fr; 2heraud@univ-corse.fr ; 3ericsambatra@yahoo.fr; 4razafimaheninajeanmarie@yahoo.fr

1

Abstract This pape r deals with faults de tection and localization of wound rotor induction machines base d on principal compone nts analysis me thod. Both, the localization and the de tection approaches consist in analyzing dete ction inde x which is establishe d on the latest principal components. Once the faults are de tecte d, the affecte d state variable s are localize d by the variables reconstruction approach. The e xponentially we ighte d moving average filte r is applie d to improve the faults de tection quality by re ducing the rate of false alarms. An accurate analytical mode ling of the e lectrical machines is propose d and impleme nte d on the Matlab software to obtain the state variables data of both healthy and faulte d machines. Se veral simulation results are prese nte d a nd analyze d. Keywords Principal Components Analysis; Wound Rotor Induction Machines; Faults Detection and Localization; Detection Index; Reconstruction Approach; EWMA filter

Introduction The necessity for having reliable electric machines is more important than ever and the trend continues to increase. Now, advances in engineering and materials science allow building lighter machines while having a considerable lifetime. Although researches and improvements have been carried out, these machines still remain the most potential failures of the stator and the rotor. The faults can be resulted by normal wear, poor design, poor assembly (misalignment), improper use or combination of these different causes. Indeed, for many years, faults detection in electrical machines has been the subject of reflection and research projects in various industrial and academic laboratories.

262

Several diagnosis and control methods exist and already used for the electrical machines monitoring. In this paper, Wound Rotor Induction Machines (WRIM) faults detection and localization based on Principal Components Analysis (PCA) is proposed. PCA is a statistical method used for data or state variables measurement of systems in operation to monitor their behavior. The PCA principle consists in reducing the size of the representation space of the system [1]. In fault detection approach based on PCA, two methods are proposed [2, 3], Hotelling’s T2 statistical method and Squared Prediction Error (SPE) indicator. The T2 statistical is calculated with the “l” first principal components while the SPE indicator achieves detection with the residual space. However the two methods have limitations in faults detection [3, 4]. In case of sensors detection, the T2 indicator is not very efficient because the variations due to the failure may be masked by normal variations of the variables in the first principal components space. And when the considered systems are no longer linear, residues having high variance contain the modeling errors generated by the PCA. Thus the residue having a low variance will have less influence on the SPE quantity with respect to the residues having a higher variance, so that they correspond to the linear redundancy relations or quasi-linear. This sensitivity of the SPE indicator to the modeling errors can create many false alarms. F. Harkat, G. Mourot, and J. Ragot [2] proposed a new method for faults detection and localization based on the sums of squares of the last principal components.


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

For our case, the method proposed by [2, 4] will be used for the WRIM faults detection and localization. To improve the faults detection and to reduce the false alarms, the Exponentially Weighted Moving Average (EWMA) filter is used. The first part of this article deals with the reconstruction principle of the PCA model followed by the WRIM modeling. The s econd part is focused on the faults detection method by the detection index (D i). The third part is focused on variables reconstruction combined with the fault indicator Di for fault location aim. The last part is reserved to the applications of the PCA approach on the WRIM. Several simulations result built with Matlab software are presented and analyzed to show the PCA method sensitivity.

PCA decomposes X as follows: (3)

T = XP

T ∈ ℜ N *m and N is the number of carried out measures of variables to be monitored. Determination of the structure of the model To obtain the model structure, the components number “l” to be retained must be determined. This step is very important for PCA construction. Component number can be determined by using:

 l  ∑ λi  i =1  m  ∑ λk  k =1

   *100 ≥ thc   

(4)

Pca Method Imple mentation

With l<m

PCA methods

Where thc is an user defined threshold expressed as percentage. Now, user should retain only the components number “l” which was associated in the first term of (4). By reordering the eigenvalues, the minimum numbers of components are retained while still reaching the minimum variance threshold, [9, 10]. The vector of principal components is noted by:

The PCA method is based on a transformation of space representation of simulation data. The new space is smaller than that of the original space. This method is classified as without model methods [5] and can be seen as a full-fledged system identification method [6, 7]. Each variable to be monitored for the state of the WRIM are expressed by different units and scales. For that, it is preferable to apply a PCA on a centered and reduced measures matrix X (columns of zero means and units standard deviations) [8]. The orthogonal space defined by PCA is generated by the eigenvalues and eigenvectors of the matrix correlation R of X. These values are sorted in descending order in a diagonal matrix. The eigenvalues analysis of the correlation matrix R provides information on the number of principal components to be retained “l” for the PCA model reconstruction [1]. The orthonormal projection matrix P formed by the m eigenvectors associated with eigenvalues of the correlation matrix R is expressed as:

P = [ p1 , p2 ,..., pm ]

(2)

With λ1 ≥ λ2 ≥ ... ≥ λm The orthogonal matrix which represents the projection of X in the PCA new space is T. Mathematically, the

(5)

Since the aim of PCA is to reduce the space dimension, the “l” first principal components (l << m) are the most significant and sufficient to explain the variability of a process. Therefore, the expression of centered and reduced matrix X can be written as follows:

= X Xp +E

(6)

The matrix X p is the estimated principal part and the matrix E the residual part of X which represents information looses due to the X matrix dimension reduction. They are expressed as follow: l

' X p = ∑ PT i l

(7)

i =1

(1)

The diagonal matrix Λ of the correlation matrix R generated by the eigenvectors associated with eigenvalues λ sorted in descending order is done by:

Λ =diag (λ1 , λ2 ,..., λm )

T = [t1 , t2 ,..., tm ]

E=

m

∑ PT

i = l +1

'

i l

(8)

T ' is the transpose of the orthogonal matrix. Wrim Analytical Modeling Fig.1 shows the equivalent electrical circuit of the WRIM. Each coil, for both stator and rotor, is modelised with a resistance and an inductance connected in series configuration (Fig. 2).

263


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

V j, Ij and Φj (j : A, B, C for the stator phases and a, b, c, for the rotor phases) are respectively the voltages, the electrical currents and the magnetic flux of the stator and the rotor phases, θ is the angular position of the rotor relative to the stator.

[RS] and [RR ] are the resistances matrix, [LS] and [LR] the own inductances matrix, and [MSR ] and [M RS ] the mutual inductances matrix between the stator and the rotor coils. Equations (9) and (10) become: d {[ LS ][ I S ]}

[VS ] = [ RS ] [ I S ] +

dt

[VR ] = [ RR ] [ I R ] +

d {[ LR ][ I R ]} dt

+

+

d {[ M SR ][ I R ]}

(13)

dt d {[ M RS ][ I S ]}

(14)

dt

By applying the fundamental principle of dynamics to the rotor, the mechanical motion equation is [12]: Jt

dΩ + fv= Ω Cem − Cr dt

FIG. 1 EQUIVALENT ELECTRICAL CIRCUIT OF THE WRIM

Ω=

dθ dt

Cem =

FIG. 2 EQUIVALENT ELECTRICAL CIRCUIT OF THE WRIM COILS

Rj and Lj are the resistances and the own inductances of the stator and the rotor phases. We note the voltages vector ([V S], [V R]), the currents vector ([IS], [IR]) and the flux vector ([ΦS], [ΦR]) of respectively the stator and the rotor:

(9)

2

 LSC  0   0 [ L] =   M SR f1  M SR f 2   M SR f 3

0 LSC 0 M SR f 3

0 0 LSC M SR f 2

M SR f1 M SR f 3 M SR f 2 LRC

M SR f 2 M SR f1 M SR f 3 0

M SR f1 M SR f 2

M SR f 3 M SR f1

0 0

LRC 0

M SR f 3  M SR f 2  M SR f1   0  0   LRC 

(18)

With (10)

= [VR ]

[ RR ] [ I R ] +

= [φS ]

[ LS ] [ I S ] + [ M SR ] [ I R ]

(11)

= [φR ]

[ LR ] [ I R ] + [ M RS ] [ I S ]

(12)

264

Jt is the total inertia brought to the rotor shaft, Ω the shaft rotational speed, [I]=[IA IB IC Ia Ib Ic] ’ the current vector, fv the viscous friction torque, Cem the electromagnetic torque, Cr the load torque applied to the machine, θ the angular position of the rotor with respect to the stator, and [L] the inductance matrix of the machine.

2

dt d [φR ]

(17)

of the each phase of the stator and LR is the own inductance of the each phase of the rotor), the mutual inductances between the stator and the rotor coils MSR and pole pair number p, the inductance matrix of the WRIM car be written as follow:

Taking into account the above assumptions, both stator and rotor three phase voltages and currents are connected to the total magnetic flux by differential equations systems [11]. The stator and rotor voltages vectors expressions are given by: d [φS ]

1 t d ( [ L] ) *[ I ] [I ] * dθ 2

rotor LSC = 3 LS and LRC = 3 LR (LS is the own inductance

Va  Ia  φa      ; ; [VR ] = Vb  [ I R ] =  I b  [φR ] = φb  Vc   I c  φc 

[ RS ] [ I S ] +

(16)

Introducing the cyclic inductances of the stator and the

VA  IA  φ A  [VS ] = VB  ; [ I S ] =  I B  ; [φS ] = φB  VC   I C  φC 

= [VS ]

(15)

dt

(19)

f1 = cos( pθ ) f 2 = cos( pθ +

2π ) 3

(20)


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

f 3 = cos( pθ −

2π ) 3

(21)

Differential equations system modelling In choosing the stator and rotor currents, the shaft rotational speed and the angular position of the rotor relative to the stator as state variables, the differential equations a system modeling the WRIM is given by: (22)

= [ X ] [ A]−1 ([U ] − [ B][ X ])

With [ L] 0 [ X ] [ I A I B I C I a I b I c Ω θ ]' ; [ A] =  0 J t   0 0

0 0  ; 1 

codes and allows us to obtain several matrix data for the PCA applications on WRIM faults detection and localization. The WRIM is considered faulted from t=2s and coupled to a mechanical load at time 2s. Nine state variables (m=9) have been chosen to be monitored and 10000 measures (N=10000) during 4s are considered. Fig.3 represents the temporal variations of some state variables (Stator current, rotor current, shaft rotational speed, angular position and electromagnetic torque) showing the steady and transient states of faulted WRIM. Fig.4 shows the zoom of the same state variables variations but only the part during which the machine is in loaded.

 [V ]  ' [U ] =  −Cr  ; [V ] = [VA VB VC Va Vb Vc ] ;  0  d [ L]   0 0 [ R] + Ω dθ   d [ L] 1 fv 0 [ B ]=  − [ I ]t  2  dθ   0 1 0 −    

This model of the WRIM will be used to simulate both healthy and faulted operation case of the stator and the rotor. The considered faults are resistances values increases of the stator or rotor windings due to a rise of their temperatures. The following table presents the different parameters of the WRIM:

FIG. 3 STATE VARIABLES VARIATIONS VERS US TIME OF THE FAULTED WRIM LOADED (STEADY AND TRANS IENT STATES)

TABLE I WRIM PARAMETERS Symbol

Parameter

Value

Units

Lsp

S tator princ ipal induc tanc e

0.397

H

Lrp

Rotor princ ipal induc tance

0.397

H

Lsl

S tator leakage induc tance

9.594

mH

Lrl

Rotor leakage induc tanc e

9.594

mH

M sr

S tator-rotor mutual induc tanc e

0,3953

H

p

Number of pole pairs

1

-

Jt

Moment of inertia

0.024

Kg.m2

Rs

S tator resistance

2.86

Rr

Rotor resistanc e

2.756

fv

Viscous fric tion coeffic ient

1,444

mNm/rad/s

The model has been implemented on Matlab in source

FIG. 4 STATE VARIABLES VARIATIONS VERS US TIME OF THE FAULTED WRIM (STEADY S TATE)

Considered faults The considered faults are on the resistance values which increase due to a rise of their temperature. In normal operation, a resistance value variation

265


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

compared to its nominal value (in ambient temperature, 25°C) is faulted machine due to machine overload or coils fault [10, 13]. The resistance versus the temperature is expressed as:

R = R0 (1 + α∆T )

(23)

R0 is the resistance value at T0 = 25°C, α the temperature coefficient of the resistance and ΔT the temperature variation. Faults Detection Approach Residues generation For any measures vector x(k) the equation (6) becom es:

= x ( k ) x p ( k ) + e( k )

(24)

estimations vector and the estimation errors vector. The principal components vector t(k) corresponds to x(k) is expressed as:

t(k) = [tp(k) te (k)]

(26)

t p ∈ ℜ N *l and te ∈ ℜ N *( m −l ) are respectively the “ l ” first

principal components vector and the “ m-l ” last principal components vector. With this expression (26), there is an similarity on the residue vector e(k) and the final components vector te(k). Detection index “Di” calculation The fault detection index is based on successive sums of squares of the last principal components [2, 4] and is defined as follows: m

∑t

(k )

(28)

Di (k ) > τ i2,α EWMA filter

To reduce false alarm and to improve the faults detection quality, the EWMA filter is applied at time k, and then the “jnth” filtered vector of the last principal components can be written as follow [3, 6 and 15]:

tefj (k ) = (1 − γ )tej (k − 1) + γ tej (k )

γ =1 − exp(−1/ ∆ t )

(27)

Finally, the detection index is expressed as:

Di f (k ) =

m

∑t

2 efj j = m −i +1

(k )

(31)

And the filtered detection threshold of faults is given by [3, 10]:

τ if2 ,α =

γ 2−γ

τ i2,α

(32)

It should be noted that many research works uses the threshold detection for sensor faults, but our case concerns faults detection of systems. Faults Localisation Approach When a fault is detected, it is necessary to localize or identify the involved variables. There are several methods for faults localization: residues structuring approach,

At time k, systems are malfunctioning sensing if Di is greater than a threshold index noted τ i2,α . α is the false

partial PCA approach,

detection probability according to the Khi-2 law with “m” degree of freedom [14]. One can note a strong similarity between the detection index SPE and the detection index D i. Indeed, D i corresponds to the SPE indicator calculated by PCA model with (m-l)

(30)

∆ t is the time step.

i = 1, 2,…, (m-l)

266

(29)

γ is the forgetting factor (0< γ <1) in taking as initial

(25)

t (k ) = P x(k ) '

Di (k ) =

The process is considered in default at time k if:

condition tej (0) = 0 and can be calculated by [6, 16]:

x p (k ) and e(k ) vectors represent respectively the

2 ej j =m−i +1

principal components. Thus, this threshold detection can be calculated with an argument similar to that exposed in [4, 14].

Calculation of variables contributions to the detection indicator approaches. But [45] showed the disadvantage of the methods mentioned above. Then, in this paper, faults location of WRIM state variables is based on the variables


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

reconstruction combined with the filtered detection index. The localization of the WRIM affected state variables by the combined approach is based on two methods combination: variables reconstruction by PCA, detection index The method consists in eliminating the fault influence on Di when the affected variable is reconstructed. To localize faults on the indicator, faults directions projected in the residual space should not be collinear [8]. After the j variable number reconstruction, fault indicator in noted Di j . One can also use the EWMA filter to reduce the localization false alarms and to improve the localization indicator quality. If Difj is the filtered detection index of the j variable number, the localization indicator can be obtained by:

Lifj =

Difj

τ if2 ,α

(33)

The variable for which the localization indicator Lifj is less than one is the offending variable. This method can be used for the multiple faults localization in reconstructing the supposed faulted variables simultaneously. Simu lation Results and Discussion To validate the proposed models and the efficiency of the chosen approaches, the established models have been implemented in Matlab. Nine WRIM state variables (stator three phase currents, rotor three phase currents, shaft rotational speed, angular position and electromagnetic torque) have been considered. The matrixes data of the healthy and faulted WRIM obtained by the analytical model of the machine are introduced in PCA model to show faults detection and localization performances. For the electrical machines diagnosis, many methods are used to detect the presence or absence of faults, occurred at t=2s, and to locate the time when it began to appear on the machine windings. Two types of faults levels are considered in the system (10%, 30%). These values correspond to the rise of the stator or rotor coils resistance. We can mention the temporal

representation (Fig. 5, Fig. 6 and Fig. 7) and the signal frequency analysis. Although they have demonstrated their efficiency, the state variables representations between them also show their advantages. They can be performed without mathematical transformation (Fig. 7). Also, the electromagnetic torque variations versus the shaft rotational speed clearly show the WRIM operation zone in the presence of faults (Fig. 7 ). After several simulations, we suggest some of these methods to highlight the place of PCA among them. In the Fig.5 and Fig.6, the figures clearly show that it is difficult to visualize changes in signals and the fault appearance time. However, by analyzing the residues of the stator current by PCA (Fig.8, Fig.9), the fault appearance time is located on the two signals. The case of a healthy machine that has a zero residue is almost coincident with the x-axis. These observations are found in the case of the rotor current (Fig. 8). In Fig.5 and Fig.6, the presences of faults with the conventional temporal representation are no more evident than that using PCA method (Fig. 8 and Fig. 9). This one shows the residue analysis interest on PCA method. Fig.6 and Fig.9 expose the real and residue variations of the electromagnetic torque versus time. With Fig.6, the fault appearance time is not easy to locate. However, with PCA method, the variation of residues in the electromagnetic torque with and without faults can be easily detected. The real (Fig. 6) and residue (Fig. 9) variations of the electromagnetic torque versus shaft rotational speed of the machine show again that it is much more interesting to treat the state variables of the machine with PCA m ethod to detect the presence or absence of faults on the windings. The difference between healthy and faulted operation (Fig. 9) are clearer. It is almost not found in the real variation representations (Fig. 6). Fig.8 to Fig.11 highlight the major potential benefits of state variables treatment by PCA method which easily shows faults detection and locate time of fault appearance. With PCA method application, all representation (Fig. 10 and Fig. 11) well shows the differences between healthy and faulted WRIM. In the healthy case, residues are zero. When faults appear, the residue representations have an effective value with an absolute value greater than zero. In the Fig.10 and Fig.11, the healthy case is represented by a right line placed on the x-axis. Also, in taking into account “1” last principal component, Fig.10 shows peak variations

267


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

With “2” last principal components, the detection is improved because after the first peaks, other peaks (in the presence of faults) are greater than those in the case of the healthy machine. In the case of the faults detection, Fig.12 and Fig.13 show the variations of the filtered and the no filtered detection index of both “1” and “2” last principal components versus the measure number of the faulted WRIM. The threshold detection is represented on both figures. In the part where the machine is coupled to a mechanical load, the shapes exceed repeatedly the threshold index. This overrun corresponds to the presence of faults. The last principal components numbers do not have large influences on the curves shape for both filtered and no filtered detection index. 10 Healthy 30 % 10 %

8

Stator current IA [A]

6

Healthy 10.5

10.45

10.4

10.35

30% 10.3

4

287.5

10% 289

288.5

288

290

289.5

Shaft rotationnal speed [rad/s]

FIG. 7 REAL VARIATIONS OF ELECTROMAGNETIC TORQUE VERS US THE S HAFT ROTATIONAL SPEED OF THE WRIM Residues:Healthy & Faulted WRIM 1

Healthy

10%

30%

0.5

Phase "a" Rotor current

WRIM Healthy & Faulted: Stator current [A]

important in faults detection process to avoid alarm false.

Electromagnetic torque [Nm]

at t=2s (measure number 5000), time at which the faults are introduced. The peaks are attenuated immediately after but the signals are shifted.

0

-0.5

2

-1 0

Early faulted

-2

-1.5 1.99

-4

1.992 1.994 1.996 1.998

2

2.002 2.004 2.006 2.008

Time [s] -6 -8 -10 1.995 1.996 1.997 1.998 1.999

2

2.001 2.002 2.003 2.004 2.005

Time[s]

FIG.8 EARLY FAULTED IN VARIATIONS OF THE ROTOR CURRENT RES IDUES VERS US OF THE HEALTHY AND FAULTED WRIM

FIG.5 REAL VARIATIONS VERS US TIME OF THE STATOR CURRENT OF THE HEALTHY AND FAULTED WRIM

Residues:Healthy & Faulted WRIM 0.25 0.2

30%

Healthy & Faulted WRIM

0.15

40

Electromagnetic torque [Nm]

30

Electromagnetic torque

Healthy 10 % 30 %

35

25 20

Healthy

0 -0.05

15

-0.1

10 5

-0.2 -0.03

0

-10 -50

10%

-0.15

-0.02

-0.01

0

0.01

0.02

0.03

0.04

Shaft rotational speed

-5

0

50

200 100 150 Shaft rotational speed [rad/s]

250

300

350

FIG.6 REAL VARIATIONS OF ELECTROMAGNETIC TORQUE VERS US THE S HAFT ROTATIONAL SPEED OF THE WRIM

However, in the case of the no filtered shape, excessive values appear. These values show a bad detection of the no filtered data compared to those of the filtered data. This behaviour can be corresponding to alarm false for some cases. Data filtering is therefore

268

0.1 0.05

FIG. 9 VARIATIONS OF ELECTROMAGNETIC TORQUE RES IDUES VERS US THE S HAFT ROTATIONAL S PEED RES IDUES OF THE WRIM

Fig.14 and Fig.15 represent respectively the no filtered and the filtered localization index versus the WRIM state variables. The threshold of the localization index is represented on both figures. All variables having a localization index founding below the threshold variation are the affected variables.


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Detection index of 1 last principal component

-3

2.5

x 10

Dectetion index of 2 last principal components

-12

x 10

1.5

No filtered Di Filtered Di Thershold index

2

False alarm

1.5

1

30% 10%

1 Thershold index

0.5

0.5

0

Healthy 2000

3000

4000

5000

6000

7000

8000 0

Measure number

FIG. 10 DETECTION INDEX OF “1” LAST PRINCIPAL COMPONENT VARIATIONS VERS US THE MEAS URE NUMBER OF THE WRIM -3

x 10

Non filtered localization index

30%

4 3 2

10%

1 0

8000

9000

10000

4200

4400

4600

4800

1

0.8

0.6

0.4

0.2

Healthy 5000

5200

5400

5600

5800

6000

0

Measure number

FIG. 11 DETECTION INDEX OF “2” LAST PRINCIPAL COMPONENTS VARIATIONS VERS US THE MEAS URE NUMBER OF THE WRIM

1

2

3

4

5

6

7

8

9

WRIM state variable

FIG. 14 NO FILTERED LOCALIZATION INDEX VARIATION VERS US THE STATE VARIABLES

Detection index of 1 last principal component No filtered Di Filtered Di Thershold index

State variable Threshold localization index

Filtered localization index

1.2

1

False alarm

0.8

0.6

Thershold index 0.4

0.2

0

7000

State variable Threshold localization index

1.2

5

1.2

6000

1.4

6

-12

5000

Detection index of 2 last principal components

7

x 10

4000

FIG. 13 FILTERED AND NO FILTERED DETECTION INDEX OF “2” LAS T PRINCIPAL COMPONENT VARIATIONS VERS US MEAS URE NUMBER OF FAULTED WRIM

8

1.4

3000

2000

1000

Measure number

9

-1 4000

0

1

0.8

0.6

0.4

0.2

0

1000

2000

3000

4000

5000

6000

7000

8000

9000

10000

0

1

Measure number

FIG. 12 FILTERED AND NO FILTERED DETECTION INDEX OF “1” LAS T PRINCIPAL COMPONENT VARIATIONS VERS US MEAS URE NUMBER OF FAULTED WRIM

In the case of the no filtered localization index, only variable “4”and “6”correponding to the phase “a” and phase “c” rotor currents are not affected. In the filtered localization index case, stator phase “a” and phase “b” are not affected by faults. This last better reflects the WRIM behaviour in the case of the considered fault type. As in the case of the fault detection approach, the use of filter is necessary for faults localization.

2

3

4

5

6

7

8

9

WRIM state variable

FIG. 15 FILTERED LOCALIZATION INDEX VARIATION VERS US THE STATE VARIABLES

Conclusion PCA method based on residues analysis has been established and applied on WRIM diagnosis. In the case of temporal variation and without PCA, the electromagnetic torque and the shaft rotational speed are the more affected by the considered fault type. An accurate analytical model of the machine has been proposed and simulated to perform the healthy and

269


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

faulted data for PCA approach need.

for

WRIM faults detection and localization approaches based on PCA method are proposed. For that, an accurate analytical modeling of the WRIM has been carried out. The established models are implemented in Matlab. Nine state variables of the machine have been considered. Simulation results show the efficiency of the detection and localization based on respectively the detection index and localization index. The use of EWMA filter on both detection and localization has helped to avoid some false alarm. Also, filtered localization index better show the affected variables.

Engineering Practice, vol.17, pp. 494-502, 2009.

This research was supported by MADES/SCAC Madagascar project. Authors are grateful to french cooperation for technical and financial support.

Y. Tharault, G. Mourot, J. Ragot, and D. Maquin, “Fault and

isolation

with

robust

principal

compone nt analysis,” International Journal of Applied Mathematics and Computer Science, ED 11, vol.4, pp. 429442, 2008. [2]

F. Harkat, G. Mourot, and J. Ragot, “Diffé re ntes

vol.11, pp.19-33, 2001. [8]

J. Karhune n, “Robust PCA me thods for comple te and missing data,” Aalto Unive rsity School of Scie nce1, De pt. of Information and Compute r Scie nce , Espoo, Finland 2011.

[9]

G.R. Halligan, “Fault de tection and pre diction with application to rotating machine ry,” PhD, Missouri

[10] J.F. Ramahale omiarantsoa, E.J.R. Sambatra, N. Hé raud,

and J.M. Razafimahe nina, Performances of the PCA method in electrical machines diagnosis using Matlab, [11] M. Wieczore k, E. Rosołowski, “Mode lling of induction

motor for simulation of inte rnal faults,” Modern Electric Power Systems, MEPS'10, Wroclaw, Poland, p. 29, 2010. [12] A. Ste fani, “Induction Motor Diagnosis in Variable

Spee d Drives,” PhD in Ele ctrical Enginee ring Final Dissertation, University of Bologna, 2010.

CIFA’02,

Analysis Me thod Application in Electrical Machines

Francophone

d’Automatique,

Nantes, France , 2002.

Diagnosis,” International Conference on Informatics in

J. Mina, C. Ve rde , “Fault dete ction for large scale

Control, Automation and Robotics, ICINCO, Pays Bas,

systems using dynamic principal compone nts analysis

2011. [14] A.

Be naicha, M. Gue rfe l, N. Bouguila, and K.

Communication & Control, vol. II, N°2, pp. 185-194, 2007.

Be nothman, “Ne w PCA-base d me thodology for se nsor

J. F. Ramahaleomiarantsoa, N. Hé raud, E.J.R. Sambatra

fault

rotor induction

motor

by

principal components

analysis,” International Conference on Production, ICPR, Stuttgart Germany, 2011. L. Lui, “Robust fault de tection and diagnosis for pe rmane nt

magne t

synchronous

motors,”

PhD

dissertation, Colle ge of Enginee ring, The Florida State Unive rsity, USA, 2006.

270

principal component analysis,” Journal of Process Control,

and J.M. Razafimahe nina, “Principal Components

principales,”

and J.M. Razafimahe nina, “Fault dete ction of a wound

[6]

B. Huang, “Process ide ntification base d on last

Conférence

composantes

with adaptation,” International Journal of Computers,

[5]

Control

[13] J.F. Ramahale omiarantsoa, N. Hé raud, E.J.R. Sambatra

Internationale

[4]

diagnostics,”

mé thodes de localisation de dé fauts basées sur les de rniè res

[3]

e ngine

INTECH, MATLAB / Book 1, ISBN 979-953-307-774-0.

REFERENCES

de tection

aircraft

Unive rsity of Scie nce and Te chnology, 2009.

ACKNOWLEDGMENT

[1]

[7]

on-line

de tection

Conference

of

and Modeling

localization,”

International

Simulation,

MOSIM’10,

Hammame l, Tunisia, 2010. [15] G. Spitzlspe rger “Fault de tection for a via Etch process

using

adaptive

multivariate

me thods”

IEEE,

Transaction on semiconductor manufacturing, vol. 18, N° 4, pp. 528-533, 2005. [16] J. Wang, S. J. Qin, “EWMA Kalmar Filter and Recursive

S. Borque t and O. Lé onard, “Coupling principal

Least Squares - re lationships and modifications,”

compone nt analysis and Kalman filtering algorithms

TWMCC, Spring Meeting, Austin, 2003.


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Ramahaleomiarantsoa Jacques was born in November 1966 in Antananarivo Madagascar. He has obtaine d the Dipl. of e nginee ring in e lectrical e nginee ring powe r transmission option in 1995 at the Polytechnic School of Antsiranana (ESPA) Madagascar. PhD stude nt at the Unive rsity of Corsica, France and the ESPA. Research professor at the ESPA and at the High Te chnology Institute of Antsiranana (IST D). His research focuses on fault diagnosis system and rural e le ctrification base d on re ne wable e nergy. Heraud Nico las was born in France on se ptember 15, 1962. He rece ive d his Ph.D. de gree in Automatic and Ele ctrical Enginee ring from Institut National Polytechnique de Lorraine in 1991. Since 1992, he teaches at the Unive rsity of Corse as professor and he is at the CNRS (UMR 6134). His fie ld of inte rest includes data reconciliation and process diagnosis on re ne wable ene rgy systems.

Sambatra Eric Jean Roy was born in Antsirabe , Madagascar, on December 11, 1975. He rece ive d his Ph.D. de gree in Ele ctrical Engineering from the Ele ctrical and Automatic Re search Team of Le Havre Unive rsity (GREAH) in 2005. Since 2009, he teaches e lectrical e ngineering and rene wable e nergy systems at the IST-D (Institut SupĂŠ rie ur de Technologie ) and ESPA (Ecole SupĂŠ rie ure Polyte chnique ), Antsiranana, Madagascar. His research inte rests are rene wable ene rgy systems, e le ctrical machines and diagnosis. Razafimahenina Jean Mar ie was born in Fianarantsoa in August 1950; he obtaine d Dipl. of e nginee ring e le ctricity and powe r e lectronics in 1979 from the Polytechnic School of Antsiranana. He graduate d Compre hensive Study (DEA) in powe r e le ctronics at the ESPA in 1981, PhD in photovoltaic ene rgy in 1986 and doctorate in e lectrical ne tworks in 2005. He is a Full Professor at the ESPA and Higher Institute of Technology of Antsiranana, Madagascar.

271


Inte rnational Journal of Ene rgy Scie nce Vol. 2 Iss. 6, De ce mbe r 2012

Evaluation of the Quality of Service Parameters for Routing Protocols in Ad-Hoc Networks Zeyad Ghaleb Al-Mekhlafi1, Rosilah Hassan2, Zurina Mohd Hanapi3 Universiti Putra Malaysia (UPM) 43400 UPM Serdang, Selangor, Malaysia

1.3

Universiti Kebangsaan Malaysia, 43600 UKM Bangi, Selangor, Malaysia

2

Ziadgh2003@hotmail.com; rosilah@ftsm.ukm.my; zurina@fsktm.upm.edu.my

Abstract Recently, many researchers have focused on the Ad-Hoc networks especially the routing protocols which include reactive and proactive routing protocols. The ultimate goal of routing protocols is forwarding data packet from the source to the destination. Consequently, several proactive routing protocols, such as routing information protocol (RIP), and reactive routing protocols, such as Dynamic Source Routing (DSR), are based on exploring, maintenance, and recuperating the route path. The likely problem in the Ad-Hoc networks is how to establish the best routing protocol that assures the requirements of the application concerning about some criteria. This work presents the evaluation of RIP and DSR utilizing the QualNet simulation. Furthermore, the achievement of these routing protocols was assessed based on the throughput, average jitter, average end-to-end delay, and energy consumption metrics. This paper demonstrates that the RIP has superior evaluation performance as compared to DSR in two different scenarios (effect of the number of nodes and effect of packet size). Keywords Routing Protocols; Average Jitter; Average End-to-End Delay; Throughput; Energy Consumption

Introduction The new revolutions in wireless technology have led to the emergence of a new wireless system which is called Ad-Hoc Network. Ad-Hoc Network is a kind of wireless system which allows direct communication with each other. In Ad-Hoc network, each node plays a dual role; a router and a host in the sense at the same time. The process of sending and receiving data packages is controlled by getting some information regarding the surrounding network and dealing with algorithm. This combination between these functions is known as a routing protocol. A number of studies have recently gained attention in using the routing protocols, particularly, proactive routing

272

protocol and reactive routing protocol [1, 2]. Proactive routing protocols are those protocols which carry out the function of keeping track of routes for all the destinations in the Ad-Hoc networks. They are supported to be available in the form of tables. Furthermore, proactive routing protocol periodically exchange routing information in the whole network and maintains routes between different nodes dynamically. They have low latency and high overhead, and the routes are reliable. These protocols cannot scale well with the increase in network size. It is stated that one advantage of applying such kinds of protocols is that they facilitate communication to undergo minimal initial delay in the application procedure. However, their disadvantage is represented by the fact that they require additional control traffic to constantly update the entries of the stale route. On the other hand, reactive routing protocols attempt to identify a path to the destination only when a packet of data sent to the destination is received by the network protocol. This is one advantage of such kind of protocols as the degree of uncertainty in the node position is found to be high. They have also proved to be more suitable and more distinguished by their better performance in Ad-Hoc networks. However, taking more time to find a route and requiring more flooding which results into clogging the network are among the disadvantages of such protocols. Therefore, the arrangement of forwarding data packet from the source to destination is the ultimate aim by utilizing routing protocols. The differences between these protocols are due to the differences in the searching, maintenance and recovering the route path. The decision of choosing the best routing protocol should take into account some considerations such as mobility of nodes, packet size, cost of path, application type, number of nodes, type of traffic, and Quality of Services (QoS).


Inte rnational Journal of Ene rgy Scie nce Vol. 2 Iss. 6, De ce mbe r 2012

On the whole, QoS shore up in wireless is an extremely demanding issue because of their dynamic character [3, 4]. Diverse techniques, as of physical layer capable of application layer, have been wished-for to supply QoS shore up in wireless Ad-Hoc networking surroundings [5]. Recently, a cross-layer design move toward in QoS conditioning in wireless networks has gained more research interest [6, 7]. Consequently, this paper focuses on the most important factors, namely end-to-end delay, average jitter, throughput and energy consumption. The end-to-end delay is important for the Ad-Hoc networks due to the fact that some of the real-time applications are very sensitive to the delay which means that the data packet sent from the source node should be delivered to the final target node within a specific period of time without any delay. Therefore, the routing protocol will be selected based on the shortest path from the source node to the destination node. The average jitter assesses the variability over time of the packet latency across a network which associated with the delay. The network with constant delay has no jitter. Therefore, the routing protocol that satisfies the constant delay without any variation during the time will be more suitable to be selected for data routing. Moreover, the significance of throughput come from the needs to deliver the more messages to destination nodes during a specific period of time which means that the routing protocols should use some mechanisms to avoid the congestion in some paths which are more frequently used to prevent the packet drops during the data routing. Hence, the reactive routing will be getting a better chance as compared to the proactive routing, to be chosen as it can find alternative paths to be used rather than the congested one. Another mechanism to increase the throughput of routing protocols, in order to be chosen, is how to deal with the failures of the paths during the data delivery; meaning that if the current path used no more available either by the node failure or moving from the current position, the routing which deals with this issue will be more preferred by the user. Beside these, energy consumption is an important factor especially in mobile Ad-Hoc networks which has restricted energy. Therefore, the routing protocol should consider this factor by chosen the paths that consume small energy to extend the lifetime of the node and give the chance to the connectivity of the network to be longer. Moreover, the nodes of paths which routed the data packets will deplete their energy very fast and run-out their batteries. Therefore, the routing protocol must look for new paths to avoid using the same path repeatedly and consuming much energy. Again, the reactive

protocols will be more preferred because of their ondemand property. Related Works In [8], an Ad-Hoc routing protocol, namely Ad-Hoc On demand Distance Vector (AODV) has been evaluated. According to this model, the performance of AODV in homogeneous Ad-Hoc was better than heterogeneous one. A performance analysis of proactive and reactive routing protocols for Ad-Hoc networks Dynamic Destination-Sequenced Distance Vector (DSDV), AODV and Dynamic Source Routing (DSR) showed that the performance of AODV was better in dense environment except packet loss [9]. Moreover, it was found that both DSR and AODV performed well, and they proved to be better than DSDV. However, it is not clear which protocol is the best for all scenarios, even though there are rapid growth and development in the field of Ad-Hoc network. A comparison of the parameters of routing protocols between these previous studies is shown in table 1. TABLE 1 COMPARISON OF THE PARAMETERS OF ROUTING PROTOCOLS BETWEEN PREVIOUS STUDIES Parameter

(Tyagi& Chauhan, 2010)

(Ismail& Hassan,2010)

Numberof nodes

10-200

5,7

Simulationtime

1200sec(20Min)

3000s

Simulationarea

800Х1200 m

500Х500 m,1000Х1000 m, 1500Х1500 m,2000Х2000 m, 2500Х2500 m.

Routing protocols

DSDV,AODV,DSR

AODV

Transmission range

250 m

250 m

Packetsize

512 bytes

100,200,300,400,500,600,700,800,900 and1000 bytes

MAC protocol

802.11

802.11

Mobility type

Randomway point

Randomway point

Type oftraffic

CBR

CBR

Packetrate

54 Mps

54 Mps

Speed

(10-100) m/s

2 Mps

Program simulation

NS-2

OMNeT++

A comparative review study on reactive and proactive routing protocols in MANETs provided information about several routing schemes proposed for Ad-Hoc networks [10]. These schemes were classified according to the routing strategy (i.e., Proactive and Reactive). It is

273


Inte rnational Journal of Ene rgy Scie nce Vol. 2 Iss. 6, De ce mbe r 2012

shown that each protocol has definite advantages and disadvantages and is well studied for certain situations. Despite of the rapid growth in the field of Ad-Hoc networks, many challenges still exist and need more attention and consideration from researchers so that it is possible for such networks to be used more widely within the next few years. Recently, we have evaluated the routing information protocol and dynamic source routing [11]. According to this model, Routing Information Protocol (RIP) was found to be better as compared to Dynamic Source Routing (DSR).

network protocol requires knowing the next node in the path and the outgoing interface on which to send the packet [15]. A routing protocol computes routing information such as homogeneous and heterogeneous networks [8, 16]. Overall, routing protocols can be classified into two categories: proactive (table driven) routing protocols and reactive (on-demand) routing protocols. Popular proactive routing protocols are (DSDV) [17], Open Shortest Path First (OSPF) [18, 19], and RIP [20], whereas reactive routing protocols include DSR [21] and AODV [22].

Performance evaluation of AODV, DSDV, and DSR Routing Protocol in Grid Environment was described in a previous study [12]. According to this model, the AODV, DSR, and DSDV perform very well when the mobility is high. However, simulation results showed that the traditional routing protocols like DSR have a dramatic decrease in performance when the mobility is high. In [13], the performance of routing protocols in mobile Ad-Hoc network was compared for DSDV, AODV, and DSR and showed that DSR outperforms AODV. The DSR has less routing overhead when nodes have high mobility considering the throughput, end-to-end delay and packet delivery ratio metrics while DSDV produces low end-toend delay compared to AODV and DSR. In [14], the evaluation four Ad-Hoc network protocols (AODV, DSDV, DSR and TORA) in diverse network scales taking into contemplation the mobility factor. Based on this model, the throughput and energy consumption in tiny size networks did not disclose any momentous differences. On the other hand, for medium and huge AdHoc networks the TORA concert proved to be incompetent in this research. Above all, the concert of AODV, DSDV and DSR in tiny size networks was equivalent. Other than in medium and large size networks, the AODV and DSR formed good results and the concert of AODV in terms of throughput is good in all the scenarios that have been investigated.

Routing Information Protocol

Thus, our work in this present study is to use the more widely used traditional mobility models and traffic sources to create observations based on more standardized methodology that can be used to evaluate which protocol, proactive routing protocol (RIP) or reactive routing protocol (DSR), is more stabile for AdHoc networks based on some criteria in QualNet simulation. Ad-Hoc Routing Protocols The routing protocol resolves the path of a packet from the source to the destination. To forward a packet, the

274

RIP is a routing protocol which is dynamic as OSPF, but it is widely used in both local and wide area networks. It is classified as an Interior Gateway Protocol (IGP) which makes a use of the distance-vector routing algorithm as proposed in 1988 [23]. Since then, RIP Version 1 has been extended and updated to RIP Version 2 in 1998 [20]. It is indicated that both RIP versions are still being used today, but they have been technically supported by more advanced techniques such as OSPF and Open Systems Interconnection (OSI) protocol; Intermediate System to Intermediate System (IS-IS). Moreover, RIP has been updated to IPv6 network which is known as a standard RIP next generation (RIPng). One of the advantages of employing RIP is that it is simple to understand and easy to configure as it is capable of being supported by all routers, support load balancing, and in general, it is free from loop. However, among the disadvantages, RIP is not efficient, slow when it is used in large networks due to its configuration, supports equalcost load balancing, its congestion raises a problem and its scalability is limited since it is only measured as 15 hop maximum. Dynamic Source Routing Dynamic Source Routing (DSR) is defined by Johnson and Maltz [24] as a routing protocol which is still on demand and in which the sender of data can determine exactly the required sequence of nodes to propagate a packet. This packet header includes a number of intermediate nodes for routing. Each node works to maintain the route cache which cashes the source route being learned. It is stated that “Route Discovery” and “Route Maintenance” are the two main components of DSR which work together to determine and maintain routes to random destinations. The purpose of designing such protocol is to make restrictions to the large consumption of bandwidth caused by control packets in Ad-Hoc wireless networks. This process is done by deleting the messages of the


Inte rnational Journal of Ene rgy Scie nce Vol. 2 Iss. 6, De ce mbe r 2012

periodic updates required which usually appears in the table-driven approach [25]. The possibility of establishing a route when necessary makes the sender to be able to choose and control routes by reducing the load of data and including routing which is free from loop containing unidirectional links in networks is all the main advantages of DSR. However, DSR may lead to significant overheads because the source route has to be included with each packet. It uses cashing excessively and lacks mechanisms by which it can detect the freshness of the routes which causes delay and reduction; hence, the route mechanism for maintenance is unable to repair a broken link locally. Therefore, this makes the delay of the connection setup higher than that found in table-driven protocols [26].

b. Average Jitter Average Jitter is known as the time variation measured between the arrival of the packets due to the congestion of the network, the drift in timing, or changing of the route [2]. c. Throughput Throughput is the number of delivered packet per unit of time [28]. d. Energy Consumption It is defined as the amount of energy consumed in a process or system, or by an organization or society. It is the summation of the idle mode, transmit mode, and receive mode [29].

Metrics for Evaluation

Simulation Environments

Corson and Macker showed that the evaluation metrics are possible to be made a use of in evaluating the quantitatively Mobile Ad-Hoc Network (MANET) routing protocols [27]. Such quantitative measurement is useful as a prerequisite for assessing or evaluating the performance of network or even to compare the performance using different routing protocols.

In this paper, the QualNet simulation was implemented; 802.11 MAC [30]. The parameters in the simulation such as number of nodes, time of simulation, packet size, and type of traffic were summarized in Table 2.

Materials and Methods Simulation Tools The objective of this QualNet Version 5 simulation is to evaluate the proactive routing protocol and reactive routing protocol in Ad-Hoc networks in two scenarios. In a previous study [11], the effect of the number of nodes was evaluated. Beside this effect, the current study also covered the effects of packet size. It has five experiences with different number of nodes for scenario I (effects the number of nodes), and seven experiences with different packet size for scenario II (effects of packet size). The evaluation metrics used are throughput, end-to-end delay, average jitter, and energy consumption. a. Average End-To-End Delay This refers to the interval taking place between the data packet generation time and the time of the arrival of the last bit to the destination i.e. the average amount of time taken by a packet to move from source to destination. The process includes all possible delays which happen due to buffering during route discovery latency, queuing at the interface queue, retransmission delays at the Media Access Control (MAC) and propagation and transfer times [9].

TABLE 2 PARAMETERS SETUP Parameter

Scenario I

Scenario II

Numberof nodes

50,90,130,170,210

7

SimulationTime

1200sec(20Min)

3000s

Simulationarea

800ĐĽ1200m

500ĐĽ500m

Routing protocols

RIP andDSR

RIP andDSR

TransmissionPower

25dBm

25dBm

TransmitPower Consumption

100mW

100mW

Receive Power Consumption

130mW

130mW

Idle PowerConsumption

120mW

120mW

Transmissionrange

270m

270m

TransmissionPower

25.0

25.0

ItemSize

512bytes

100,200,300,400,500,600 and700 Bytes

PHY

802.11b

802.11b

Type oftraffic

CBR

CBR

Data Rate

11Mbps

11Mps

Speed

(10-100) m/s

(10-100) m/s

The number of nodes ranges from 50 to 210 nodes which divided into 50, 90, 130, 170, and 210 and the packet size range from 100 bytes to 700 bytes which divided into 100, 200, 300, 400, 500, 600, and 700 bytes. Five reasons

275


Inte rnational Journal of Ene rgy Scie nce Vol. 2 Iss. 6, De ce mbe r 2012

experiences with different number of nodes and seven reasons experiences with different packet size were implemented in this work. Evaluation of Results Results are obtained after the experiments have been conducted. The present paper aims to demonstrate the evaluation performance of each routing protocol with respect to the effects of the number of nodes and effects of packet size. The evaluation metrics considered for average jitter, end-to-end delay, throughput, and energy consumption. The tests highlight the evaluation performance of RIP and DSR in Ad-Hoc network. Scenario I Average End-To-End Delay Data set of the effects of the number of nodes by QualNet simulation of Average End-to-End Delay (scenario I) is shown in Table 3. TABLE 3 DATA SET OF AVERAGE END-TO-END DELAY Scenario I

Throughput

Average End-to-End Delay(s) No of Nodes

DSR

RIP

50

0.079186

0.058514

90

0.197886

0.069717

130

0.207281

0.052935

170

0.063845

0.03455

210

0.191009

0.04776

FIG. 1 AVERAGE END-TO-END DELAY BETWEEN RIP AND DSR IN SCENARIO I

276

Figure 1 shows the influence of the number of nodes on network average end-to-end delay for two routing protocols. The average end-to-end delay values increased according to the number of nodes for DSR. The maximum average end-to-end delay gained simulation with 130 numbers of nodes from DSR and the minimum average end-to-end delay gained from simulation 170 numbers of nodes from DSR. The increase average endto-end delay values the increase and the decrease according to the number of nodes for RIP. The maximum average end-to-end delay gained simulation with 90 numbers of nodes from RIP and the minimum average end-to-end delay gained from simulation 170 numbers of nodes from RIP. From the graph, it is clear that RIP out performs DSR for scenario I or II of varying pause time, varying simulation time, varying speed and varying number of nodes. In case of DSR, delay time increased sharply with increasing number of nodes. However, a sharp decrease was noticed when the number of nodes is 170. On the other hand, RIP increased and then decreased with increasing number of nodes. It is important to note that RIP gave a low end-to-end delay as compared to DSR.

Data set for the effects of the number of nodes by QualNet simulation of Throughput (scenario I) is demonstrated in Table 4. TABLE 4 DATA SET OF THROUGHPUT Scenario I Throughput (bits/s) No of Nodes

DSR

RIP

50

2312

2320

90

3

2301.75

130

6

1532.33

170

14

2285

210

6

2343.25

Figure 2 shows the influence of the number of nodes on network throughput for two routing protocols (RIP and DSR). The throughput values increased according to the number of nodes for RIP while in DSR it first increased when the number of nodes rose to 50 after which it starts to decrease sharply with increasing number of nodes. The maximum throughput was gained from simulation with 210 nodes for RIP and the minimum throughput was


Inte rnational Journal of Ene rgy Scie nce Vol. 2 Iss. 6, De ce mbe r 2012

gained from simulation with 130 nodes. The maximum throughput was gained from simulation with 50 nodes from DSR and the minimum throughput has gained from simulation with (90,130,170,210) numbers of nodes. RIP have higher throughput value compared to DSR.

FIG. 3 AVERAGE JITTER BETWEEN RIP AND DSR IN SCENARIO I

Energy Consumption

FIG. 2 THROUGHPUTS BETWEEN RIP AND DSR IN SCENARIO I

Average Jitter Data set for the effects of the number of nodes by QualNet simulation of Average Jitter (scenario I) is shown in Table 5.

In energy consumption, the result was calculated by collecting Idle mode + Transmit mode + Receive mode. The energy consumption was represented in two tables: Table 6 for the Idle mode, Transmit mode and Receive mode and Table 7 for the collected energy consumption (Idle mode + Transmit mode + Receive mode). TABLE 6 ATA SET FOR ENERGY CONS UMPTION FOR IDLE MODE, TRANSMIT MODE AND RECEIVE MODE DSR

TABLE 5 DATA SET OF AVERAGE JITTER Scenario I Average Jitter (s) No of Nodes

DSR

RIP

50

0.0365204

0.015466

90

0

0.036365

130

0.0248375

0.018677

170

0.0143463

0.000938

210

0.0224834

0.01431

The two kinds of routing protocols have different jitter with the increased number of nodes, as shown in Figure 3. Overall, RIP showed a better jitter than DSR when the number of nodes is greater than 50 while DSR showed the better jitter than RIP, when the number of nodes is 90 but when the number of nodes is above 90, the RIP gave a better jitter than DSR.

No of Nodes

50

90

130

170

210

Receive mode

0.066248

26.4599

33.6272

36.8386

29.895

Transmit mode

0.020879

0.008001

0.013737

0.007616

0.007551

Idle mode

39.9387

15.5754

8.95939

5.99503

12.4046

RIP No of Nodes

50

90

130

170

210

Receive mode

2.25513

2.35914

2.96544

3.51517

4.11252

Transmit mode

0.21928

0.398631

0.577919

0.718834

1.01791

Idle mode

37.9164

37.8188

37.2575

36.7488

36.1947

277


Inte rnational Journal of Ene rgy Scie nce Vol. 2 Iss. 6, De ce mbe r 2012

TABLE 7 DATA SET OF THE COLLECTED ENERGY CONS UMPTION

TABLE 8 DATA SET OF AVERAGE END-TO-END DELAY Scenario II

Scenario I

End-to-End Delay(s) Energy Consumption No of Nodes

DSR

RIP

50

40.02583

40.39081

90

42.0433

40.57657

130

42.60033

40.80086

170

42.84125

40.9828

210

42.30715

41.32513

Packet Size

DSR

RIP

100

6.5376

0.00089

200

6.54139

0.00085

300

6.43877

0.000777

400

6.73125

0.000939

500

6.06969

0.000761

600

6.41203

0.000566

700

6.81644

0.000714

Fig. 5 Average End to EndDelay between RIP andDSR in scenario II.

Throughput FIG. 4 ENERGY CONSUMPTION BETWEEN RIP AND DSR IN SCENARIO I

The energy consumption for the two routing protocols increased at the beginning of this work, as shown in Figure 4. DSR has a longer consumption than RIP. Therefore, RIP has the better energy consumption than DSR except when the number of nodes is 50 nodes.

Data set of the effects of packet size by QualNet simulation of Throughput (scenario II) is shown in Table 9. TABLE 9 DATA SET OF THROUGHPUT Scenario II Throughput (bits/s) Packet Size

DSR

RIP

Scenario Ii

100

6.5376

0.00089

Average End-to-End Delay

200

6.54139

0.00085

Data set of the effects of packet size by QualNet simulation of average End-to-End Delay (scenario II) is presented in Table 8.

300

6.43877

0.000777

400

6.73125

0.000939

500

6.06969

0.000761

600

6.41203

0.000566

700

6.81644

0.000714

Figure 5 shows that the average end-to-end delay for two routing protocols decreased; except when the packet size of DSR was higher than 100 bytes. Thus, DSR has longer delay than RIP and RIP exhibits shorter delay than DSR.

278


Inte rnational Journal of Ene rgy Scie nce Vol. 2 Iss. 6, De ce mbe r 2012

FIG. 6 THROUGHPUTS BETWEEN RIP AND DSR IN SCENARIO II

FIG. 7 AVERAGE JITTER BETWEEN RIP AND DSR IN SCENARIO II

Figure 6 shows the influence of the packet size on the network throughput for two routing protocols. Overall, the throughput value increased with the packet size for the two routing protocols. The maximum throughput gained from simulation with 700 bytes packet size, while the minimum throughput gained from simulation with 100 bytes packet size. On the other hand, DSR has the maximum throughput values according to increase packet size compared to RIP. Therefore, the DSR has better throughput than RIP.

The two kinds of routing protocols have different jitter with increased packet size (Fig 7). In general, RIP had better jitter than DSR while DSR showed longer delay than RIP. Thus, RIP showed the best evaluation performance.

Average Jitter Data set of the effects of packet size by QualNet simulation of Average Jitter (scenario II) is presented in Table 10. TABLE 10 DATA SET OF AVERAGE JITTER Scenario II Average Jitter (s)

Energy Consumption There are two tables to show the energy consumption: table 11 for the Idle mode, Transmit mode and Receive mode while table 12 was for the collected result (Idle mode + Transmit mode + Receive mode). TABLE 11 DATA SET FOR ENERGY CONS UMPTION OF IDLE MODE, TRANSMIT MODE AND RECEIVE MODE DSR Packet Size

100

200

300

400

500

600

700

Receive mode

0.01 317

0.01 5901

0.0175 71

0.016 744

0.0186 13

0.0188 96

0.019 2

Packet Size

DSR

RIP

Transmit mode

0.04 5568

0.05 6014

0.0631 08

0.060 627

0.0687 09

0.0692 31

0.071 318

100

0.956555

0.001107

Idle mode

149. 958

149. 948

149.94 2

149.9 44

149.93 7

149.93 7

149.9 35

200

1.03527

0.000909

300

0.997965

0.000897

400

1.04567

0.001143

500

1.03995

0.000736

600

1.04009

0.000409

700

1.05922

0.000677

RIP Packet Size

100

200

300

400

500

600

700

Receive mode

0.00 8079

0.00 7198

0.0126 69

0.008 29

0.0077 53

0.0100 93

0.009 364

Transmit mode

0.02 9998

0.02 7191

0.0467 71

0.031 318

0.0295 84

0.0360 45

0.036 08

Idle mode

149. 973

149. 975

149.95 7

149.9 72

149.97 3

149.96 7

149.9 67

279


Inte rnational Journal of Ene rgy Scie nce Vol. 2 Iss. 6, De ce mbe r 2012

TABLE 12 DATA SET OF THE COLLECTED ENERGY CONS UMPTION (IDLE MODE + TRANSMIT MODE + RECEIVE MODE) Scenario II

scenarios. REFERENCES [1]

S.-J. Lee, J. Hsu, R. Hayashida, M. Gerla, R. Bagrodia, Selecting a routing strategy for your ad hoc network.

Energy consumption

Computer Communications 26 (2003) 723-733.

Packet Size

DSR

RIP

100

150.0167

150.0111

and simulations of routing protocols in ad hoc networks.

200

150.0199

150.0094

Computer Communications 30 (2007) 1890-1898.

300

150.0227

150.0164

400

150.0214

150.0116

500

150.0243

150.0103

600

150.0251

150.0131

700

150.0255

150.0124

[2]

[3]

L. Layuan, L. Chunlin, Y. Pe iyan, Performance evaluation

D. Wu, QoS provisioning in wire less networks. Wireless communications and mobile computing 5 (2005) 957-969.

[4]

A. Duda, C.J. Sreenan, Challenges for quality of service in ne xt ge neration mobile networks. parameters 4 (2003) 3G.

[5]

P. Mohapatra, J. Li, C. Gui, Qos in mobile a hoc networks. Wire less Communications, IEEE 10 (2003) 44-52.

[6]

Q. Zhang, F. Yang, W. Zhu, Cross-layer QoS support for multimedia delivery over wire less internet. EURASIP Journal on Applie d Signal Processing 2005 (2005) 207-219.

[7]

S.B. Lee , G.S. Ahn, X. Zhang , A.T. Campbell , INSIGNIA : An IP-base d quality of service frame work for mobile ad hoc networks. Journal of

Parallel and Distribute d

Computing 60 (2000) 374-406. [8]

Z. Ismail, R. Hassan, Evaluation of ad hoc on demand distance vector routing protocol in He tMAN architecture. Journal of Computer Science 6 (2010) 830-836.

[9]

S. Tyagi, R. Chauhan, Performance Analysis of Proactive and Reactive Routing Protocols for Ad hoc Ne tworks. International Journal of Computer Applications IJCA 1

FIG. 8 ENERGY CONSUMPTION BETWEEN RIP AND DSR IN SCENARIO II

(2010) 31-34. [10]

revie w study of reactive and proactive routing protocols in

The two types of routing protocols have different energy consumption with increasing packet size as shown in Figure 8. DSR has longer energy consumption than RIP, while RIP has smaller energy consumption than DSR. As a result, the RIP showed the best evaluation performance in energy consumption.

[11]

Conclusion

[12]

In the present paper, an evaluation for routing protocols was carried out on acquired simulation results of two routing protocols, RIP and DSR using QualNet V5. RIP and DSR were selected to represent the Proactive routing protocols and Reactive routing protocols, respectively. We found that Routing Information Protocol preformed better than DSR for all evaluation metrics in 2 different

280

S. Mohseni, R. Hassan, A. Pate l, R. Razali, Comparative MANETs, in, IEEE, 2010, pp. 304-309. Z.G. Al-Me khlafi, R. Hassan, Evaluation Study on Routing Information Protocol and Dynamic Source Routing in AdHoc Ne twork, in, 7th Conference on IT in Asia 2011(CITA11). Kuching Sarawak Malaysia, 2011,pp.256-259. N.S.M. Usop, A. Abdullah, A.F.A. Abidin, Performance evaluation of AODV, DSDV & DSR routing protocol in grid e nvironment. IJCSNS 9 (2009) 261-268.

[13]

P. Manickam, T.G. Baskar, M. Girija, D.D. Manimegalai, Pe rformance Comparisons of Routing Protocols in Mobile Ad Hoc Networks. Arxiv preprint arXiv:1103.0658 (2011).

[14]

V. Kanakaris, D. Ndzi, D. Azzi, Ad-hoc networks energy


Inte rnational Journal of Ene rgy Scie nce Vol. 2 Iss. 6, De ce mbe r 2012

consumption: a revie w of the ad-hoc routing protocols.

[22]

(JESTR) 3 (2010) 162-167. [15]

[23]

Based

Approach

for

Reducing

Broadcast

[24]

Re dundancy with Fast Network Coverage in Wire less AdHoc Ne tworks, in, 5th International Conference on IT in [16]

[20]

[21]

[25]

G. Acs, L. Buttyan, I. Vajda, Provably secure on-demand

Q. Guo, X. Xu, J. Zhu, H. Zhang, A QoS-guaranteed cell

Transactions on Mobile Computing (2006) 1533-1546. [26]

M. Uma, Padmavathi, A

Comparative Study and

Journal 28 (2006) 77-83.

Pe rformance Evaualtion of Reactive Quality of Service

C.E. Perkins, P. Bhagwat, Highly dynamic destination-

Routing Protocols Mobile ADHOC Ne tworks. Journal of

sequenced distance-vector routing (DSDV) for mobile

The oretical and Applie d Information Technology 6 (2009) 2. [27]

S. Corson, J. Macker, Mobile Ad hoc Networking

Re vie w 24 (1994) 234-244.

(MANET): Routing Protocol Performance Issues and

R. Coltun, D. Ferguson, J. Moy, A. Lindem," OSPF for IPv6,

Evaluation Considerations[S], RFC2501, 1999.

in, RFC 5340, July, 2008. [19]

hoc wireless networks. Mobile computing (1996) 153-181. source routing in mobile ad hoc networks. IEEE

computers. ACM SIGCOMM Computer Communication [18]

D.B. Johnson, D.A. Maltz, Dynamic source routing in ad

Asia (CITA'07), 2007. selection strategy for heterogeneous cellular systems. ETRI [17]

C.L. He drick, RFC1058: Routing information protocol. RFC Editor Unite d States (1988).

M.J. Islam, M.A. Khaer, M.N. Islam, M.M. Islam, Simplified XOR

C. Perkins, E. Belding-Royer, S. Das, Ad hoc on-demand distance vector (AODV) routing. (2003).

Journal of Engineering Science and Technology Revie w

[28]

N. Mishra, N.N. AmitPandey, R. Sinha, D.S. Tapaswi,

J. Moy, RFC2328: OSPF Version 2. RFC Editor Unite d States

Selection of Ad Hoc Network Routing Protocols by

(1998).

Pe rformance Analysis, in, 2008, pp. 113-116.

G. Malkin, RFC2453: RIP Version 2. RFC Editor United

[29]

M.C. Domingo, Packet Size Optimization for Improving

States (1998).

the Energy Efficiency in Body Sensor Networks. ETRI

D.B. Johnson, D.A. Maltz, J. Broch, DSR: The dynamic

Journal 33 (2011).

source routing protocol for multi-hop wire less ad hoc ne tworks. Ad hoc networking 5 (2001) 139â&#x20AC;&#x201C;172.

[30]

Y. Tay, K.C. Chua, A capacity analysis for the IEEE 802.11 MAC protocol. Wireless networks 7 (2001) 159-171.

281


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

An Investigation of Power Performance of Small Grid Connected Wind Turbines under Variable Electrical Loads Md. Alimuzzaman1 , M.T.Iqbal2 , Gerald Giroux3 Faculty of Engineering and Applied Science, Memorial University of Newfoundland, St. Johnâ&#x20AC;&#x2122;s, NL Canada A1B 3X5 1,2

Wind Energy Institute of Canada (W EICan), 21741 Route 12, North Cape, PEI, Canada, C0B 2B0

3

ma6762@mun.ca; 2tariq@mun.ca; 3gerald.giroux@weican.ca

1

Abstract In this study, the power pe rformance of two small grid connecte d wind turbine s has bee n investigate d. The objective was to study the impact of load power factor on the wind turbine powe r curve . Two small wind turbine s were teste d in a numbe r of load conditions and test data were collecte d for about two months. A se t of resistors, inductors and capacitors were use d as load in addition to the grid connection. Eve ry second set of te st data was collecte d for at least two days in each load condition. Data was analyse d for active power, power factor and reactive powe r. Wind turbinesâ&#x20AC;&#x2122; powe r curves are plotte d with load and without any load connecte d be twee n the wind turbine and the grid. Results indicate that the type of load does not significantly affect the powe r curve of a small wind turbine . It was also obse rve d that above 200W the powe r factor of a small grid connecte d wind turbine was also constant. Keywords Small wind turbine power pe rformance; grid connected small wind turbine; wind turbine under variable electrical loads; reactive power of small wind turbine; power factor of small wind turbine

Introduction The wind industry has been contributing a significant percentage of electric power generation all over the world. The power performance and power quality of wind turbines and their interaction with the grid is becoming an important issue [1]. Small wind turbines are being widely used to fulfil local demands. They are used for dairy farms, water supplies for small communities, small industry, irrigation and greenhouses. Many of these turbines are grid connected, so when there is excess electricity they can sell the extra power to a grid and when there is a lack of electric power from the wind turbine, they can purchase electricity from the grid.

282

To produce electric energy, doubly-fed induction generators (DFIG) in large wind turbines and permanent magnet generators (PMG) in small wind turbines have been widely used. Since a PMG has its own permanent magnet, it does not require an external excitation current, so it does not consume reactive power from the grid. Also, it presents high efficiency and a small size compared to a DFIG. That is why PMGs are dominant in small wind turbine systems [2]. Power performance for small wind turbines is very important. The manufacturers provide a power curve for their wind turbines, which is essentially turbineproduced active power versus wind speed. Depending upon the situation, the small wind turbine load may be resistive, inductive or sometimes capacitive in nature. As the small wind turbines are in dispersed locations, they can be used to provide local reactive power consumption. That may decrease the reactive power flow and also decrease the overall power losses [3]. In this article, the power performance of small wind turbines has been investigated. Two wind turbines were tested at the Wind Energy Institute of Canada (WEICan) with a resistive load, inductive load and capacitive load. Recorded data has been compared and analysed for active power, reactive power and power factor. The results and discussion are presented below. Grid Connected Small Wind Turbines A small wind turbine system consists of a rotor, generator, rectifier and inverter. After an inverter, system is connected to an AC panel; the local load is also connected to an AC panel. Before connection to the grid, typically, power goes through a transformer


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

to change the voltage level. For our experiment, the system was arranged in a similar way. To convert the wind energy into mechanical energy, a wind rotor is used. Most of the commercially available small wind turbines use furling, flapping, passive pitching and soft stall for their over speed and power control capacity [4]. For electric power generation, various types of generators have been used. Amon g them, a permanent magnet generator is widely used. As the wind speed and wind direction change every second, so does the total extracted power also change every second. As a result, the produced electric power from the generator is not uniform. The AC power from the generator is first converted to DC power with the help of a rectifier. Then, the DC power is again converted to the desired AC power so that the output can be connected with grid. The inverter plays a vital role in this system. The inverters must produce good quality sine-wave output and must follow the frequency and voltage of the grid. The inverter must observe the phase of the grid, and the inverter output must be controlled voltage and frequency variations [5]. Most of the commercially available grid tie inverters have an active power factor controller to reduce the Total Harmonic Distortion (THD).

Rated wind speed: 12.5 m/s Power Regulation: Non‐stop output control Maximum output power: 4KW Permanent magnets synchronous) Over Speed Regulation

generator

(3

phase,

Control/Protection:

Stall

Inverter output Voltage: 120V/208V Voltage Tolerance: ± 5% Grid Frequency: 60Hz Frequency Tolerance: ± 0.00083% Recommended font sizes are shown in Table 1. 1)

Power performance without any local load:

Experiments and Results

The turbine is grid connected. If any load is connected between the grid and turbine, in this article, it will be called local load. For comparison with local load and non-local load, the power performance data of the turbine with non-local load was collected for ten days. Then, 1 minute average values were calculated and normalized. Finally, by using the bin method, the power curve, reactive power curve and power factor curve were plotted.

To investigate the power performance of small wind turbines, two different turbines were used. For our experiment, we can say that they are turbine A and turbine B.

Fig. 1(a) shows the power curve of the wind turbine. In Fig 1(b), the power factor is plotted and it indicates that, when wind speed is more than 9 m/s, the power factor is close to unity.

For data logging, a Campbell scientific 1000 data logger was used, collecting the data for every second and storing it in a pc. This creates a file for the whole day (24 hours). Raw data (second data) is later converted to 1 minute average data. Then, one minute data was normalized and the power curve was plotted using the bin method [6]. The bins method was also used to compare the reactive power and power factor.

Power factor =active power/apparent power Apparent power 2= active power 2+ reactive power 2 Reactive power is plotted in Fig 1(c) and it can be observed that reactive power is almost linear and it is less 200 var.

For the experiment with a motor, another Hioki meter was used. It can measure three phases - active power, power factor and reactive power - every second and store the data in a memory card. Later, collected data can easily be moved from the memory card to a PC. Experiments with turbine A Turbine ‘A’ is a 1.1 KW small wind turbine. It has a PMG. It uses stall regulation for its control system. The specifications of this turbine are as follows: Rated power: 1.1 KW

(a)

283


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

FIG. 2 BLOCK DIAGRAM FOR AN EXPERIMENTAL SETUP WITH RES ISTIVE LOAD (HEATER)

(b)

(c)

Heaters were connected for 6 days and 1 second data was collected. Data was converted to 1 minute average data. Using the bin method, the power performance curves were plotted and these are shown in Fig. 3.

(a)

FIG. 1 POWER PERFORMANCE OF TURBINE A WITHOUT ANY LOCAL LOAD A) POWER CURVE, B) POWER FACTOR, C) REACTIVE POWER

Active power output depends on wind speed and it increases as the wind speed increases. But with wind speed increase, there is no significant change in reactive power compared to active power change. So, at high wind speed, apparent power is almost equal to active power and thus the power factor improves at high wind speed. 2)

Experiment with Heaters

To experiment on whether resistive load can have an effect on the power performance of wind turbines, three heaters with phase b and/or phase c were connected, as shown below in Fig. 2. It is noted that the turbine is connected to phase b and phase c of the transformer. The heaters were adjusted to maximum point. One of the heaters was a turbine shed heater. It was configured for 208 volt and it was connected to phase b and phase c. The other heaters were portable. They were configured for 120V and they were connected to phase b-neutral and phase c-neutral. They were kept outside the shed.

284

(b)

(c) FIG. 3 POWER PERFORMANCE COMPARISON OF TURBINE A WITH RES ISTIVE LOAD A) POWER CURVE, B) POWER FACTOR, C) REACTIVE POWER


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

From Fig. 3(a ), it can be seen that there is no change in power curve. In Fig. 3(b), there is no change in power factor also. In both cases, when wind speed is greater than 9m/s power factor was very close to unity. Fig. 3(c) shows that there is no significant change in reactive power. In both cases, reactive power decreases gradually and it is always less than 200 var. Therefore, we can say that there is no significant effect of heaters on the output power, power factor and reactive power. 3)

Experiment with a 5hp induction motor:

For the experiment with inductive load, a 5hp induction motor was used as the load. The motor was connected before the transformer. The motor was in no load mode, so its active power consumption was low but reactive power was high. The motor was a three phase motor. The motor was left for three days and 1 second data was collected for those days. After that, a set of compensation capacitors was connected for two more days and 1 second data was collected again. Data was analysed using the procedure mentioned earlier. In this case, another meter (Hioki m eter) was connected before the transformer. This meter measured the whole building power performance for every second. Here, the whole building includes the wind turbine, inductive motor, data logger, and building lights. Data was analysed using the bin method. The connection diagram is below in Fig. 4 and data is plotted in Fig. 5.

FIG. 5 (A) POWER FACTOR COMPARISON OF THE WHOLE BUILDING WITH INDUCTIVE AND CAPACITIVE LOAD

FIG. 5 (B) REACTIVE POWER COMPARIS ON OF THE WHOLE BUILDING WITH INDUCTIVE AND CAPACITIVE LOAD

Experimental result & comparison in the turbine transducer:

(a) FIG. 4 BLOCK DIAGRAM FOR EXPERIMENTAL SETUP WITH AN INDUCTIVE LOAD (MOTOR) AND CAPACITOR

Experimental result & comparison in the Hioki meter: From Fig .5(b), we can find that, when the inductive load was connected, the whole building power factor was decreased and when compensation capacitors were connected, the power factor improved. Fig. 5(c) shows that the total demand of reactive power for the whole building was more than 1600var, but with compensation capacitor it was similar to reactive power as without a motor.

(b)

285


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

curve and it is not consistent. Sometimes, it goes into stall regulation too early, say at 8m/s or 10 m/s and remains stalled above that wind speed; it may be a synchronizing problem with grid. At that time, it does not produce any power. The following three power curves are for the same conditions (i.e. without any local load, turbine is connected to the grid directly).

(c) FIG. 6 POWER PERFORMANCE COMPARISON OF TURBINE A WITH INDUCTIVE AND CAPACITIVE LOAD A) POWER CURVE, B) POWER FACTOR, C) REACTIVE POWER

From Fig. 6(a) it can be observed that the power curve for all the three cases i.e. without motor, with motor and with motor & capacitor are about same. There is no effect of the induction motor and capacitor. Next, two curves indicate that the power factor is almost similar and reactive power is still less than 200var. So, the inductive load and compensation capacitor have no effect on the power performance of this small grid connected wind turbine. Experiments with Turbine B Turbine B is a 1.3 KW small wind turbine. It also has a PMG and it uses the stall regulation method for its control system. The specifications of this turbine are given here: Rated power: 1.3 KW Rated wind speed: 12 m/s

FIG. 7 POWER CURVE VARIATION FOR TURBINE B UNDER THE SAME CONDITIONS

Therefore, it is very difficult to compare power curves with and without local load. For our experimental comparison, 1 second data for 14 days without any local load was collected and plotted following the same procedure as described above for wind turbine A. Th e power curve and reactive power against wind speed was plotted. As the active power varies for the same wind speed, here we have plotted the power factor against the active power instead of power factor versus wind speed. Fig. 8 shows that the power factor is very constant. When output power is more than 200W, the power factor becomes unity. Reactive power is always less than 150 var.

Power Regulation: Active – Inverter Maximum continuous output power: 1.4KW Utility interconnection voltage and frequency trip limits and trip times: Programmable, Utility specific Total Harmonic Distortion (current): < 3% Trip limit and trip time accuracy:<10% Grid Voltage: Single phase, 208V

(a)

Tolerance: ± 5% Grid Frequency: 60Hz Tolerance: ± 0.00083% 1)

Power performance without any local load:

Fig. 7 below shows the turbine power curve generated from the data collected over a number of days. It indicates that wind turbine B has a nonlinear power

286

(b)


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

(c)

(a)

FIG. 8 POWER PERFORMANCE OF TURBINE B WITHOUT ANY LOCAL LOAD A) POWER CURVE, B) POWER FACTOR, C) REACTIVE POWER

2)

Experiment with Heaters

(b)

FIG. 9 BLOCK DIAGRAM FOR EXPERIMENT SETUP WITH RES IS TIVE LOAD (HEATER) FOR TURBINE B

Experiments were repeated similar to wind turbine A for resistive load. Turbine B was connected to phase A and phase B in the AC panel, so all the heaters were connected to phase A and/or Phase B, as shown in Fig. 9 and data was collected and analysed. Recorded data is plotted in Fig. 10. From Fig. 10(a) we can see that there is a little change in the power curve. As mentioned earlier, the power curve is not constant for this wind turbine, so it is better to look at the power factor and reactive power. The power factor is also unchanged. From Fig. 10(b), for both cases, the power factor was close to one when active power was greater than 200 watts. Fig. 10(c) also shows that there is no significant change in reactive power. In both cases, reactive power increases gradually and it was always less than 150 var. Therefore, it could be concluded that there is no significant effect of resistive load on the power performance of turbine B. 3)

Experiment capacitors:

(c) FIG. 10 POWER PERFORMANCE COMPARISON OF TURBINE B WITH RES ISTIVE LOAD A) POWER CURVE, B) POWER FACTOR, C) REACTIVE POWER

mentioned earlier, a Hioki meter was used before the transformer to measure the whole building power performance. Collected data was analysed and it is plotted in Fig. 12.

with 5hp induction motor and

Th e experim ent was r epeated for t urbin e B w ith inductive and capacitive load. Fig. 11 below shows connection diagram for this set of experiments. As

FIG. 11 BLOCK DIAGRAM FOR THE EXPERIMENTAL SETUP WITH AN INDUCTIVE LOAD (MOTOR) AND CAPACITOR FOR TURBINE B

287


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

Experimental result & comparison in the Hioki meter:

can be seen that there is no significant effect of the inductive load and compensation capacitor on the turbine power factor and reactive power.

FIG. 12 (A) POWER FACTOR COMPARISON OF THE WHOLE BUILDING WITH INDUCTIVE AND CAPACITIVE LOAD CONNECTED TO TURBINE B

(a)

FIG. 12 (B) REACTIVE POWER COMPARIS ON OF THE WHOLE BUILDING WITH INDUCTIVE AND CAPACITIVE LOAD CONNECTED TO TURBINE B

(b)

In Fig. 12(a), we can see that, when active power is greater than 200W in both directions (from grid or to grid), the power factor is very close to unity for without motor. When we connected the motor power factor decreased, but when compensation capacitors were added, the power factor improved and came back to unity. From Fig. 12(b), without the motor reactive power is more or less 200 var. But when the motor was connected, the whole building demand for reactive power increased and it was greater than 1600 var. After that, capacitors were connected and the reactive power demand for the whole building was decreased. More test results are presented in Fig. 14. Turbine transducer data was collected and analysed as described previously for turbine A. Experimental result & comparison in the turbine transducer: From the plots in Fig. 14(a), we can see that the power curve varied a little bit. As this turbine power curve is not a constant, it is better to compare the power factor and reactive power. From Fig. 14(b) and Fig. 14(c), it

288

(c) FIG. 14 POWER PERFORMANCE COMPARISON OF TURBINE B WITH INDUCTIVE AND CAPACITIVE LOAD A) POWER CURVE, B) POWER FACTOR, C) REACTIVE POWER

Conclusions This paper described the power performance of two small wind turbines under variable load conditions. The active power performance, the power factor condition and the reactive power performance under resistive load, inductive load and compensation capacitor load have been presented in this paper. From the data, it is concluded that there is no significant


International Journal of Energy Science Vol. 2 Iss. 6, December 2012

effect of load type on the power performance of a small grid connected turbine. When experiments were done with a 5hp inductive motor, which consumes about 1600var, it was found that the turbine also had no effect on its reactive power production; thus, the power factor remained unchanged. Th e motor consumed the reactive power from the grid. Therefore, these experimental results conclude that the small wind turbine cannot produce any reactive power for induction or capacitive load. The wind turbine inverter basically acts as a current source and its current phase angle is very close to the grid voltage phase angle. Most of the small wind turbines are situated in remote locations isolated from the national power plant. If somebody buys a small wind turbine to run a small motor for his/her business and connects the turbine to the grid, then the system will still consume reactive power from the grid and there will be loss or utility penalty for the reactive power. A control system should be developed so the owner can adjust the reactive power and power factor for an optimal operation and minimal reactive power from the grid.

REFERENCES [1]

M.Koulouvari,

P.

Ladakakos,

D.

Foussis,

E.

Morfiadakis, “Measure ment Systems De dicate d to wind

turbine

powe r

quality

applications”

8th

inte rnational confe rence on Harmonics and quality of powe r, 1998. [2]

Carlos E. A. Silva Re ne T. Bascope , De mercil S. Olive ira Jr, “Three -phase

powe r factor correction rectifier

applie d to wind e ne rgy conve rsion syste ms” APEC 2008, Twe nty –Third Annual IEEE, Page (s): 768-773. [3]

Pe iyuan Chae n, Pie rluigi Siano, Birgitte Bak-Jensen, Zhe che n, “Stochastic optimization of wind turbine powe r factor using stochastic mode l of wind powe r” , IEEE transactions on Sustainable Energy ,April 2010.

[4]

Md. Arifujjaman, M.Tariq Iqbal, John E. Quaicoe, M. Jahangir Khan, “Mode ling and control of a small wind turbine ” Canadian Confere nce on Ele ctrical and computer Engineering, 2005.

[5]

Varin

Vongmancee,

Vee rapol

Monyakul,

“Grid

connecte d inverte r with unity powe r factor for wind ACKNOWLEDGEMENTS

powe r applications” industrial

The authors would like to thank the National Science and En gineering Research Council (NSERC), Canada for providing financial support for this research. The authors also would like to thank the Wind Energy Institute of Canada (WEICan), for giving all kinds of technical support for the experiments.

Applications, 2009. [6]

Ele ctronics

&

A. Llombart, S. J. Watson, D. Llombart and J. M. Fandos, “Powe r Curve Characte rization I: improving the bin me thod” Inte rnational confe re nce on re ne wable e nergies and powe r quality, 2005.

289



International Journal of Energy Science