With the advancement in the use of neural network, differenttechniqueshavebeenproposedtomakethefault detection intelligent by detecting and predicting the occurrence and location of faults in the engine processes. ThiscanbeseenintheworkofStephenetal[9]wherethe useofneuralnetworkindetectingandpredictingfaultwas proposed.Theirideawasbasedonaproofofconceptwhich wasverifiedin[10]throughsimulationstudy.In[11],Gas TurbineFaultDiagnosisUsingProbabilisticNeuralNetworks (PNN) was proposed using three different classification methodscomparedwiththePNNandthetechniquewasable topredictthefutureoccurrenceoffaultsingasturbines.In [12],aneuralnetwork basedschemeforfaultdetectionofan aircraft engine. using a set of dynamic neural networks (DNN) was developed to detect fault occurrence in the engine. Despite the advancement in the use of neural networkindetectingfaultingasturbineengine,majorityof thetechniquessofararecomplex,expensive,andtooslowto
Inourworldtoday,gasturbineenginesplayimportantrole especiallyintheaviationindustry,mechanicaldrivers,and power generation [1]. Gas turbine engine also known as combustion turbine is a type of combustion engine consistingofacompressor,combustor,andturbinewitha basicoperationbasedontheBraytoncircle[2].Ithasbeenin existencesince1500andcontinuoustoevolvetilltoday.The basicoperationofthegasturbinebasedonBraytoncircleis showninFig.1[3].Fromthefigure,thegasturbinenormally followsfourthermodynamicprocessesnamelycompression (isentropic),combustion(isobaric,constantpressure),and isentropicexpansionandheatrejection. Basedontheabovedescription,differentresearchershave beeninvolvedindesigninggasturbineenginesofrecentdue to advancement in technology. This is done in order to improvethesystemperformanceandreliability.Areasthat seen increase in research is mainly in the detection and analyzing of faults from different aspects of the engine processes.
International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page1096
Abstract - In the modern aviation industry, gas turbine engines play important role in in the aircraft design. However, because of the complex nature of the design and operation of such engines, there is need to constantly monitor and maintain the engines to perform at its best. One of the most challenges aspects of such engines is the detection of fault as it can occur at any instance. Thus, detecting the fault and predicting it occurrence is of paramount important. In this paper, a novel technique of detecting and predicting faults in a gas turbine engine using a robust Neuro Fuzzy controller. The gas turbine mathematical model is design and validated analytically. The fault detection and prediction system are then designed using the Neuro fuzzy technique and validated through simulation study. The results obtained shows that the Neuro Fuzzy system was able to predict when the gas turbine is in error by checking with the train signal on the relay with a high signifying the detection of fault and 0 otherwise. In addition, the Neuro Fuzzy controller outperformed the PID controller, and this can be attributed to the fact the Neuro fuzzy has the ability to train and correct the error based on the previous training results.
1Mechanical engineer, Ministry of Information & Media Complex, Al Soor St, Al Kuwayt, Kuwait 2Lead Innovator, University of Sheffield & The University of Sheffield, Western Bank, Sheffield, S10 2TN, UK ***
1.INTRODUCTION
Fault Analysis and Prediction in Gas Turbine using Neuro-Fuzzy System Bader Alhumaidi Alsubaei1 , Ali Imam Sunny2
Fig 1:GasturbineenginebasedonBraytoncycle[3]. ThemostnotableworksincludesthatofWaleligneetal [3]whereageneralreviewongasturbinecondition based diagnosis was presented. They did by looking at different diagnosticsmethodsandtheiradvantagesanddisadvantage in the gas turbine engines. In [4], a stride in the engine performancebyalteringtheenginefuelcontrolframework, altering power turbine (PT) nozzle guide vane (NGV) and compressor inlet guide vane (IGV) are keys qualities was presented. Their technique was able to the turbine inlet temperaturetherebyincreasingtheturbineefficiency.In[5], a fault isolation technique in gas turbine engine based on componentqualificationwaspresented.Theapproachwas abletodeterminewhichcomponent(subsystem,system)is failingorhasfailed,andwhichkindoffaultmodeexisted.
Key Words: Dynamicmodelling,Gasturbine,Faultdetection, Mu Lawcompressor,Neuro fuzzycontroller,PIDcontroller.
Also there are other NDT and fault detection techniques available which are well established and applicable in multipleindustrialsectors[6,7,8],however,theirapproach needstobecorrelatedwithwatchedreactiontoadefective conditionwhichreducestheefficiency.
2.1 Gas Turbine Modelling
Theprocessstartsbyfirsttakinganairthroughtheairinlet andtheairgetcompressedatthecompressorwhosemain function is to increase the inlet air pressure. It then goes through the Combustion Chamber where a fuel is then sprayedinacontinuousmannertoignitetheairtomakethe combustiontoproduceaHighflowTemperature.Thenext stageissendingthehighpressureandhightemperaturegas toentertheturbinesectionwhichinturnsendsenergytothe turbinebladesmakingthemtorotate.Throughthisprocess, anelectricitycanthenbegenerated.
Based on the above model, the relationship between the compressorandtheshaftcanbeexpressedintermsofthe rotordynamicsusingtheenergybalanceequationgivenby [13] (1) WhereEistheenergy,tisthetime, istheefficiency, and and arethe work done in the turbine andthe compressorrespectively. In terms of the volume dynamics, the relationship betweentheinputandtheoutputofthegasturbinecanbe expressedusing (2) WherePisthepressure,Tisthetemperature,Risthegas constant,Visthevolume,andisthemassflowrate. Theaboverelationshipscanthenbeusedtomodelthegas turbineinSimulinkasshowninFig.3adoptedfrom[14].
International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page1097
Inthissection,theresearchmethodadoptedforthedesign andsimulation of the gas turbine systemisdiscussed.The modelwillbedevelopedbasedontheBraytoncyclemodel. Themathematicalmodelofthesystemisdesignedfirstafter whichthemodelwillbeverifiedanalytically.Thereafter,the modelwillbevalidatedthroughnumericalsimulation.The intelligent controller system using Neuro fuzzy is then designed after which the system will compared with PID controller method to show the efficacy of the Neuro fuzzy controllerinanalyzingandpredictingfaultsinagasturbine system.ThesummaryofthemethodologyisshowninFig.2.
2.3 Simulink Design for Fault Detection System
train the system. In this work, a novel technique using neuro fuzzyispresentedthatisrobust,simple,andfastin detecting and predicting fault in gas turbine engine. are developedtolearnthedynamicsofthejetengine.
Tomodelagasturbine,thedesignismostlybasedonBrayton cycle design as shown in Fig. 1. From the figure, the gas turbine consists of a compressor, combustor, and turbine.
2. FAULT DETECTION IN A GAS TURBINE
2.2 Simulink Design for Fault Detection System
Fig 2:Blockdiagramofthemethodologyadoptedforthis work.
Aftermodellingthegasturbine,inreallife,faultcanoccurin the gas turbine and a system to detect and analyse it is needed. In most gas turbine design, fault detection mechanism/techniqueareaddedtoknowwhenthesystemis operatingnormallyornot.Todothis,amethodtodetectthe faultisdevelopedandthenintegratedintothegasturbine modeldevelopedinSection2.1.thisisdonetodetectfaultsin the system, and it is an important and effective tool when operatorswanttochangefrompreventivemaintenanceto predictive maintenance with the goal of reducing the maintenancecost.Inthisstudy,theMu lawcompressorand thethermocouplefunctionareusedtomodelatechniqueof detectingfaultinagasturbine.Thedesignofthistechniqueis doneinSimulinkasshowninFig.4(adoptedfromthework in[15]).Fromthefigure,theatmosphericairgoesthrough theenginethroughtheairinletmodelledasstepinputinthe Simulink.Thereafter,areferencefilterisusedtomitigatefor anyerrorasa resultofdustfromtheair.Thisismodelled usingatransferfunctionshowninthefigure.Thenextstage is the compressor design based on Mu Law with internal circuitdesignasshowninFig.5.
Fig 3:Simulinkdesignofthegasturbineenginebasedon Braytoncycledesign.
Theremainingpartofthepaperisorganizedasfollows: Section 2 presents the fault detection technique used to detectthefaultatdifferentstagesinthegasturbineengine, Section3dealswiththeresultsanalysisanddiscussion,and section 4 deals with the summary and conclusion of the work.
Using the above table, the aggregation is done after viewingtheruleafterwhichtheoutputiddefuzziedintothe ANFISasshowninFig.8.
Journal
2.4 Design of a Neuro Fuzzy Controller System
Fig 5:SimulinkinternaldesignoftheMu Lawcompressor.
Condition (R1)Residue (R2)Residue (R3)Residue Normal 0 0 0 Fault1 1 0 0 Fault2 0 1 0 Fault3 0 0 1
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified |
AsshowninFig.4,thecompressoristhenconnectedto thecombustorandtheexhaustusingvariabledelayblocksas shown.Asummerisusedtoaddthegasconstantwhichis0.5 forourcase.Theiroutputcanthenbeviewedonascopeto showthepresent operational stageofthegasturbine.The combustordelayoutputisthenconnectedtoatemperature transfer function which is then linked to a relay to be switchingbetweenwhenafaultisdetected.Onceafaultis detected,the information from the relay is then compared withoriginalmessagetoknowwhenthereisafaultbasedon a threshold set and for our case, the error should not go beyond0.271,else,afaultisdetected.
Fromtheabovedesignacontrollerisneededtobedesign thatcanbeusedtocontrol thegasturbinesystemandthe designispresentedinthenextsection.
Forthecontrollerdesign,duetoadvancementintheuse of gas turbine engine especially in the modern aircraft, an intelligent controller is needed. In the literature, there are differenttechniquesofdesignintelligentcontrollerssuchas theworksin[16].Inthiswork,weadoptedtheNeuro fuzzy controller based on the Adaptive Nuero fuzzy Inference system(ANFIS)duetoitssimplicityanditsabilitytotrainthe system fast using the artificial neural network (ANN). In addition, a PID controller is design to compare the ANFIS design and the PID controller to show the efficacy of the system.TomodelthecontrollerusingtheANFIS,thereare four stages involved namely Fuzzification, Rule evaluation (inference),Aggregationoftheruleoutputs(composition), andDefuzzificationasshowninFig.6[17].
Fig 6:StagesinthedesignoftheNeuro fuzzycontroller. IntheFuzzificationstage,thecrispinputsaredefined.Inthe gasturbine,faultsoccurbasedonthreedifferentscenarios namelyfuelflowratefaultF1,faultattheexittemperature F2, and fault at the turbine output torque F3. These three faultsareusedasthecrispinputtocreatethemembership functionsasshownbythefuzzydesignin7. Fig 7:Fuzzydesignanditsmembershipfunction. The next stage is to add the rules governing the fault detection and the fault can be detected based on the conditionsshowninTable1.Thetableparametersareadded as rule governing the fuzzy design which can then be importedinANFIS.Theruleissettoflag1wheneverafaultis detectedwiththepositionofthe1indicatingwhichfaultis detected.Whenthereisnofault,theconditionis0. table.
Table 1: RulesEvaluation
Page1098
Fig 4:Simulinkdesignofthefaultdetectiontechniquein gasturbinesystem.
International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072
Certified Journal | Page1099
3. RESULTS AND DISCUSSION
Fig 8:Neuro Fuzzydesign.
3.1 Simulation Set Up and Parameter Definitions
Theneuro fuzzycontrollerdesignisshowninFig.9and the Simulink design for the modified fault detection techniqueincludingtheneuro fuzzycontrollerisshownin Fig.10. Fig 9:Neuro Fuzzycontrollerblockdiagram. Fig -10:Overalldesignforthefaultdetectionand predictionusingNeuro Fuzzycontroller. Thesimulationresultsarepresentedinthenextsection.
neuro fuzzy controller resultsare thencompared with the conventional PID controller to show the efficacy of the techniquedeveloped.
Forthegasturbine,thesimulationparameterswereset based on the Brayton Cycle settings with the following configurations: Compressor, alternator, and turbine mass flowrateof , poweroftheturbine100kW,the systeminputsareconfiguredbasedontheparametersshown intheplotinFig.11.Basedonthesesettings,thesimulation of the gas turbine can then be carried out with the results presentedinthenextsection.
Fig 11:Systeminputsconfigurationforthegasturbine.
International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008
In this section, the simulation results based on the methodsdescribedinsection2arepresented. Theresults start by first validating the model developed for the gas turbinethroughsimulationinMATLAB/Simulinkplatform. Thereafter, the model developed to detect the fault is simulatedandvalidatedusingperformancemeasuressuchas rotospeed,loadangleandtemperatureentropydiagram.The
Using the model developed in section 2 for the gas turbine,thesimulationwascarriedoutandtheresultsare presented as shown in Fig. 12. From the results, the validationofthemodelisdonebystudyingtheoutputpower andtheefficiencyofthemachine.Fromtheresults,itcanbe seenthatfortheoutputpower,thepowerwaslowsignifying the onset of the machine where the compressor is compressingtheairconvertingtohighpressure,thentothe turbine to get high temperature before the power is being generatedatatimeofaround500seconds.Themachinethen startedgeneratingpowerinastepsequenceuntilitstabilises at around 100 kW which is the maximum power of the turbine. For the efficiency of the system, it is low at the beginning before it also increases as the machine started generating power although the efficiency needs to be improvedbecauseitisabitlowlikelyduetosomefaultinthe system.Thus,integratingthegasturbinewithfaultdetection andpredictioncapabilityisparamount.
3.2 Simulation Results of the Gas Turbine System
Basedonthesimulationresultofthegasturbine,todetect theexistenceofanyfaultinthesystem,themodifiedneuro fuzzycontrollersysteminFig.10isused.Thesimulationwas runtotesttheoperatingconditionofthegasturbinewhich canbemonitoredbyobservingcombustorandtheexhaust delays plot from the scope in Simulink and the results are showninFig.14.Fromthefigure,itcouldbeobservedfor boththecombustorandtheexhaustdelays,thestepinput responsecanbeclearlyseentobeworkingwithlittleripples attheedgeswhenitisswitchingtohighvalues.
(a) (b) Fig -13:Resultsforthe(a)Turbineand(b)APU. 3.3 Simulation Results of the Fault Detection and Prediction using Neuro-Fuzzy System
Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072 2022, IRJET Impact Factor value: 7.529 | ISO 9001:2008
Certified Journal | Page1100
3.3 Comparison between Neuro Fuzzy and PID Controller Toshowtheefficacyoftheneuro fuzzycontroller,theresults obtained is compared with a conventional PID controller havingtheSimulinkdesignshowninFig.16adoptedfromthe work in [18]. The controller is used to control the speed regulator,temperatureregulator,andsurgemarginregulator asshowninthefigure.Itisincorporatedinthegasturbine model design and the simulation was run with the results showninFig.17.
Fig 15:ResultsforthedetectionoffaultusingNeuro fuzzy controller.
International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056
©
Fig 14:Resultsfortheexhaustandcombustordelaysof thegasturbineengine. ForthefaultdetectionandasshownbyFig.10,arelayis addedtodetectthefaultbyswitchingtohigh(1)whenthe anyofthe3faultsisdetectedandzerootherwise.Theresults forthefaultdetectionwithfaultandthecorrectedversion usingNeuro FuzzyisshowninFig.15.Fromtheresults,itcan be seen that an error was detected by the relay which the Neuro fuzzy quickly detect and correct the output to the closervalueoftherelay.Theerrorcanthenbemitigatedas shownfromtheresult.Inaddition,theintelligentsystemcan next predict whether in the turbine is in error by always checkingwiththetrainsignalwhichontherelaywhenitis highsignifyingthedetectionoffaultand0otherwise.
Fig 12:Resultsfortheoutputpowerandefficiencyofthe gasturbine. In addition, the turbine, and the auxiliary power unit (APU)characteristicforthegasturbineareshowninFig.13. Fromtheresults,itwasobservedthatfortheturbine,itcan be choked off at around 30% based on the operating line whilefortheAPU,itischokedataround50%.Thisindicates thattheturbineneedsimprovementinthedesign.
|
Fig 17:PIDcontrollersimulationresultsforthegas turbine. Based on the PID controller design and the simulation validation,thetwocontrollersweresimulatedtogetherfor thegasturbinefaultdetectionandtheresultsareshownin Fig.18.Fromtheresult,itcanbeseenthatintermsdetecting and correcting the error, the Neuro Fuzzy controller outperformthePIDcontroller,andthiscanbeattributedto thefacttheNeuro fuzzyhastheabilitytotrainandcorrect theerrorbasedontheprevioustrainingresults.Thus,itcan beconcludedthatusingintelligentcontrollerssuchNeuro fuzzyoffersbetterresultindetectingandpredictingfaultin gas turbine engine compared to the conventional PID controller.
4. CONCLUSIONS
[2] M.P.Boyce,GasTurbineEngineeringHandbook,4thed., Oxford:Butterworth HeinemannLtd,ELSEVIERScience &Technology,2011,1000pp.
International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056 Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page1101
Fig 18:resultsforthecomparisonbetweentheNeuro FuzzyandthePICcontrollerforthegasturbine.
In conclusion, this paper presented a novel technique of detectingandpredictingfaultsinagasturbineengineusing arobustNeuro Fuzzycontroller.Atfirst,thegasturbinewas modelled based on the components of compressor, combustion, and turbine. The mathematical model was developedbasedonthisandverifiedanalytically.Thereafter, aNeuro fuzzycontrollerandPIDcontrollerweredesigned and used for comparison. Simulation results obtained showedtheefficacyoftheNeuro Fuzzycontrolleroverthe conventional PID controller. From the results, From the results,itwasshownthatanerrorwasdetectedbytherelay which the Neuro fuzzy quickly detect and correct it. The intelligentsystemwasabletopredictwhetherintheturbine isinerrorbyalwayscheckingwiththetrainsignalwhichon therelaywhenitishighsignifyingthedetectionoffaultand 0 otherwise. In addition, the Neuro Fuzzy controller outperformedthePIDcontroller,andthiscanbeattributed tothefacttheNeuro fuzzyhastheabilitytotrainandcorrect theerrorbasedontheprevioustrainingresults.Thus,itcan beconcludedthatusingintelligentcontrollerssuchNeuro fuzzyoffersbetterresultindetectingandpredictingfaultin gas turbine engine compared to the conventional PID controller.
REFERENCES [1] K.Young,D.Lee,V.Vitali,andY.Ming,“Afaultdiagnosis methodforindustrialgasturbinesusingbayesiandata analysis,”JournalofEngineeringforGasTurbinesand Power,vol.132,ArticleID041602,2010.
Fig 16:PIDcontrollerSimulinkdesignforthegasturbine.
[3] Waleligne Molla Salilew, Zainal Ambri Abdul Karim, Aklilu Tesfamichael Baheta, Review on gas turbine condition based diagnosis method, Materials Today: Proceedings,2021.
[11] Duk,"WikimediaCommonsWebpage,"2013.[Online]. [Accessedhttp://en.wikipedia.org/wiki/File:Brayton_cycle.svg.Available:14June2013].
Bader Alhumaidi Alsubaei currentlyworksfortheMinistryof Information of Kuwait. He has received his bachelor in mechanical engineering from FIU (Florida international university) 2017followedbyMScinAdvanced Mechanical Engineering from BrunelUniversity.
[16] A.G.Memon,K.Harijan,M.A.UqailiandR.A.Memon, "Thermo Environmental and Economic Analysis of Simple and Regenerative Gas Turbine Cycles with Regression Modeling and Optimization," Energy Conversion and Management, vol. 76, pp. 852 864, 2013.
BIOGRAPHIES
Journal | Page1102
[7] SunnyAI,ZhangJ,TianGY,TangC,RafiqueW,ZhaoA, Fan M. Temperature independent defect monitoring using passive wireless RFID sensing system. IEEE SensorsJournal.2018Nov21;19(4):1525 32.
[13] H.Asgari,X.Q.ChenandR.Sainudiin,"Considerationsin ModellingandControlofGasTurbines AReview,"in the 2nd International Conference on Control, Instrumentation,andAutomation(ICCIA2011),Shiraz, Iran,2011.R.Nicole,“Titleofpaperwithonlyfirstword capitalized,”J.NameStand.Abbrev.,inpress.
Dr Ali Imam Sunny is working as an Engineering Manager and a LeadInnovatorwithintheNuclear AMRCofUniversityofSheffield.He has received his Ph.D. from NewcastleUniversityin2017and has held numeric academic and industrial positions including King’sCollegeLondon,MediWise Ltd.,NewcastleUniversity,etc. His research interest lies in Machine learning,NDTandSHM.
Volume: 09 Issue: 02 | Feb 2022 www.irjet.net p ISSN: 2395 0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified
[12] H. Cohen, G. Rogers and H. Saravanamut, Gas Turbine Theory,5thed.,PearsonEducation,2001.
[17] S.M.Hosseinalipour,E.AbdolahiandM.Razaghi,"Static and Dynamic Mathematical Modeling of a Micro Gas Turbine,"JournalofMechanics,vol.29,no.2,pp.327 335,2013.
[4] S. Sanaye, S. Hosseini, Off design performance improvement of twin shaft gas turbine by variable geometryturbineandcompressorbesidesfuelcontrol, Proc. Inst. Mech. Eng. Part A J. Power Energy 234 (7) (2020)957 980
[8] TangC,RashvandHF,TianGY,HuP,SunnyAI,WangH. Structural health monitoring with WSNs. Wireless Sensor Systems for Extreme Environments: Space, Underwater, Underground, and Industrial. 2017 Aug 7:383 408. [9] S. S. Tayarani Bathaie, Z. N. Sadough Vanini and K. Khorasani,"Faultdetectionofgasturbineenginesusing dynamic neural networks," 2012 25th IEEE Canadian Conference on Electrical and Computer Engineering (CCECE),2012 [10] G.V.BeardandM.R.Rollins,"EngineeringEconomics," in Power Plant Engineering, V. Black, Ed., Springer Science&BusinessMediaInc.,1996.
[18] S.M.Hosseini,A.Fatehi,A.K.SedighandT.A.Johansen, "Automatic Model Bank Selection in Multiple Model Identification of Gas Turbine Dynamics," Journal of Systems and Control Engineering, vol. 227, no. 5, pp. 482 494,2013.
International Research Journal of Engineering and Technology (IRJET) e ISSN: 2395 0056
[5] A.M.Y.Razak,IndustrialGasTurbines:Performanceand Operability, first ed., Cambridge, England: Woodhead PublishingLimited,2007,602pp.
[14] W. Rachtan and L. Malinowski, "An Approximate Expression for Part Load Performance of a Micro Turbine Combined Heat and Power System Heat RecoveryUnit,"Energy,vol.51,pp.146 153,2013.
[6] SunnyAI,ZhaoA,LiL,KantehSakilibaS.Low costIoT based sensor system: A case study on harsh environmentalmonitoring.Sensors.2021Jan;21(1):214.
[15] L.MalinowskiandM.Lewandowska,"AnalyticalModel BasedEnergyandExergyAnalysisofaGasMicroturbine atPart LoadOperation,"AppliedThermalEngineering, vol.57,no.1 2,pp.125 132,2013.