Pervasive Sensor Environment A dissertation submitted in partial fulfillment of the requirements for the degree of Master of Technology by Rahul Kumar Mishra (Roll No. 09307902)
Under the guidance of Prof. S. N. Merchant
DEPARTMENT OF ELECTRICAL ENGINEERING INDIAN INSTITUTE OF TECHNOLOGY BOMBAY June 27, 2012
To my Creator who has given me the senses to sense and act responsibily.
Dissertation Approval The dissertation entitled
Pervasive Sensor Environment by
Rahul Kumar Mishra (Roll No. 09307902) is approved for the degree of Master of Technology
Examiner
Examiner
Guide
Chairman
Date: June 27, 2012 Place: IIT Bombay
Declaration of Academic Ethics I declare that this written submission represents my ideas in my own words and where others’ ideas or words have been included, I have adequately cited and referenced the original sources. I declare that I have properly and accurately acknowledged all sources used in the production of this thesis. I also declare that I have adhered to all principles of academic honesty and integrity and have not misrepresented or fabricated or falsified any idea/data/fact/source in my submission. I understand that any violation of the above will be a cause for disciplinary action by the Institute and can also evoke penal action from the sources which have thus not been properly cited or from whom proper permission has not been taken when needed.
Date: ............
Rahul Kumar Mishra (Roll No. 09307902)
Acknowledgements I would like to express my deepest gratitude to Prof. S. N. Merchant, Prof U. B. Desai and Prof. P Rajalakshmi, for their persistent guidance, constant support, inspiration, encouragement and stimulating suggestions during this work. Apart from the technical inputs I am indebted to them for motivation and encouragement provided to make this project a learning experience. I am thankful to Mr. Mirza Sami, student IIT Hyderabad for the technical help provided during the initial phase of this project at IIT Hyderabad. I am indebted to my project mates Mr. Biprangshu Saha, Mr. Praveen Tamhankar, Mr. Prashant Verma, and Mr. Mehul Shah for their important contribution in data collection, network deployment and other phases of the project. I am also grateful to Mrs. Shasikala Nair and Mr. Chandrakant for taking care of the lab and other office work so that we could devote our time to project. Mr. Prasad Lavande and Nimish Shah, Student K.J.Somaiya College of Engg helped with their contributions in dynamic network deployment and assessing the limitations of hardware. Prof. R.S. Patil, CESE was always more than helpful in introducing us to the notions of environment and pollution monitoring. My sincere thanks to Prof. Steve Hailes, UCL for the timely tips related to intelligence in WSN. I would also like to thank all SPANN lab members and all the people in IIT Bombay for many things that I am grateful for. I am thankful to IUATC for providing the funding for this project. Special thanks to my parents and family for providing me support and positive energy to keep me going. Finally, my obeisances to God who has been taking care of me always.
Date:
Rahul Kumar Mishra
Abstract Pervasive Sensor Environment is an IU-ATC funded project that aims to support a diverse range of sensor-network based services across a scalable infrastructure.The application area identified is environmental monitoring. This has intentionally been selected as being of significant social and economic interest in both the UK and India. Within this context, we have undertaken air pollution level measurement and building a context-aware framework for various applications like traffic monitoring and actuators based pollution control systems.
Air pollution has significant influence on the concentration of constituents in the atmosphere leading to effects like global warming, acid rains and more. To avoid such adverse imbalances in the nature, an air pollution monitoring system is utmost important. This project attempts to develop an effective system for pollution monitoring using wireless sensor networks (WSN) on a real time basis and adding intelligence to make a context-aware framework. Commercially available sensors are calibrated and used for sensing concentration of various environmental parameters like temperature, atmospheric pressure, humidity, CO2 , N Ox , CO, SO2 and O2 . These pre-calibrated gas sensors are then integrated with the wireless sensor motes for field deployment using multi-hop data aggregation algorithm. A light weight middleware and a web interface to view the live pollution data in the form of numbers and charts from the test beds was developed and made available from anywhere on the internet. Data mining algorithms and machine learning have been used to find insights into the data for decision making and adding intelligence needed for building context-aware framework. Experimentation carried out using the developed wireless air pollution monitoring system under different physical conditions show that the system collects reliable source of real time fine-grain pollution data.
iv
Contents Abstract
iv
List of Figures
viii
List of Tables
x
List of Abbreviations
xi
1
Introduction
1
1.1
Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
1.2
Objective of the work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
3
1.3
Related work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
4
1.4
Organization of dissertation . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
2
Pollution Monitoring
6
2.1
Air Pollution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6
2.1.1
Major Air Pollutants . . . . . . . . . . . . . . . . . . . . . . . . . . .
8
WSN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9
2.2
2.2.1 2.3 3
Key Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
11
WSN being Pervasive . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13
System Design
16
3.1
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16
3.2
Calibration and Conditioning of sensors . . . . . . . . . . . . . . . . . . . . .
17
3.3
Configuring Wireless Sensor Nodes . . . . . . . . . . . . . . . . . . . . . . .
18
3.4
Lightweight middleware and web interface . . . . . . . . . . . . . . . . . . . .
19
v
4
Real time monitoring
20
4.1
Static Multi-hop WSN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
20
4.2
Dynamic Multi-hop WSN . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
23
4.2.1
AODV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
23
4.2.2
Routing Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
4.2.3
Advantages and Disadvantages . . . . . . . . . . . . . . . . . . . . .
25
Individual motes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
26
4.3 5
Data Observation & AQI
29
5.1
Observations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
5.1.1
Air Contaminant . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
29
5.1.2
Carbon Dioxide – CO2 . . . . . . . . . . . . . . . . . . . . . . . . . .
30
5.1.3
Nitrogen Dioxide – N O2 . . . . . . . . . . . . . . . . . . . . . . . . .
31
5.1.4
Oxygen – O2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
32
5.1.5
Relative Humidity . . . . . . . . . . . . . . . . . . . . . . . . . . . .
33
5.1.6
Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
34
AQI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
35
5.2.1
Calculating AQI . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
37
5.2.2
Air Quality Index Trends . . . . . . . . . . . . . . . . . . . . . . . . .
37
5.2
6
Data Analysis and Context-aware PSE
40
6.1
Data Warehouse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
40
6.1.1
Database Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
40
Descriptive models using data mining . . . . . . . . . . . . . . . . . . . . . .
43
6.2.1
Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
43
6.2.2
Decision Trees . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
44
6.2.3
Association Rules
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
45
Context-awareness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
6.3.1
Elements of CAPS . . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
6.3.2
Service framework . . . . . . . . . . . . . . . . . . . . . . . . . . . .
48
6.3.3
Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
49
6.2
6.3
vi
7
Challenges, Conclusion and Future scope
51
7.1
Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
51
7.2
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
52
7.3
Future Scope . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
53
vii
List of Figures 1.1
Conceptual view of the system . . . . . . . . . . . . . . . . . . . . . . . . . .
2
1.2
Pollution monitoring stations in India . . . . . . . . . . . . . . . . . . . . . .
5
2.1
Generalized Process of Atmospheric Pollution . . . . . . . . . . . . . . . . . .
8
3.1
System overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
17
3.2
Various stages in calibrating gas sensors . . . . . . . . . . . . . . . . . . . . .
18
3.3
Waspmote Board . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
18
3.4
Waspmote Gas Sensor Board . . . . . . . . . . . . . . . . . . . . . . . . . . .
19
3.5
Multihop mesh network system architecture . . . . . . . . . . . . . . . . . . .
19
4.1
Multi-hop architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
21
4.2
Static multi-hop implementation overview . . . . . . . . . . . . . . . . . . . .
22
4.3
Deployment in Kukatpally a suburb in Hyderabad . . . . . . . . . . . . . . . .
22
4.4
Deployment in IIT Hyderabad campus . . . . . . . . . . . . . . . . . . . . . .
22
4.5
AODV implementation Sample terminal output on gateway . . . . . . . . . . .
24
4.6
AODV implementation Sample terminal output illustrating packets and process
26
4.7
Waspmotes deployed at Y-Point gate IIT Bombay . . . . . . . . . . . . . . . .
27
4.8
Mote Placement at IIT Bombay Gate . . . . . . . . . . . . . . . . . . . . . . .
27
4.9
Output of Programmed Mote . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
4.10 Sample Sensor Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
5.1
Air Contaminant Trends: (A)24-hour (daily) (B)7-day (weekly) (C)17-day (fortnightly) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
30
5.2
CO2 Trends: (A)24-hour (daily) (B)7-day (weekly) (C)17-day (fortnightly) . .
31
5.3
N O2 Trends: (A)24-hour (daily) (B)7-day (weekly) (C)17-day (fortnightly) . .
32
viii
5.4
O2 Trends: (A)24-hour (daily) (B)7-day (weekly) (C)17-day (fortnightly) . . .
5.5
Relative Humidity Trends: (A)24-hour (daily) (B)7-day (weekly) (C)17-day (fortnightly) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5.6
33
34
Temperature Trends: (A)24-hour (daily) (B)7-day (weekly) (C)17-day (fortnightly) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
35
5.7
Air Quality Index (US Standards) . . . . . . . . . . . . . . . . . . . . . . . .
36
5.8
AQI Sub-index and breakpoint for India as proposed by IITM (MoES) . . . . .
36
5.9
Calculation of AQI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
38
5.10 Air Quality Index Trends: (A)24-hour (daily) (B)7-day (weekly) (C)17-day (fortnightly) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
38
5.11 AQI Vs Hours of the day . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
39
5.12 AQI Vs Days . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
39
6.1
Data warehouse ETL and staging . . . . . . . . . . . . . . . . . . . . . . . . .
41
6.2
Clustering using Microsoft Clustering algorithm . . . . . . . . . . . . . . . . .
43
6.3
Decision tree using Microsoft Decision tree algorithm . . . . . . . . . . . . . .
45
6.4
Dependency network of decision tree . . . . . . . . . . . . . . . . . . . . . . .
46
6.5
Association rules using Microsoft ARM algorithm . . . . . . . . . . . . . . . .
47
6.6
Abstract layered architecture of CAPS . . . . . . . . . . . . . . . . . . . . . .
48
6.7
Air pollution parameters of node#1 . . . . . . . . . . . . . . . . . . . . . . . .
50
7.1
Hardware cost for fine-grained air pollution monitoring in Mumbai, India . . .
52
ix
List of Tables 6.1
Database table maindata as DWH table . . . . . . . . . . . . . . . . . . . . .
41
6.2
Database table Mote information at individual base stations . . . . . . . . . .
42
7.1
Software Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
52
x
List of Abbreviations WSN
Wireless Sensor Network
AODV
Adhoc On-demand Distance Vector
SSAS
SQL Server Analysis Services
SSIS
SQL Server Integration Services
PSE
Pervasive Sensor Environment
OLAP
On-Line Analytical Processing
ETL
Extraction Transformation and Loading
ARM
Association Rule Mining
DWH
Data Warehouse
QoS
Quality of Service
CAPS
Context-aware Pervasive System
AJAX
Asynchronous JavaScript and XML
JSP
JavaServer Pages
API
Application Programming Interface
xi
Chapter 1 Introduction Since the 1970s, there has been an exponential increase of industries and automobiles. These industries and automobiles have caused complex and serious problems like stratospheric ozone depletion, acid rain, climate change, and increased occurences of diseases like lung cancer, asthma and chronic pulmonary diseases. In addition to industries and automobiles, agricultural activities, and even ordinary homes have contributed towards this environmental pollution. Stationary and mobile sources release various chemical pollutants, including suspended particulate matter (SPM), carbon monoxide (CO), oxides of nitrogen (N Ox ), oxides of sulfur (SOx ), and other toxics. Therefore, its become increasingly important to have real-time air pollution monitoring system along with a framework that aids in various stages of decision making. Wireless Sensor Network(WSN) is a fast evolving technology and an active research area mainly due to the number of potential applications. In this thesis, we expound the system which monitors environmental parameters by use of dynamic WSN for air pollution monitoring in Mumbai and Hyderabad in three-part framework. First part is the pre-deployment phase which includes sensor calibration, second part is the deployment phase and data collection, and finally the third phase is the data warehouse on the remote PC. We propose an intelligent dynamic Wireless Sensor Network Air Pollution Monitoring System to monitor air pollution through the use of WSN. Ad hoc On-Demand Distance Vector (AODV) routing is implemented over this zigbee WSN. AODV has been extensively chosen in research community because of its quick adaptation to dynamic link condition, low processing and memory overhead and low network utilization. The proposed system makes use of an ambient Air Quality Index (AQI) for the indian context. AQI is also used as prediction parameter in data mining algorithms to find data insights and environment context. The various data mining algorithms used for gaining insights 1
and pattern recognition are decision trees, unsupervised and semi-supervised clustering, and association rule mining. The data warehouse makes use of Microsoft SQL Server 2005 Database, Online analytical processing (OLAP), SQL Server 2005 Analysis Services(SSAS) and Integration Services(SSIS). This thesis is application-centric and describes the system entirely in the practical context. The figure 1.1 describes the conceptual view of the system. Firstly, hardware motes are configured along with calibrated sensors mounted on it. Here, we have used Libelium waspmotes. Then a multi-hop network for sensing environment parameters is deployed on the field. The collected data is then sent to the base station and then to a central server. The data is visible on the internet through a web interface. This web interface also provides analysis of data as numbers, figures and charts for getting key insights into data which is further used for decision making using the context provided by the analysis engine.
Figure 1.1: Conceptual view of the system The objective of this work is to come up with cost effective, reliable, scalable and accurate real-time air pollution monitoring system with wireless sensor networks. Commercially available electrochemical and resistive heating type sensors were used to sense the gases like O2 , CO2 , CO and N O2 . Appropriate calibration technologies were developed to calibrate these sensors, which are then interfaced to wireless sensor motes. Zigbee based wireless sensor networks with multihop data transfer algorithm were implemented to [31] extend the range of monitoring area. The calibration technology for the gas sensor is out of scope of this thesis. System architecture of real time wireless pollution monitoring system, development and deployment of the system, a context framework by adding intelligence are discussed in the remaining chapters of this thesis. 2
1.1
Motivation
Air pollution emerged in many parts of the world as a result of explosive industrial growth. Road transport is also one of the major contributors of air pollution, which contribute to climate change that has perilous domestic and global consequences [27]. Generation and transport of pollutant materials are governed not only by the distributions of their sources but also by the dynamics of the atmosphere. Pollutant clouds are sometimes observed traveling along the wind directions [20]. To understand the involved processes in more detail, we need more thorough data on the spreads of fine-grain pollutants and their variations with time. Hence, there is a growing demand for the environmental pollution monitoring and control systems. Pollution monitoring and display to the citizens is essential to compare the impact of measures taken by municipalities and public institutions and raise public awareness. For example, monitoring of pollution in Stockholm city center made its citizens to approve in a referendum approve a congestion tax for accessing to downtown. The results were a 22% reduction in CO2 emissions and a 18% reduction in the average time of jams. Other cities such as London, Brisbane and Singapore have adopted similar measures.Hence, to understand the nature of pollutants and taking appropriate action timely is of utmost importance. An air pollution monitoring system that is comprehensive in terms of spatial, temporal and pollutant coverage, and is relatively inexpensive and autonomous is the priority.
1.2
Objective of the work
• WSN deployment to measure air pollution parameters. • Fine-grained pollution data collection for analysis • Analysis of the test-bed and defining notions of context • Building a context-aware framework • a pervasive air-quality sensors network and prediction models
3
1.3
Related work
Some of the existing instruments for air pollution monitoring are Fourier transform infrared (FTIR) instruments, gas chromatographs, and mass spectrometers. These instruments provide fairly accurate and selective gas readings. However high cost, large size & maintenance cost made them unfavorable for monitoring applications on large scale . A gas sensor that is compact, robust, with versatile applications, and low cost could be an equally effective alternative [23]. Some of the gases monitoring technologies are electrochemical, infrared, catalytic bead, photo ionization, and solid-state . The existing monitoring system largely uses smart transducer interface module (STIM) with semiconductor gas sensors, which uses the 1451.2 standard. STIM was found to be an efficient monitoring system but for the power requirements and ability to expand for large deployment. One of the large scale sensor networks for monitoring and forecasting is Environment Observation and Forecasting System (EOFS) [39], but the size of the system and initial cost is too high. Air pollution monitoring system based on geosensor network with control action and adaptive sampling rates proposed in [22] also cannot be vast deployment due to high cost. RESCATAME [9] is a project funded by the European Union through its LIFE program. Its main goal is to achieve sustainable management of the traffic in the city of Salamanca, Spain by using two key-elements: a pervasive air-quality sensors network as well as prediction models. In India , a project on Pollution Monitoring and Evaluation system using Sensor based Wireless Mesh Network for the protection of Public spaces" was initiated at IIM Kolkata under the umbrella of Department of Science and Technology. Indian Institute of Tropical Meteorology (IITM), Pune, a constituent under the Ministry of Earth Sciences, Government of India, is spearheading country’s first major initiative named as “System of Air Quality forecasting and Research (SAFAR)� [18]. Figure 1.2 shows the pollution monitoring stations in India.
4
Figure 1.2: Pollution monitoring stations in India
1.4
Organization of dissertation
In Chapter 2, air pollution notions are provided along with an overview of WSN. Also, pervasiveness has been emphasized in context with WSN. System design has been expounded in stages in Chapter 3. Network deployment has been dealt in detail in Chapter 4 covering the aspects of static and dynamic WSNs Data collected using the deployed WSN is analyzed in Chapter 5 using plots. Chapter 5 also introduces the notion of AQI which is very crucial for data mining In Chapter 6, a thorough data analysis is provided using data mining algorithms. Data warehouse and context-awareness are also expounded here. Finally, the thesis is concluded by mentioning the challenges met, future scope and the critical summary.
5
Chapter 2 Pollution Monitoring 2.1
Air Pollution
Air pollution is one of the major growing problems all over the world. Many different sources such as factories, power plants, automobiles and even from natural causes such as windblown dust, smoke from bush fires and volcanic eruptions are responsible for the pollution. The air quality can get affected in many ways due to the pollutants emitted from these sources. The deterioration of air quality thus results into a corresponding increase in health problems, eventually inducing the monitoring of air quality as a prime necessity in day to day life. There is a growing awareness of the linkages between human health, the weather and climate. Timely air quality information can assist the public in coping with health problems caused by groundlevel ozone, sulphur dioxide, nitrous oxide, particulate matter and other pollutants. Air quality advisories or alerts issued when predetermined pollutant thresholds exceeds should result in actions to reduce pollution levels and encourage people to avoid polluted areas thereby alleviating adverse effects on health. Though it is widely known that high concentrations of extended periods of prolonged exposure can have a highly detrimental effect on health [4, 17], it is not common knowledge that even short term exposure to relatively low-level of air pollutants can have serious health effects [40]. In the short-term, high pollution episodes can trigger increased admissions to hospital and contribute to the premature death of those people that are more vulnerable to daily changes in levels of air pollutants. High concentrations or extended periods of exposure can have serious health effects. Air pollution also has negative impacts on our environment, both in terms of direct effects of pollutants on vegetation, and indirectly through effects on the acid and nutrient status of soils and waters. 6
Nowadays, the highest percentage of air pollution comes directly from road traffic and not anymore from large industries, currently placed outside metropolitan & urban areas. Road traffic is considered to be responsible for 25% of all emissions in Europe. Loss of environmental quality is one of the biggest threats of our century to health and human well-being, together with environmental impacts. Research advises that present increase of respiratory and other related diseases is due to air pollution, as well as the increase of allergies that diminish in so many aspects people’s quality of life. According to European Union official data, 225,000 people died in Europe of diseases related with emissions from cars. To overcome this threat, the European Union legislation has become stricter and intends to reduce car emissions by 20% until 2020. As the European Environment Agency (EEA) says: "Ozone and PM are the most problematic pollutants for health, potentially causing or aggravating cardiovascular and lung diseases and leading to premature death. Eutrophication, an oversupply of nutrient nitrogen in terrestrial and aquatic ecosystems is another major problem caused by air pollutants. Nitrogen oxides (NOx) from combustion processes are now the main acidifying and eutrophying air pollutants, as sulphur pollution has fallen in recent years." Particulate air pollution is consistently and independently related to the most serious effects, including lung cancer and other cardiopulmonary mortality. Ambient air pollution, in terms of fine particulate air pollution (PM2.5), causes about 3% of mortality from cardiopulmonary disease, about 5% of mortality from cancer of the trachea, bronchus, and lung, and about 1% of mortality from acute respiratory infections in children under 5 yr, worldwide. This amounts to about 0.8 million (1.2%) premature deaths and 6.4 million (0.5%) years of life lost. This burden occurs predominantly in developing countries; 65% in Asia alone. These estimates consider only the impact of air pollution on mortality (i.e., years of life lost) and not morbidity (i.e., years lived with disability), due to limitations in the epidemiologic database. If air pollution multiplies both incidence and mortality to the same extent (i.e., the same relative risk), then the DALYs for cardiopulmonary disease increase by 20% worldwide [8]. A Even at low levels of pollution, health effects are heavily amplified for people with respiratory problems, weak immune systems and/or cardiovascular problems [40]. Air pollution can generally be identified by objective indicators such as: • reduction in visibility • presence of smog • acidic rain 7
• temperature increase (greenhouse effect) • unseasonal effects
Figure 2.1: Generalized Process of Atmospheric Pollution
2.1.1
Major Air Pollutants
The EPA (Environmental Protection Agency) Office of Air Quality Planning and Standards (OAQPS) has set National Ambient Air Quality Standards for the US. There are six major criteria pollutants [12] defined by EPA for which ambient air standards have been set to protect human health and welfare. • Sulfur dioxide (SO2 ): This primary pollutant is a colorless gas with a strong odor. At relatively high concentrations SO2 causes severe respiratory problems. Sulfur dioxide is an acid precursor, which is a source of acid rain when SO2 combines with water droplets to form sulfuric acid, H2SO4. Sulfur dioxide is also a precursor of sulfate particulates which affect the radiation balance of the atmosphere. • Nitrogen oxides (NOx): It is an indeterminate mixture of nitric oxide, NO, and nitrogen dioxide, N O2 , which is formed mainly from N2 and O2 during high-temperature combustion of fuel in vehicles. NOx causes the reddish-brown haze in city air, which contributes to heart and lung problems and may be carcinogenic. • Carbon monoxide (CO) and Carbon dioxide (CO2 ): Carbon monoxide is produced from incomplete combustion and is known to be major pollutant of an urban air. CO is highly poisonous to humans and most animals. Carbon dioxide is a complete oxidation product of fuel combustion. Also, in the atmosphere, CO oxidizes to CO2 which is a key greenhouse gas. 8
• Lead (Pb): Found as impurities in fuels. There are some other metals that are included in the pollutant group of lead, such as, mercury, cadmium, chromium and nickel. Its anthropogenic sources are metal mining and processing facilities, motor vehicles, etc. Lead is highly toxic. • Particulate matters (aerosols) – PM10: PM10 are particles with diameter smaller than 10 µm and include dust, soot, smoke, sulfates, nitrates, asbestos, pesticides, bio-aerosols (e.g. pollen, spores, bacterial cells, and fragments of dead insects). They are very harmful to human respiratory system and contribute to urban haze causing visibility reduction. • Ozone (O3 ): Ozone is a gas which is not emitted into atmosphere as it is formed from the ozone precursors, VOCs (Volatile Organic Carbons), and nitrogen oxides (NOx). Ozone is a hazard at ground level — it is a major constituent of photochemical smog which has diverse effects on human health and ecology causing damage to trees and vegetable.
2.2
WSN
Wireless Sensor Networks (WSNs) have become one of the most interesting areas of research in the past few years primarily due to its large number of potential applications. WSN is an enabler technology, many believe that it can revolutionize ICT(Information and Communication Technologies), the way microprocessor revolutionized chip technology nearly 30 years ago. The proliferation in MicroElectro-Mechanical Systems (MEMS) technology has facilitated the development of smart sensors. These recent advances in Wireless Sensor Networks (WSNs) have also lead to rapid development of real time applications. In this article we take a tour on how WSNs have evolved over the last decade with emphasis on the applications. In 2003, Technology Review from MIT, listed WSN on the top, among 10 emerging technologies that would impact our future. The increasing interest in wireless sensor networks can be promptly understood simply by thinking about what they essentially are: a large number of self-powered small sensing nodes which gather information or detect special events and communicate in a wireless fashion, with the end goal of handing over their processed data to a base station. There are three main components in a WSN: the sensor, the processor, and the radio for wireless communication. Processor and Radio technology are reasonably mature. Nevertheless, cost is still a major consideration for
9
large scale deployment. If we look back, a lot of work have been done in the field of protocols, collaborative information processing, dedicated OS like Tiny OS, dedicated database systems like Tiny DB, programming languages like nesC, 802.15.4 standardization in form of Zigbee, and large number of test deployments. Also, major initiatives in WSN R&D have been taken by MNCs like Microsoft(Project Genome), Intel(WISP), IBM(IBM Zurich Sensor System lab and Testbeds), SUN Microsystems(SPOT), etc. Sensor networks provide endless opportunities, but at the same time pose formidable challenges, which include deployment, localization, self-organization, navigation and control, coverage, energy, maintenance, and data processing. The fact that energy is a scarce and usually non-renewable resource, power consumption is a central design consideration for wireless sensor networks whether they are powered using batteries or energy harvesters. However, recent advances in low power VLSI, embedded computing, communication hardware, and in general, the convergence of computing and communications, are making this emerging technology a reality in terms of processing, memory and energy. In general, WSNs are deployed using a nonrenewable, but there have been applications like pervasive sensor environment(described later) which uses renewable source of energy, here, solar power. In many current projects, applications are executing on the bare hardware without a separate operating system component. Hence, at this stage of WSN technology it is not clear on which basis future middleware for WSN can typically be built. Another key challenge in WSN is the middleware. Middleware as the name suggests, sits right in between the operating system and the application. The main purpose of middleware for sensor networks is to support the development, maintenance, deployment, and execution of sensing-based applications. Currently, programmers deal with too many low levels details regarding sensing and node-to-node communication and the programming abstractions provided by middleware become a key aspect in its pursuit. Middleware also plays a very crucial role in leading to pervasive sensor systems. Ad-hoc sensor networks although related to WSNs but are very different in terms of energy supply, number of sensor nodes, computational capabilities, memory and global identification. Wireless Sensor Networks are state of the art technologies that have a wide range of potential applications. Sensor network generally consists of a large number of distributed nodes that organize themselves into a multi-hop wireless network. Each node has one or more sensors, embedded processors and low-power radios, and is normally battery operated because of small size. Currently, monitoring is done through large and expensive devices, which are not always connected to their control center and are in a small
10
number for the coverage area. To connect ambient monitoring to a wireless network creates new possibilities. A variety of mechanical, thermal, biological, chemical, optical, and magnetic sensors may be attached to the sensor node to measure properties of the environment. Since the sensor nodes have limited memory and are typically deployed in difficult-to-access locations, a radio is implemented for wireless communication to transfer the data to a base station (e.g., a laptop, a personal handheld device, or an access point to a fixed infrastructure). Battery is the main power source in a sensor node. Secondary power supply that harvests power from the environment such as solar panels may be added to the node depending on the appropriateness of the environment where the sensor will be deployed. Depending on the application and the type of sensors used, actuators may be incorporated in the sensors. In a structured WSN, all or some of the sensor nodes are deployed in a pre-planned manner. The advantage of a structured network is that fewer nodes can be deployed with lower network maintenance and management cost. Fewer nodes can be deployed now since nodes are placed at specific locations to provide coverage while ad hoc deployment can have uncovered regions.
2.2.1
Key Issues
The design of the sensor network is influenced by many factors such as tolerance, scalability, energy, hardware constraints etc. • Scalability - A large number of sensor nodes in the range of hundreds may be deployed in a sensor network and this requires that the sensor network should be scalable. It should be easy to add additional nodes or remove damaged nodes without drastically altering the existing architecture of the network. • Fault Tolerance - Sensor networks in which large number of sensors are deployed need to have a good level of fault tolerance because some sensors or sensor nodes can fail and this should not affect the overall performance of the network. • Lifetime - Most of the scenarios in which the use of WSN is depicted relate to environmental habitat monitoring and security application, where the WSN is usually deployed out in field where they are supposed to lay for months unattended. The node has to be designed in such a manner as to maximize the lifetime of WSN by managing its local supply of energy in an efficient manner. Various methods (at the cost of degraded performance) can be employed to maintain the energy supply for a longer duration. The 11
most crucial factors affecting the lifetime of a WSN is the energy supply [15]. Each node has to be designed in such a manner as to maximize the lifetime of WSN by managing its local supply of energy in efficient manner. Various methods (at the cost of degraded performance) can be employed to maintain the energy supply for a longer duration. – One method could include usage of solar cells or piezoelectric generators to scavenge energy from the environment. – Most of the power in a pod is consumed by radio link, and this could be managed by lowering the transmission output power, though this could affect the other performance parameters, such as communication range. – Another method could be usage of optical methods instead of radio link for communication though this will pose its own limitations, one being the requirement of having a clear Line of Sight (LOS) between the pods, which would be quite unrealistic. • Energy Consumption - The consideration of energy needed to power the sensors plays a very important part in the overall design of the sensor network. Since the number of sensors could be quite large, ideally the energy required to power these sensors should be small enough as to keep the operating costs lower and at the same time enough power should be available to the sensor(s) to accomplish the assigned tasks. – The energy requirements for sensors differ, based on the type of the sensor and the task they are assigned to perform. – Another scenario that can effect the energy consumption of the sensors is if they are active or passive. A radar based sensor is an example of an active sensor that is continuously propagating radio waves .Hence will consume power continuously. An acoustic sensor on the other hand will sit idle listening for a sound, and in case of detecting a sound will become active for a short duration and pass it on the sink and then will go back to an idle mode. – A sensor that is continuously propagating radio waves, actively looking for an object and transmitting the data to the sink will of course consume more power than a sensor that is only active for a very short duration and only transmits to the sink when some event had occurred. 12
We need to use efficient routing algorithms as well as use spaced out readings in order to save power. Sleep and hibernate modes can also be used for power saving. • Coverage - Area of coverage provided by a WSN is another important factor since it increases system’s value to the end user. Various networking and data routing techniques can considerably increase the range of a network, for example multi-hop communication technique can extend the coverage of the network well beyond the range of radio technology alone [25]. Multi-hoping though has greater power consumption which could result in a shorter lifetime of the network and also a denser node network is needed for multi-hoping to work, which would mean additional costs.
2.3
WSN being Pervasive
WSNs are biggest proponents with regard to the third wave in computing. i.e.; Ubiquitous computing and its subsuming. First were mainframes, each shared by lots of people. Now we are in the fading end of personal computing era, person and machine staring uneasily at each other across the desktop. Ubiquitous computing has just entered, or the age of calm technology, when technology recedes into the background of our lives. The inherent nature of WSNs makes them deployable in a variety of circumstances. They have the potential to be everywhere, on roads, in our homes and offices(smart homes), forests, battlefields, disaster struck areas, and even underwaters. This very pervasive nature leads us to “everyware”phenomenon of ubiquitous computing. Today, we have entered the third wireless revolution, “Internet of Things”. The third wave is utilizing wireless sense and control technology to bridge the gap between the physical world of humans and the virtual world of electronics. The dream is to automatically monitor and predict or respond to forest fires, avalanches, land slides, earthquake, hurricanes, traffic, hospitals and much more over wide areas and with thousands of sensors. It has come in reaching grasp due to the development of Wireless Sensor Networks (WSN) more often called Ubiquitous Sensor Networks (USN). Pervasive computing refers to millions of computers embedded in the environment, allowing technology to recede into the background. Pervasive computing (also called ubiquitous computing) is the growing trend towards embedding microprocessors in everyday objects so they can communicate information. The words pervasive and ubiquitous mean “existing everywhere.”and are part of the bigger “Internet of Things”. Main features as described in [32] of Pervasive system are 13
• Invisibility - Invisibility is one of the major features of a pervasive system. Ideally there should be complete disappearance of pervasive computing technology from a user’s mind. In practice it implies is minimal user interaction. If a pervasive computing environment continuously meets user expectations and rarely presents him with surprises, it allows him to interact almost at a subconscious level [38]. • Scalability - As penetration of computing devices in the space grows, the intensity of interactions between a user’s personal computing space and his/her surroundings increases. This has severe band-width and energy implications for a wireless mobile user. The presence of multiple users will further complicate this problem. Scalability, in the broadest sense, is thus a critical problem in pervasive computing. Here, the density of interactions has to fall off as one moves away; otherwise, both the user and his computing system will be confused by distant interactions that are of little relevance. Although a mobile user far from home will still generate some distant interactions with sites relevant to him, the preponderance of his/her interactions will be local • Effective Use of Smart Spaces - A space may be an enclosed area such as a meeting room or corridor, or a well-defined open area such as a courtyard or quadrangle. By embedding computing infrastructure in building infrastructure, a smart space brings together two worlds that have been disjoint until now. The fusion of these worlds enables sensing and control of one world by the other. A simple example of this is the automatic adjustment of heating, cooling, and lighting levels in a room based on the surroundings. Influence in the other direction is also possible: software on a user’s computer may behave differently depending on where the user is currently located. Smartness may also extend to individual objects, whether located in a smart space or not. Main design issues in pervasive systems are as follows: • User Intent - The pervasive system should be able to track the user intent. It should be able to take decisions which actions will help the user. The main area of concern is that how to judge the user intent. Should it be provided specifically (eg. by a file) or should it be inferred. Even if the system can determine user intent, how to verify its correctness. • Energy Management - Energy has always been one of the major area of concern for such a system. Performing so many tasks like user intent determination, context aware 14
actions increases the energy demand of a software on mobile computer. At the same time, relentless pressure to make such computers lighter and more compact places severe restrictions on battery capacity. • Context Awareness - A pervasive computing system that strives to be minimally intrusive has to be context-aware. In other words, it must be aware of its user’s state and surroundings, and must modify its behavior based on this information. A user’s context can be quite rich, consisting of attributes such as physical location, physiological state (e.g., body temperature and heart rate), emotional state (e.g., angry, distraught, or calm), personal history, daily behavioral patterns, and so on. In making such decisions, system should not disturb user at inopportune moments except in an emergency. A key challenge is obtaining the information needed to function in a context-aware manner. In some cases, the desired information may already be part of a user’s personal computing space. The aim is to design a pervasive pollution monitoring system that is hidden from the users yet is continuously reading the data. Context awareness is one area of importance for building prototype. Energy management although crucial is beyond the scope of this thesis. This thesis present pervasive sensor environment from the point of view of context-aware notion.
15
Chapter 3 System Design 3.1
Overview
The design and development of the pollution monitoring system constitutes the following stage: • Calibration and Conditioning of sensors • Network deployment (middleware) – Programming motes – Programming gateway – Connecting gateway to central server • Field Deployment • Web-interface (Central Server) – Real-time display of sensed data as numbers and charts – Analysis of data – Publish analysis results In view of the above discussion, the figure 3.1 shows the system overview .
Air pollution monitoring is done using a network of Libelium Waspmotes. It is simple and cost effective for prototype development due to it is features of wireless communication via the protocol 802.15.4 / Zigbee. Each of the stages of the system design are described in the sections below. 16
Figure 3.1: System overview
3.2
Calibration and Conditioning of sensors
Each gas sensor is unique, i.e., though the type and the gas sensed are the same, different sensors may differ from each other in terms of the output characteristics. Therefore it becomes essential to calibrate [5] each and every sensor before interfacing to the wireless sensor mote for accurate readings. Calibration of gas sensors are carried out in laboratory by exposing the sensors to different concentrations of gas. The various stages involved in the calibration process are shown in Figure 3.2. A specially design closed chamber setup is arranged with varying temperature and humidity. Chamber has a provision for sending gas and taking the electrical output. Gas is sent from highly precise equipment which can maintain constant flow rates with MASS Flow controllers. Precise gas chromatography (GCC) equipment is used to measure the PPM (parts per million) of gas which is flowing through the air tight chamber. The measured raw output voltage from the sensor is found to be highly unstable and of very low magnitude. Signal conditioning circuits were designed to stabilize and amplify the measured signal from the sensors during the calibration process. The sensors along with the conditioning circuit are placed in the chamber and readings are noted down for regular PPMs of the gas. Each sensor produces a voltage value corresponding to the input concentration of gas. These observed values are plotted and a characteristic equation is formulated to map voltage signals into corresponding concentrations in PPM. Calibration and conditioning process is out of the
17
scope of this thesis and is not dealt with any further.
Figure 3.2: Various stages in calibrating gas sensors
3.3
Configuring Wireless Sensor Nodes
The pre-calibrated commercially available gas sensors are interfaced to wireless sensor motes/modules through the gas sensor board, which are programmed for air pollution monitoring application. Libelium WASPmotes are used as the basic wireless communication module, which comprises of the processing unit and the communication unit refer Figure 3.4 and 3.3. ADC (analog to digital converter) ports of the wireless nodes are programmed to periodically sample the various gas sensors interfaced to the sensor board on a rotational basis. The collected samples are packetized and sent to base station [12] [16] at regular intervals from each of the sensor node, which forms the mesh network [ refer Figure 3.5 ]. To increase the monitoring range, multi-hop data transfer algorithm [31] was implemented. To configure RF Xbee module, gain of signal conditioning and other modules on the WASP system refer [11]. The real time pollution monitoring test bed was developed and deployed with five node network.
Figure 3.3: Waspmote Board
18
Figure 3.4: Waspmote Gas Sensor Board
Figure 3.5: Multihop mesh network system architecture
3.4
Lightweight middleware and web interface
Base station or the sink node receives data at regular intervals of time from the deployed network. Light weight middleware is developed for effective storage and retrieval of data. An application to read data from serial port and convert to appropriate format is developed using C#. Parsed data is logged in to the database in the form of tables along with the time stamp of each packet. A web based graphical user interface (GUI) is developed to view the live data in the form numbers and charts, which is made accessible from anywhere on internet. Data is sent to the base station through multihop network. Network deployment and the web interface are discussed in detail in further chapters.
19
Chapter 4 Real time monitoring WSN deployment of the air pollution monitoring system was carried out in three different modes: static network , dynamic network and individual motes.
4.1
Static Multi-hop WSN
Static multi-hop is often used in the context of structured WSN scenarios. In a structured WSN, all or some of the sensor nodes are deployed in a pre-planned manner. The advantage of a structured network is that fewer nodes can be deployed with lower network maintenance and management cost. Fewer nodes can be deployed now since nodes are placed at specific locations to provide coverage.Another clear advantage of static WSN is the low power consumption because of less computation required. In this, all the nodes are not equal. The leaf nodes do not have relaying capability inbuilt and thus consume lesser power in comparison to the intermediate nodes. This scenario is similar to data aggregation using clustering where there are heterogeneous nodes and one needs to choose cluster heads accordingly. Without further going into the clustering details , we look into the implementation of static multi-hop WSN. In a static network, the nodes are not moving i.e they are still and hence the addresses they need to transmit the data to are fixed. Implementing a static network is relatively simple, as the algorithm is simple. We can assign the destination address within the code and can even implement multi-hopping in order to increase the distance from the central server. If a particular intermediate mote in a multi-hop path is not working we can code the transmitting mote to hop to another mote. With this, the network will survive and the central server will receive the packets. However, the major disadvantage is that we cannot scale the network beyond a limit. 20
This would increase code complexity and increase network congestion. Also if destination mote addresses are hardcoded into the mote then we need to program every mote to also accommodate change in address. This happens when a particular mote is replaced with a new mote or additional motes are added to expand the area of coverage. The following is an implementation of a static network with multi hopping. The motes are equipped with temperature, oxygen and atmospheric pressure sensors. Also there is on board RTC through which we get the time stamp i.e the time at which data was collected. In Figure 4.1, we present the basic architecture of 5 node network that were used for deployment in IIT Hyderabad campus and Kukatpally, Hyderabad.
Figure 4.1: Multi-hop architecture The network consists of 4 motes. Mote number N4 is farthest from sink(base-station, here N1) and is not in the range whereas motes N2 and N5 are within the range of the base-station. Therefore, N4 transmits its packet to N5. An acknowledgement is given by receiver if it has received data properly. N5�s job is to transmit its own packet as well as N4’s packet. This is multi-hop transfer. N2 and N5 transmit their own packets to the server. Now if N5’s power is down or is not responding then N4 sends the data to N3. Each packet consists of mote number, sensor values and time stamp. In the multi-hop transfer path may also be recorded. In Figure 4.2, the implementation flow diagram of Figure 4.1 is shown. The code for the same is done using C++ like language in the Waspmote IDE provided by the manufacturers of waspmote. The deployment at IIT Hyderabad campus and at Kukatpally, Hyderabad is shown in 4.4 and 4.3 respectively. 21
Figure 4.2: Static multi-hop implementation overview
Figure 4.3: Deployment in Kukatpally a suburb in Hyderabad
Figure 4.4: Deployment in IIT Hyderabad campus 22
4.2
Dynamic Multi-hop WSN
Algorithms for static networks are not particularly useful for mobile networks as the topology of the network varies from time to time. A network where topology changes dynamically is defined as Ad hoc network.. If the wireless nodes are within the range of each other, the routing is not necessary. If a node moves out of this range, and they are not able to communicate with each other directly, intermediate nodes are needed to organize the network and take care of the data transmission. Implementing a routing algorithm to route the information is a convenient choice. The purpose of a routing algorithm is to define a scheme for transferring a packet from one node to another. This algorithm should choose some criteria to make routing decisions, for instance number of hops, distance etc. Ad-hoc On-demand Distance Vector routing (AODV) is one such routing protocol. AODV is reactive protocol i.e. path is set only when need arises. The operation of AODV is made loop-free by using concept of sequence numbers thus avoiding the Bellman-Ford “count-toinfinity�problem. A sample output of the implementation grabbed on the serial terminal on gateway is shown in Figure 4.5.
4.2.1
AODV
Route Requests (RREQs), Route Replies (RREPs), and Route Errors (RERRs) and HELLO are the message types defined by AODV. As long as the endpoints of a connection have valid routes to each other, AODV does not play any role. Information about routes are maintained in a routing table. When a route to a new destination is needed or there is link break in existing connection, the node broadcasts a RREQ to find a route to the destination. The destination address and source address is mentioned in the RREQ. Any node receiving an RREQ determines if it itself is the destination or if it has a fresh route to the destination. The freshness is determined by comparing a special id called sequence number in the packet and sequence number in the routing table for corresponding destination. If sequence number in routing table is greater and the route is valid then it means there is a fresh route to the destination. This route is made available to the source by unicasting a RREP back to the origination of the RREQ. Each node receiving the request RREQ records a route back to the originator of the request in its routing table, so that the RREP can be unicast from the destination along a path to that originator, or
23
Figure 4.5: AODV implementation Sample terminal output on gateway from any intermediate node that has the required link. The originator, after receiving the RREP, stores the address of next node. In this way a link between originator and destination is established. RREQ is rebroadcasted by nodes who received an RREQ and they do not have any information related to the destination. The RREQ broadcast stops once a node with the information has responded. Nodes also monitor the link status of next hops in active routes. When a link break in an active route is detected, a RERR message is used to notify other nodes that the loss of that link has occurred. The RERR message indicates those destinations which are no longer reachable due to the link break. In order to enable this reporting mechanism, each node keeps a "precursor list", containing the address for each its neighbors that are likely to use it as a next hop towards each destination. The information in the precursor lists is feeded during the transmission or reception of RREP packet.
24
4.2.2
Routing Table
AODV is a routing algorithm which mainly deals with routing table management. Each mote has to maintain a routing table within itself which acts as lookup table when it receives or wants to transmit any information. The important fields in routing table are: • Destination Address • Source address • Destination Sequence number • Source Sequence number • Hop Count • Next Hop • Mote number/ Identity • Valid flag • List of precursors • Time to live A sample terminal output on the serial terminal software is shown in figure:4.6
4.2.3
Advantages and Disadvantages
The biggest advantage of AODV is loop free networking and route discovery. However there are limitations which affect network optimization. For example if there are hundreds of nodes then most of the packets would be control packets and the data packets would be few. This also poses a problem when the network needs to be scaled. Increase in nodes may cause network congestion and loss of data packet. Also with increase in broadcast messages, power requirement increases. Hence it needs to be ensured that continuous power is available or there are any means to charge if power is low. Security is drawback of AODV. Care needs to be taken that all nodes belong to the same network and any external intervention by another network should be discarded. Also, since the it requires more computation there is more power consumption. In our case, we also came to notice that the the programming stack often got corrupted and 25
Figure 4.6: AODV implementation Sample terminal output illustrating packets and process therefore not very highly complex programs can be written on Waspmote. This was the main reason we averted AODV implementation and preferred static WSN as well as individual mote architectures.
4.3
Individual motes
A number of motes were deployed at the Y-point gate IIT Bombay for sensor data collection. The motes were put up at the Security Kiosk at YP Gate (Market gate) of IIT Bombay. Data collected on memory card with date timestamps and sensor values in comma separated values (CSV) format Two sets of sensor motes were put to measure the following: Temperature, Pressure, Humidity, CO2 , O2 , N O2 , Air Contaminant RTC (Real Time Clock) was used to give a time stamp to all the data samples collected. The motes were programmed to collect data at intervals of 3 minutes 45 seconds (225 seconds) as can be seen in figure 4.10 on page 28. Collecting the data at a higher rate would drain the battery at a higher rate and causing the system to shut down after some hours. Thus the data collection events were spaced out in order to run the system for a long time (practically indefinitely with a live power source). The output of a mote can be seen in figure 4.9 on page 28.
26
Figure 4.7: Waspmotes deployed at Y-Point gate IIT Bombay
Figure 4.8: Mote Placement at IIT Bombay Gate
27
Figure 4.9: Output of Programmed Mote The data was stored in a CSV file on an SD card mounted on the mote. The comma-separated values (CSV) file format is a set of file formats used to store tabular data in which numbers and text are stored in plain textual form that can be read in a text editor. Lines in the text file represent rows of a table, and commas in a line separate what are fields in the tables row. And thus it enabled us to directly import the CSV file to a spreadsheet on a computer. The data was copied from the SD card onto the computer at intervals of every 3-4 days. The reason the system was not left up and running for a long time at a stretch was that if the charging power supply was disturbed or switched off due to some reasons (like loose connections, power outage, etc.), the mote would switch off within hours and all the data collected so far would be lost for ever as it was programmed to recreate the file if restarted. Also if the mote was restarted after this without reprogramming it, the new data collected would be corrupted with wrong time-stamps and hence not useful. Then I converted the values so obtained from the sensors into their respective ppm/% concentrations and stored them in a database on the central server.
Figure 4.10: Sample Sensor Data
28
Chapter 5 Data Observation & AQI 5.1
Observations
Data after collection is analyzed using SSAS and plotted using tools such as MATLAB [35] and Weka with different resolutions. For each of the parameters measured 3 representations are presented: • 24-hour (daily) • 7-day (weekly) • 17-day (fortnightly)
5.1.1
Air Contaminant
Figure 5.1 depicts the air contaminant levels measured through a 17-day period. The x-axis is in hours and the y-axis represents air contaminant level. The higher the level, more is the contaminant. As can be seen from the daily trend (figure 5.1A), starting from 12 am (midnight), the air contaminant level keeps dipping till around 5 am in the morning. Then it keeps rising till around 11 am after which it again begins a downward till around 6 pm in the evening. Then it again rises (and keeps rising) till 12 am of the next day. The reason for this is that from 12 am to 5 am there is no traffic on the roads and hence the contaminants keep on reducing. Then the rush hour starts with most people leaving for work and continues till around 11 am, after which there is again a lull in the traffic till around 5-6 pm. After that many office-goers start from their
29
offices to their homes and this increases the pollution till 12 am. This trend of there being two troughs within a day is being repeated throughout.
Figure 5.1: Air Contaminant Trends: (A)24-hour (daily) (B)7-day (weekly) (C)17-day (fortnightly)
5.1.2
Carbon Dioxide – CO2
Figure 5.2 depicts the CO2 levels measured through a 17-day period. The x-axis is in hours and the y-axis represents CO2 level (in ppm). The higher the value, more is the level of CO2 in the atmosphere. As can be seen from the daily trend (figure 5.2A), starting from 7 am to 1 pm, the CO2 level keeps dipping, after which it rises and keeps rising till 7 am next day. This is also consistent with prior research [5]. The concentration during the middle of the day is almost 30
uniform at ground level. This also agrees with prior research on CO2 [37]. Stratification of air (layers of hot air rising) during the day are responsible for the dip. The same layer comes down during the night and hence the level rises again. Also the area near the mote deployment was heavily forested, hence plant respiration has an important part to play. Due to photosynthesis, the plants consume CO2 which causes the level to fall in the day. But the plants release CO2 at night and so the level keeps rising at night till the morning. As can be seen in figure 5.2(B) and figure 5.2(C), this trend of there being one trough in the middle hours of a day is being repeated throughout. Also, it is a well known issue with CO2 measurements that fluctuations in the levels are rapid and high, and subject to large errors [5], which makes CO2 measurement difficult.
Figure 5.2: CO2 Trends: (A)24-hour (daily) (B)7-day (weekly) (C)17-day (fortnightly)
5.1.3
Nitrogen Dioxide – N O2
Figure 5.3 depicts the N O2 levels measured through a 17-day period. The x-axis is in hours and the y-axis represents N O2 level in ppm. The higher the ppm value, more is the pollution. It can 31
be seen from the trends that usually there is a spike in the N O2 levels once a day. This is usually at the time when maximum pollutants are released. This can be seen clearly in figure 5.3B and figure 5.3C. N O2 is a trace gas and hence the values are not very high. The N O2 sensor was not calibrated recently and hence there is a loss of calibration here.
Figure 5.3: N O2 Trends: (A)24-hour (daily) (B)7-day (weekly) (C)17-day (fortnightly)
5.1.4
Oxygen – O2
Figure 5.4 depicts the O2 levels measured through a 17-day period. The x-axis is in hours and the y-axis represents O2 level in percentage. The higher the percentage value, more is the level of oxygen in the air. It can be seen from the trends that the levels of O2 do not vary much. This can be seen clearly in figure 5.4B and figure 5.4C. The O2 sensor was not calibrated recently and so while the trend is confirming to what I expected, the values are not confirming (16% instead of 20%)
32
Figure 5.4: O2 Trends: (A)24-hour (daily) (B)7-day (weekly) (C)17-day (fortnightly)
5.1.5
Relative Humidity
Figure 5.5 depicts the RH (relative humidity) levels measured through a 17-day period. The x-axis is in hours and the y-axis represents RH level in percentage. The higher the percentage value, more is the RH in the air (and hence the chances of precipitation are more). It can be seen from the daily trend (figure 5.5A) that RH keeps increasing till around 10 am. This is due to layers of cool air coming down (stratification) and bring the dew and some water vapor along with it. Then due to heat from the sun, the RH keeps decreasing till around 6 pm. After sunset, which occurs at around 6 pm, the RH starts to increase again till the end of the day. It can be seen clearly in figure 5.5B and figure 5.5C that the trend repeats throughout. This also corresponds to actual trends from other sites available [36].
33
Figure 5.5: Relative Humidity Trends: (A)24-hour (daily) (B)7-day (weekly) (C)17-day (fortnightly)
5.1.6
Temperature
Figure 5.6 depicts the temperature levels measured through a 17-day period. The x-axis is in hours and the y-axis represents temperature is in â—Ś C.
34
Figure 5.6: Temperature Trends: (A)24-hour (daily) (B)7-day (weekly) (C)17-day (fortnightly)
5.2
AQI
The Air Quality Index (AQI) [21] is an index for reporting daily air quality. It tells you how clean or unhealthy your air is, and what associated health effects might be a concern. AQI (Air Quality Index) also known as the Air Pollution Index (API) or Pollutant Standard Index (PSI) is a number normally ranges from 1 to 500 which characterize the quality of the air at a given location. In general terms, AQI is a scale designed to help one understand what the air quality around you means to your health. Higher the AQI, higher the risk to health.The AQI focuses on health effects one may experience within a few hours or days after breathing unhealthy air. The AQI is calculated mainly for four major air pollutants ground-level ozone, particle pollution, carbon monoxide, and sulfur dioxide as shown in Figure 5.7. An AQI value of 50 represents good air quality with little or no potential to affect public health, while an AQI value over 300 represents air quality so hazardous that everyone may experience serious effects. An AQI value of 100 generally corresponds to the air quality standard for the pollutant, which is the level EPA
35
(Environmental Protection Agency) has set to protect public health. AQI values at or below 100 are generally thought of as satisfactory. When AQI values are above 100, air quality is considered to be unhealthy at first for certain sensitive groups of people, then for everyone as AQI values increase. AQI depends on geography and hence the norms are different for different countries as can be seen in the Figure 5.7 and Figure 5.8
Figure 5.7: Air Quality Index (US Standards)
Figure 5.8: AQI Sub-index and breakpoint for India as proposed by IITM (MoES) Each category corresponds to a different level of health concern: • Good: The AQI value is between 0 and 50. Air quality is satisfactory and poses little or no health risk. 36
• Moderate: The AQI is between 51 and 100. Air quality is acceptable; however, pollution in this range may pose a moderate health concern for a very small number of individuals. People who are unusually sensitive to ozone or particle pollution may experience respiratory symptoms. • Unhealthy for Sensitive Groups: When AQI values are between 101 and 150, members of sensitive groups may experience health effects, but the general public is unlikely to be affected. 1. Ozone: People with lung disease, children, older adults, and people who are active outdoors are considered sensitive and therefore at greater risk. 2. Particle pollution: People with heart or lung disease, older adults,1 and children are considered sensitive and therefore at greater risk. • Unhealthy: Everyone may begin to experience health effects when AQI values are between 151 and 200. Members of sensitive groups may experience more serious health effects. • Very Unhealthy: AQI values between 201 and 300 trigger a health alert, meaning everyone may experience more serious health effects. • Hazardous: AQI values over 300 trigger health warnings of emergency conditions. The entire population is even more likely to be affected by serious health effects.
5.2.1
Calculating AQI
Calculation of AQI is as illustrated as shown in figure:5.9. In this way, AQI of individual pollutants is calculated. A weighted average of the individual AQIs are then taken to calculate consolidated AQI. Individual and consolidated AQIs are calculated on the base-station and thus saving the computation cost on the motes. These AQIs are the ones which assists in providing the notions of context and predictive analytics which will be explored in the next chapter.
5.2.2
Air Quality Index Trends
Figure 5.10 depicts the AQI levels measured through a 17-day period. Here AQI means the consolidated AQI. The x-axis is in hours and the y-axis represents air-quality-index. The higher 37
Figure 5.9: Calculation of AQI the AQI Value, higher is the level of pollution. As can be seen from the daily trend (Figure 5.10(A), the AQI does not show drastic changes in its value. It depends on various pollution parameters and does not have any units. The thing worthy of notice is that AQI is consistently in the unhealthy zone (150+) throughout except at 2 major locations (as can be seen in Figure 5.10(C)). The trough (144 to 312 hours) is due to there being a low level of CO2 during that period as can be seen in figure 5.2 on page 31. Thus it is easily seen that most of the time the air that we breathe is highly polluted and will have a debilitating effect on our well-being as well as the health of our environment.
Figure 5.10: Air Quality Index Trends: (A)24-hour (daily) (B)7-day (weekly) (C)17-day (fortnightly)
38
Figure 5.11 and Figure 5.12 shows plot of AQI vs hours of day and days respectively. It also show other AQIs which are individual gas AQIs and not the consolidated AQI.
Figure 5.11: AQI Vs Hours of the day
Figure 5.12: AQI Vs Days
39
Chapter 6 Data Analysis and Context-aware PSE 6.1
Data Warehouse
In computing, a data warehouse is a database along with tools and utilities used for reporting and analysis. In the first part of work , Microsoft SQL Server 2005 had been used and in later stages the project used MySQL database due to operating system dependencies and technological compatibilities. The DWH used is a typical ETL-based data warehouse. It uses staging and integration. The staging layer stores raw data extracted from each of the networks deployed. The integration layer integrates these disparate data sets by transforming the data from the staging layer often storing this transformed data into database.The integrated data is then moved to yet another database, often called the DWH database, where the data is arranged into hierarchal groups often called dimensions and into facts and aggregate facts. In this project, databases resulting from IIT Bombay deployment and IIT Hyderabad deployment have been aggregated into DWH database. In the process data is cleaned, transformed and loaded to be readily usable for data mining and OLAP. Cleaning and transformation is done using Microsoft SSIS. Finally, the access layer helps users retrieve data. The same has been illustrated in Figure 6.1
6.1.1
Database Design
The database at the deployment site consists of two tables : Table 6.2 and a table similar to as shown in Table 6.1. Foreign key in the Table 6.1 is the primary key of Table 6.2. After ETL process, the databases are integrated into a single table at the central server for analysis and reporting. This table is shown in 6.1
40
Figure 6.1: Data warehouse ETL and staging
Column name
Comments
q_i_time
Timestamps with date and time
motenum
Mote identifier
gpserror
GPS information available?
temperature
in Celsius
humidity
Relative humidity in percentage
pressure
Atmospheric pressure in Kilo Pascals
co2
Carbon dioxide in ppm
o2
Oxygen in percentage
no2
Nitrogen dioxide in ppm
aircontm
Air Contaminant in ppm
latitude
Latitude obtained from GPS module
longitude
Longitude obtained from GPS module
altitude
Altitude obtained from GPS module
dategps
Date obtained from GPS module
timegps
Time obtained from GPS module Table 6.1: Database table maindata as DWH table
41
Column name
Comments
motenum
mote identifier
o2_param1
Oxygen parameter#1
o2_param2
Oxygen parameter#2
co2_param1
Carbon dioxide parameter#1
co2_param2
Carbon dioxide parameter#1
no2_param1
Nitrogen dioxide parameter#1
no2_param2
Nitrogen dioxide parameter#1
Table 6.2: Database table Mote information at individual base stations
42
6.2
Descriptive models using data mining
So much data exists that it overwhelms traditional methods of data analysis.Data mining provides a way to get at the information buried in the data. Data mining finds hidden patterns in large, complex collections of data, patterns that elude traditional statistical approaches to analysis. Descriptive models focus more on the intrinsic structure, relations, interconnectedness, etc. Microsoft SSAS 2005 has been used for generating these models. Here, we basically deal with unsupervised and supervised learning methodologies.
6.2.1
Clustering
Clustering is an unsupervised learning and gives the first hand information about similar and dissimilar patterns of data. Clustering analysis identifies clusters embedded in the data. A cluster is a collection of data objects that are similar in some sense to one another. A good clustering method produces high-quality clusters to ensure that the inter-cluster similarity is low and the intra-cluster similarity is high. On analyzing clusters more rigorously one often finds greater insights. Criterion functions are often used for deciding the quality of the cluster. Figure 6.2 depicts the clustering diagram and the probabilities associated with the cluster nodes using SSAS. In this representation, darker the cluster nodes greater is the number of data points in that cluster. Also, bolder the lines connecting two cluster nodes, greater is the similarity between the clusters.
Figure 6.2: Clustering using Microsoft Clustering algorithm
43
6.2.2
Decision Trees
Decision tree analysis is a widely-used method of constructing a model from a dataset in the form of a decision tree or (equivalently) a set of decision rules. It is a supervised learning method and is often claimed that this representation of the data has the advantage compared with other approaches of being meaningful and easy to interpret. The aim is to develop classification rules from the data in the training set. This is often done in the implicit form of a decision tree. A decision tree is created by a process known as splitting on the value of attributes i.e. testing the value of an attribute and then creating a branch for each of its possible values. In the case of continuous attributes the test is normally whether the value is ‘less than or equal to’or ‘greater than’a given value known as the split value. The splitting process continues until each branch can be labeled with just one classification. Decision trees have two different functions: data compression and prediction. Amongst other data mining methods, decision trees have various advantages: • Simple to understand and interpret. • Requires little data preparation. Other techniques often require data normalization, dummy variables need to be created and blank values to be removed. • Able to handle both numerical and categorical data. Other techniques are usually specialized in analyzing datasets that have only one type of variable. Ex: relation rules can be used only with nominal variables while neural networks can be used only with numerical variables. • Uses a white box model. If a given situation is observable in a model the explanation for the condition is easily explained by boolean logic. An example of a black box model is an artificial neural network since the explanation for the results is difficult to understand. • Possible to validate a model using statistical tests. That makes it possible to account for the reliability of the model. • Robust. Performs well even if its assumptions are somewhat violated by the true model from which the data were generated. • Performs well with large data in a short time. Large amounts of data can be analysed using standard computing resources. 44
Figure 6.3 depicts the decision tree and Figure 6.4 shows the dependency network using SSAS. The predict parameter used in the algorithm is AQI. The most important cause of the pollution shown by the dependency network is N O2 , followed by Air Contaminant. A decision tree is composed of a series of splits, with the most important split, as determined by the algorithm, at the left of the viewer in the All node. Additional splits occur to the right. The split in the All node is most important because it contains the strongest split-causing conditional in the dataset, and therefore it caused the first split. The weakest is on the right. Here, N O2 is strongest determiner and Hour is at level 3 for determining AQI(Air Quality Index). The Dependency Network 6.4 displays the dependencies between the input attributes and the predictable attributes(here, AQI) in the model. When the slider is moved from top to bottom it highlights from NO2 to O2 in order of decrease in relevance. The most important attribute impacting AQI is shown first here, NO2 and then the rest are highlighted according to decision tree.
Figure 6.3: Decision tree using Microsoft Decision tree algorithm
6.2.3
Association Rules
Classification rules are concerned with predicting the value of a categorical attribute that has been identified as being of particular importance. Here, we go on to look at the more general problem of finding any rules of interest that can be derived from a given dataset. Unlike classification, the left- and right-hand sides of rules can potentially include tests on the value of any 45
Figure 6.4: Dependency network of decision tree attribute or combination of attributes, subject only to the obvious constraints that at least one attribute must appear on both sides of every rule and no attribute may appear more than once in any rule. It represents an association between the values of certain attributes and those of others and are therefore called association rules. process of extracting such rules from a given dataset is called association rule mining (ARM). It basically depends on support and confidence as described below: • The support supp(X) of an itemset X is defined as the proportion of transactions in the data set which contain the itemset. • The confidence of a rule is defined conf(X⇒ Y ) = supp(X ⇒ Y )/supp(X).
Figure 6.5 depicts some of the interesting association rules derived from the dataset using SSAS. The predict parameter used in the algorithm is AQI.
46
Figure 6.5: Association rules using Microsoft ARM algorithm
6.3
Context-awareness
The context aware applications are certainly not exhaustive and have diverse behaviors in ubiquitous systems. But the custom design applications only limits them. In this work, an attempt to build a context-aware framework is made rather than building applications. This framework provides web services which can be used by applications on mobile phones and motes like waspmote using GPRS. The framework can also be used by various web applications in aiding visualization for decision making, improvising and other varied purposes. Thus the notion of reuse and software architecture becomes increasingly essential. Context aware is often interpreted as intelligence in the application due to the ability of decision making. Intelligent software agents are an emerging class of software systems that are proactive, autonomous, communicative (with people and other agents), and adaptive. Context-aware pervasive computing is a study of pervasive computer systems (a combination of hardware and software) that are aware of context and can automatically adapt and respond to such context. Context awareness enables the system to take action automatically, reducing the burden of excessive user involvement and providing proactive intelligent assistance. Here we aim to separate context-aware completely from the applications.
6.3.1
Elements of CAPS
A context-aware pervasive system can be viewed as having three basic functionalities: sensing, thinking (metaphorically), and acting. These functionalities can be realized in a centralized or a distributed architecture over one or more physical devices and sophistication can vary from
47
system to system. Figure 6.6 shows an abstract layered architecture from [3], labeled with subsystem divisions. The word sensors at the bottom layer relates to the definition of sensors used for raw data retrieval. Subsequent preprocessing or reasoning of the data is then carried out, and resulting context information stored. Storage and management of context can be sophisticated with support for querying, further reasoning, and updates, or much simpler, depending on application requirements. The three subsystems of acting, thinking, and sensing involve sensors and raw data retrieval, preprocessing and management of context, and application-dependent actions, respectively.
Figure 6.6: Abstract layered architecture of CAPS
6.3.2
Service framework
Service-oriented computing has the notion of services as its central operative idea, made popular by developments in Web service technologies. Computationally, services are an abstraction of a unit of functionality. With the development of the mobile Internet or mobile Web, such services can then be engaged over wireless networking technologies anywhere and anytime and thus supporting the universal theme of pervasive systems. Wireless Application Protocol [33] standards and applications were developed to support lightweight content that could fit within small resource-constrained devices and operated. Such content, even if more efficiently processed with less memory and computational power can be as expressive compared to the desktop Web. With the recent development of wide area high bandwidth wireless networking technologies such as 3G and 4G, even multimedia functionalities are enhanced. Waspmotes used in the network deployment cannot make complex calculations due to inherent limitations and can thus make use of the web services to facilitate it. Also, with the advent of integration of
48
sensors on mobile phones, it is no more a data collection device as it can easily be as expressive as desktop web showing charts, interacting with computer models and many more. Recent research in the field of context-awareness has predominantly adopted an infrastructurecentered approach; that is, it has assumed that the complexity of engineering context-aware applications can be substantially reduced solely through the use of infrastructure capable of gathering, managing and disseminating context information to applications that require it. In line with this approach, a variety of solutions that acquire and interpret context information from sensors, and manage repositories of information that support queries and notifications, have been proposed. These include the Context Toolkit [10], the Solar platform [6], and various context services [28], [24]. These solutions help to simplify application development and promote reuse of functionality. The most important aspect here is the query layer. The query layer provides applications and other components of our software infrastructure with a convenient interface with which to query the context management system. All the nodes just need to send an appropriate query packet to web service available to it and thus can make use of all the web computation and analytics to provide the nodes or a group of nodes a level of intelligence. This intelligence thus enhancing the context-awareness. Another important layer is the context management layer. It is responsible for maintaining a set of models and their instantiations.
6.3.3
Implementation
The implementation of the web services has been done entirely in AJAX coupled with JSP and jQuery. The services fall under 2 categories: Data trends and business intelligence. In Data trends, provisions for following services are provided, • List of environment parameters data request for any coordinate/node • List of environment parameters continuous data request for any coordinate/node • All environment parameter data request for any coordinate/node • All environment parameter continuous data request for any coordinate/node • Data trend as chart for an environment parameter specifying X and Y axis • List of available coordinates/nodes
49
Figure 6.7: Air pollution parameters of node#1 Figure 6.7 shows the display of all the air pollution parameters of node#1 on the web interface.
Following services related to business intelligence have been provisioned, • Decision tree model of a slice and dice of the OLAP cube • Clustering model of a slice and dice of the OLAP cube • ARM model of a slice and dice of the OLAP cube • List of nodes with AQI range as parameter to the query The context needs to be derived from the services enlisted above in various combinations. For example, in order to find the root cause of High AQI, one needs to first get a list of nodes having high AQI and then use this info to slice the cube to create decision tree model and then finally from dependency network, one needs to retrieve environment parameters. After providing this bare framework for finding context, a more intelligent framework is the necessity in future. For this, another abstract layer needs to be provided which can provide high level services.
50
Chapter 7 Challenges, Conclusion and Future scope 7.1
Challenges
Although, the network deployment was successful, the major challenges faced were: Sensor Calibration: Air pollution sensors like CO2 , CO and N O2 are resistive heating based sensors. They consume a lot of energy from the battery of wireless nodes which is detrimental to network life time. Chemical or MOSFET sensors need very less power but the cost is too high. The effect of temperature and humidity on resistive type gas sensors is to be considered for accurate readings. Calibration at periodic intervals is necessary but it is difficult to do for large no of sensors in field. Life of the sensors is very short (typically 6 to 9 months)[14]. High cost of waspmotes: The network deployment for even 1 city like Mumbai would cost around 6 Billion INR. If these Waspmotes are deployed only on the roads, then the cost would be 377 Million Lac INR. This is only hardware cost. Only simple implementations: Waspmotes cannot handle complex codes as the stack gets corrupted often.
51
Figure 7.1: Hardware cost for fine-grained air pollution monitoring in Mumbai, India
7.2
Conclusion
A prototype of the air pollution monitoring for collecting fine-grained pollution data is successfully deployed. The framework for data analysis and deriving context has also been suitably built. Although, offline analysis of the data is completed, an OLAP engine needs to be refined. There have been many software technologies used for the implementation of the project and are illustrated in the Table 7.1 Database
Microsoft SQL Server 2005, MySQL
Gateway Program
C#
Mote Program
C++ like
Web server
AJAX, JSP, PHP
Analysis
Weka, SSAS, Pentaho
Web Server
Glassfish, Apache Table 7.1: Software Technologies
To make it pervasive in nature, one needs to address the cost of the motes and/or use exiting pervasive gadgets like mobile phones by integrating sensors to it. The cost reduction of motes has been carried out by IIT Hyderabad team, a project partner at IUATC. Also, the integration of sensors to mobile has been recently carried out at SPANN lab, IIT Bombay. Due to better processing capability of mobiles in comparison to motes used, integrating the sensor 52
to mobiles enables the system to compute complex logic at the node itself. Thus making the system more dynamic, robust and real-time.
7.3
Future Scope
Although the system prototype has been successfully shown in working , a lot needs to be done to rescale it to large deployments and large data. Better management of the sensor data and meeting network QoS can be more efficiently achieved using cloud computing framework for sensor data. Data fusion with weather data is another area of high impact which is unexplored in this project. This project has potential of making high social and economical impact. New projects like traffic monitoring, display of environmental conditions to public, etc can be done using the framework developed here. A meta-engine is a necessity for the framework to work seamlessly with diverse applications.
53
References [1] U.S. Environmental Protection Agency. Air Quality Index : A guide to air quality and your health. http://www.epa.gov, Last accessed on 30th May 2012. [2] I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci. Wireless sensor networks: a survey. Computer networks, 38(4):393–422, 2002. [3] Matthias Baldauf, Schahram Dustdar, and Florian Rosenberg. A survey on context aware systems. Int. J. Ad Hoc Ubiquitous Comput., 2(4):263–277, June 2007. [4] R.T. Burnett, M. Smith-Doiron, D. Stieb, S. Cakmak, and J.R. Brook. Effects of particulate and gaseous air pollution on cardiorespiratory hospitalizations. Archives of Environmental Health: An International Journal, 54(2):130–139, 1999. [5] HW Chapman, LS Gleason, and WE Loomis. The carbon dioxide content of field air. Plant Physiology, 29(6):500, 1954. [6] Guanling Chen and D. Kotz. Context aggregation and dissemination in ubiquitous computing systems. In Mobile Computing Systems and Applications, 2002. Proceedings Fourth IEEE Workshop on, pages 105–114, 2002. [7] J. Chou. Hazardous gas monitors: a practical guide to selection, operation and applications. McGraw-Hill Professional, 2000. [8] A.J. Cohen, H.R. Anderson, B. Ostro, K.D. Pandey, M. Krzyzanowski, N. Kunzli, K. Gutschmidt, A. Pope, I. Romieu, J.M. Samet, et al. The global burden of disease due to outdoor air pollution. Journal of Toxicology and Environmental Health, Part A, 68(13):1301–1307, 2005. [9] European Community. Rescatame. http://www.rescatame.eu/, Last accessed on 15th June 2012. 54
[10] Anind K. Dey, Gregory D. Abowd, and Daniel Salber. A conceptual framework and a toolkit for supporting the rapid prototyping of context-aware applications. Hum.-Comput. Interact., 16(2):97–166, December 2001. [11] Engineering and Physical Sciences Research Council. The Urban Pollution Monitoring Project UK. http://www.equator.ac.uk/index.php/articles/563, Last accessed on 20th March 2012. [12] EPA. Six common air pollutants. http://www.epa.gov/oaqps001/urbanair/, last accessed on 20 June 2012. [13] D. Estrin, D. Culler, K. Pister, and G. Sukhatme. Connecting the physical world with pervasive networks. IEEE pervasive computing, pages 59–69, 2002. [14] D.J. Gaffen and R.J. Ross. Climatology and trends of us surface humidity and temperature. 2010. [15] D. Ganesan, A. Cerpa, W. Ye, Y. Yu, J. Zhao, and D. Estrin. Networking issues in wireless sensor networks. Journal of Parallel and Distributed Computing, 64(7):799–814, 2004. [16] J.C. Herrera, D.B. Work, R. Herring, X.J. Ban, Q. Jacobson, and A.M. Bayen. Evaluation of traffic data obtained via gps-enabled mobile phones: The mobile century field experiment. Transportation Research Part C: Emerging Technologies, 18(4):568–583, 2010. [17] A Ibald-Mulli, J Stieber, HE Wichmann, W Koenig, and A Peters. Effects of air pollution on blood pressure: a population-based approach. Am J Public Health, 91(4):571–577, 2001. [18] IITM and MoES. SAFAR. http://www.tropmet.res.in/, Last accessed on 15th June 2012. [19] Systems India-UK Advanced Technology Centre of Excellence in Next Generation Networks and Services. Workplan Theme 2: Pervasive Sensor Environments. http://www.iuatc.com/theme2.html, Last accessed on 15th May 2012. [20] J. C. St. John and W.L.Chameides. The climatology of ozone exceedences in the atlanta metropolitan area: 1_hour vs. 8_hour standard and the role of plume re_circulation. Environmental Science and Technology, 31:2797–2804, 1997.
55
[21] DL Johnson, SH Ambrose, TJ Bassett, ML Bowen, DE Crummey, JS Isaacson, DN Johnson, P. Lamb, M. Saul, and AE Winter-Nelson. Meanings of environmental terms. Journal of environmental quality, 26(3):581–589, 1997. [22] Young Jin Jung, Yang Koo Lee, Dong Gyu Lee, Keun Ho Ryu, and S. Nittel. Air pollution monitoring system based on geosensor network. In Geoscience and Remote Sensing Symposium, 2008. IGARSS 2008. IEEE International, volume 3, pages III –1370 –III –1373, july 2008. [23] Duk-Dong Lee and Dae-Sik Lee. Environmental gas sensors. Sensors Journal, IEEE, 1(3):214 –224, oct 2001. [24] Hui Lei, Daby M. Sow, John S. Davis, II, Guruduth Banavar, and Maria R. Ebling. The design and applications of a context service. SIGMOBILE Mob. Comput. Commun. Rev., 6(4):45–55, October 2002. [25] M.Welsh. Where do we go from here? the big problems in sensor networks. Keynote Address, Wireless Sensing Solutions Conference, September 28, 2005. [26] S. Neelamegam, C. Naveen, M. Padmawar, UB Desai, SN Merchant, and V. Kulkarni. Agrisens: Wireless sensor network for agriculture – a sula vineyard case study. In The International Conference on Distributed Computing in Sensor Systems (DCOSS 2007) Santa Fe, New Mexico, USA, pages 1–12, 2007. [27] Department
of
Transport
Stationary
office
London.
Ten
Year
Plan.
/http:/www.dft.gov.uk/about/strategy/whitepapers/previous/transporttenyearplan2000, Last accessed on 15th June 2012. [28] J. Pascoe. Adding generic contextual capabilities to wearable computers. In ISWC, pages 92–99, 1998. [29] C.E. Perkins, E. Belding-Royer, and S. Das. Ad hoc on demand distance vector (aodv) routing (rfc 3561). IETF MANET Working Group (August. 2003). [30] O. Pummakarnchana, N. Tripathi, and J. Dutta. Air pollution monitoring and gis modeling: a new use of nanotechnology based solid state gas sensors. Science and Technology of Advanced Materials, 6(3-4):251–255, 2005. 56
[31] Mohd Adib B Sarijari, Rozeha A Rashid, Mohd Rozaini Abd Rahim, and Nur Hija Mahalin. Wireless home security and automation system utilizing zigbee based multi-hop communication. 2008 6th National Conference on Telecommunication Technologies and 2008 2nd Malaysia Conference on Photonics, (August):242–245, 2008. [32] M. Satyanarayanan. Pervasive computing: Vision and challenges. Personal Communications, IEEE, 8(4):10–17, 2001. [33] S. Singhal, T. Bridgman, L. Suyranarayana, D. Manuey, J. Chan, D. Bevis, S. Hild, and J. Alvinen. The Wireless Application Protocol: Writing Applications for the Mobile Internet. Addison-Wesley Longman, 2001. [34] Quest
Technologies.
AQ5000
portable
http://www.questtechnologies.com/QuestFiles/61/QuestIAQ.pdf,
Last
detector. accessed
on
27th April 2012. [35] Inc. The MathWorks.
MATLAB - The Language Of Technical Computing.
http://www.mathworks.com/products/matlab/, Last accessed on 18 June 2012. [36] TREX.
US
Relative
Humidity
Graphs
from
Monitoring
Sites.
http://trexwww.ucc.nau.edu/cgi-bin/graph_data.pl, Last accessed on 15th June 2012. [37] J. Verduin and WE Loomis. Absorption of carbon dioxide by maize. Plant Physiology, 19(2):278, 1944. [38] M. Weiser and J.S. Brown. The coming age of calm technology [1]. Xerox PARC. Retrieved July, 8:2007, 1996. [39] N Xu. A survey of sensor network applications. IEEE Communications, 40(8):102–114, 2002. [40] Q. Yang, Y. Chen, D. Krewski, R.T. Burnett, Y. Shi, and K.M. McGrail. Effect of shortterm exposure to low levels of gaseous pollutants on chronic obstructive pulmonary disease hospitalizations. Environmental research, 99(1):99–105, 2005. [41] J. Yick, B. Mukherjee, and D. Ghosal. Wireless sensor network survey. Computer Networks, 52(12):2292–2330, 2008.
57