Vehicle Remote Health Monitoring and Prognostic...

Research ArticleVehicle Remote Health Monitoring andPrognostic Maintenance System

Uferah Shafi1 Asad Safi 1 Ahmad Raza Shahid1

Sheikh Ziauddin 1 and Muhammad Qaiser Saleem2

1COMSATS Institute of Information Technology Islamabad 44000 Pakistan2College of Computer Science and Information Technology Al Baha University Al Baha Saudi Arabia

Correspondence should be addressed to Asad Safi asadsafigmailcom

Received 11 August 2017 Accepted 6 December 2017 Published 18 January 2018

Academic Editor Andrea Monteriu

Copyright copy 2018 Uferah Shafi et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

In many industries inclusive of automotive vehicle industry predictive maintenance has become more important It is hard todiagnose failure in advance in the vehicle industry because of the limited availability of sensors and some of the designing exertionsHowever with the great development in automotive industry it looks feasible today to analyze sensorrsquos data along with machinelearning techniques for failure prediction In this article an approach is presented for fault prediction of four main subsystems ofvehicle fuel system ignition system exhaust system and cooling system Sensor is collected when vehicle is on the move both infaulty condition (when any failure in specific system has occurred) and in normal condition The data is transmitted to the serverwhich analyzes the data Interesting patterns are learned using four classifiers Decision Tree Support Vector Machine 119870 NearestNeighbor and Random Forest These patterns are later used to detect future failures in other vehicles which show the similarbehaviorThe approach is produced with the end goal of expanding vehicle up-time and was demonstrated on 70 vehicles of ToyotaCorolla type Accuracy comparison of all classifiers is performed on the basis of Receiver Operating Characteristics (ROC) curves

1 Introduction

Vehicle systems are complex both in hardware and softwareso their maintenance is challenging Maintenance strategybeing used in vehicle industry is normally reactive that resultsin reduction of lifetime of vehicle and also loss of moneyPredictive maintenance is required on this stage to overcomethese issues It is reported by European Commission thatthere will be 50 increment in transport vehicles within20 years [1] It will require effective strategies to keepup the vehicle performance Vehicles having very complexstructure need an effective maintenance strategy Three typesof maintenance strategies are being used in vehicle industrypredictive maintenance corrective maintenance and pre-ventive maintenance Preventive maintenance is performedafter a fault has occurred It is used for infrequent failureswhen the repair is extremely costly Preventive maintenanceis the common practice in the vehicle industry where vehicleparts are upgraded occasionally In contrast to preventive

and corrective maintenance in predictive maintenance [2]current condition of systemvehicle is analyzed to predictwhat is probably going to fail

Vehicle has a very complex mechatronic structure con-sisting of subsystems for example gearbox engine andbrakes [3] Normally any subsystem comprises electrome-chanical processes actuators and sensors The sensors andactuators are associated and controlled with an ECU (EngineControl Unit) whichmanages and screens the procedure It isadditionally associated with CAN (Controller Area Network)throughwhich the distinctive subsystems and the driver com-municate with each other A high level diagnostic protocolis needed to communicate with Engine Control Unit (ECU)Two well-known protocols are OBD2 and UDS [4] OBD2alludes to a vehiclersquos self-indicative and reporting ability On-Board Diagnostic (OBD) frameworks give the vehicle owneror a repair professional access to condition of current dataof different vehicle subsystems while UDS does provide alldetails Systemrsquos current condition is evaluated by diagnostic

HindawiJournal of Advanced TransportationVolume 2018 Article ID 8061514 10 pageshttpsdoiorg10115520188061514

2 Journal of Advanced Transportation

and prognostic processes Diagnostics is concerned withcurrent state of any subsystem whereas prognostic is relatedto the future state of subsystem [5]

There are serious challenges when we deal with prog-nostic maintenance Prognostic maintenance copes with on-board data Development cost of on-board diagnostic islimited in vehicle which results in limited number of sensorsThese sensors produce thousands of signals or data streamswhen vehicle is on the move These signals are continuouslysent to mobile or laptop attached with vehicle via wirelesscommunication It needs huge storage capacity which resultsin high cost Therefore such systems are not implementedStill there is tremendous increase in research on vehicle diag-nostic during last decade A detail discussion of predictivemaintenance of vehicle industry is presented by Prytz [1]in which comprehensive analysis of how machine learningapproaches are being used in automotive industry for faultprediction is also discussed ldquoConsensus self-organizedmod-els for fault detection (COSMO)rdquo was presented [6] in whichsensorrsquos data is used Heavy trucks and city bus were usedin experiments COSMO is presented to increase lifetimeof vehicle Another approach for vehiclersquos compressor faultprediction was reviewed [7] in which logged on-board datawas used Approach is demonstrated on Volvo trucks [8]an approach was presented to predict need for repair in aircompressor in buses and trucks Random Forest was appliedon logged on-board data for prediction

Due to increasing complexity of vehicles industryrsquosfocus has moved toward automated data analysis In themeantime with the minimal cost of wireless communicationand increasing trend of android applications it has becomefeasible to analyze on-board data for fault prediction Naryalet al [9] used android phone application on-board vehiclediagnostic system for vehiclersquos health monitoring wheredriver was notified in case of any alarming conditions

For each approach mentioned above there are manyoptions to perform diagnostic and prognostic maintenanceWe propose complete infrastructure of vehicle remote healthmonitoring and prognostic maintenance system using datadriven methodology which is based on real time data col-lected when vehicle is on move Toyota Corolla model 2010vehicles have been used for data generation Toyota Corollais series of cars developed by Corolla Corolla started tomanufacture these cars in 1966 and became best seller in 1997Remote health monitoring refers to monitoring the workingof different systems of vehicles remotely and prognostic refersto predicting fault in advance which is discussed in detailin Section 3 Our twofold contributions are fault predictionin four main systems of vehicle and providing real timevehicle monitoring and prognostic maintenance systemThispaper presents an approach using sensor data and machinelearning algorithms along with smart phone applications tocompletely analyze subsystems and provide fault predictionin subsystems One great advantage of the proposed systemcould be in latest project of vehicle industry including auto-nomous car where every system in car is automatic VMMScan help in automatic fault prediction of vehicle systems

The rest of the paper is structured as Section 2 con-tains some related work after that the proposed system

architecture is presented in Section 3 experiments are in Sec-tion 4 process of feature selection is presented in Section 5results are discussed in Section 6 and Section 7 presentsconclusion and future work

2 Related Work

Machine learning approaches are being used in vehicleindustry since last decade to improve vehicle up-time Theexisting technologies being used in vehicle industry includemachine learning approaches soft computing and on-boarddata analysis A new algorithm named ldquoSequential PatternMining Algorithmrdquo was proposed [10] in which proposedalgorithm learned patterns from warranty data of vehiclesand the learned patterns are converted to rule based expertsystem Choudhary et al presented an overview of applica-tion being used in manufacturing industry using machinelearning approaches including classification clustering andprediction [11] Then with development in manufacturing ofvehicle trendmoved fromdiagnostics of vehicles to prognos-tic using sensor data as large number of sensors were beingimplemented in vehicles Using these sensorrsquos data a newmodel for fault prediction ldquoConsensus Self-OrganizedModelfor Fault Predictionrdquo was presented Byttner et al [6] in whichmodel learned interesting relationships between sensorrsquos datain vehicles and also in small mechatronic systems Prytz etal [12] presented an unsupervised method to find relationsbetween sensorrsquos data in two rounds in first round goodrelations were found by MSE (Mean Square Error) then insecond round LASSO (Least Absolute Shrinkage and Selec-tion Operator) and least square error were used to determinemodel parameters Alzghoul et al [13] used linear regressionmethods for fault prediction rather than classification andresults showed that regression performed better than clas-sification Vehicle data bases containing vehiclersquos repairingrecords including sensorrsquos data are normally imbalanced Acomprehensive technique ldquoBRACIDrdquo (Bottom-up Inductionof Rules And Cases for Imbalanced Data) to deal withimbalanced data in order to learn rules was presented by[14] experiments and results showed that new algorithmperformed much better as compared to classifiers includingC45 PART RIPPER and CN2 in case of imbalanced dataAnother techniqueswas presented [8] to learn classifiers fromimbalanced data

Rodger [15] proposed a vehicle health maintenance sys-tem using Kalman model sensor data was used for faultprediction moreover engine abnormal behavior was alsoobserved by anomaly detection Then in the same yearanother vehicle maintenance system was proposed for diag-nostic and remote prognostic of vehicles based on leastsquares support vector machine classifier the system pro-moted the use of smart phones in automotive industryRemote maintenance of vehicle system is strongly based onvehicular communication network Lu et al [16] demon-strated vehicle ad hoc networks in detail and found param-eters for communication In preceding year [17] presented acomprehensive analysis of failure prediction of compressorSensor data was used and also challenges and problems havebeen discussed in detail Data was collected from logged

Journal of Advanced Transportation 3

vehicle data base which maintains maintenance record of allvehicles visiting Volvo workshop Random Forest KNN andC50 were used for fault prediction of compressor failureKargupta et al [18] presented a vehicle monitoring systemin their article The system monitors driver activity andthe status of the car engine Smart phones were used forcommunication between vehicle and back-end server usingwireless communication

Since traffic is getting crowded day by day accident dan-ger increases To overcome this situation intelligent systemsare being used with new technologies Turker and Kutlu[19] presented an overview of existing systems using OBD2tools and systems using communication through OBD2interface A comprehensive analysis of vehicle maintenanceused unsupervised and supervised techniques with the helpof telematics gateways which enables the vehicles to commu-nicate with back-end server [1] On-board and off-board datawas used for vehicle fault diagnostic and prognostics Thena novel data cloud service in Internet of things for vehiclewas proposed [20] in which two data mining algorithmswere used including Naive Bayes and Logistic Regression forwarranty analysis using warranty data record of vehicles

Sun presented multisensor fusion technique [21] to mon-itor vehiclersquos health using oil data and vibration signalsSchmalzel et al presented a comprehensive discussion onhow smart sensors could be used for health monitoring ofa system and to diagnose a problem [22] Another vehiclehealth monitoring system was presented by Murakami et alusing data base data distributionnetwork and a communica-tion system [23] Onemore system for healthmonitoring wasproposed by Ng et al that was demonstrated on a passengervehicle [24] System was able to detect any problem or faultin sensors or actuators

There are cost constraints when sensor data is being usedas large memory space and processor speed are requiredOne solution for data reduction [25] was presented byChoi et al using different machine learning techniques forengine Another prognostic health monitoring system [26]was presented by Cole et al using sensor data in which agentbased system was used Using on-board data another remotehealth monitoring system [27] was proposed by Zhang et alin which vehicle no start condition was analyzed in detailChen and Zhu presented complete model of vehicle healthmonitoring system [28] which was demonstrated on electricwheel truck Naryal and Kasliwal presented an in vehicleembedded system [9] for health monitoring to analyze theinternal condition of vehicles components by using travelersinformation That system was also able to predict futurefailures to avoid interruption in journey To monitor healthstate of machines and technical vehicles a remote diagnosticand monitoring system [29] was proposed by Manakov etal An application development project [30] was presentedby Ganesan and Mydhile to monitor vehicles health con-dition remotely using self-adaptive technique Ruddle et alpresented a prognostic vehiclersquos health monitoring system [2]for electric vehicles using failure analysis To increase safetyand reliability Baraldi et al presented a model for prognosticand health monitoring [31] for electric vehicles Hodge et alpresented a survey of wireless sensors network [32] in railway

industry Another android application based vehicle healthmonitoring system [33] was presented by Babu et al in whichenginersquos condition battery condition and emission systemwere monitored and driver of vehicle was notified about thecondition of mentioned systems via android phone

We propose a real time vehicle monitoring and faultprediction system In VMMS main subsystems of vehicleignition system exhaust system fuel system and coolingsystem are analyzed in detail Sensor data has been used forfault prediction usingmachine learning techniques DecisionTree SVM 119870-NN and Random Forest Data is generatedusing smart phone and OBD2 scanner when vehicle is ondrive Sensor data in the form of Diagnostic Trouble Codes(DTC) is transmitted to smart phone via Bluetooth and thensent to back-end server Classification algorithms are appliedand interesting patterns are learned which can cause anysystem to fail User is notified in case of abnormal conditionvia push notification on smart phone and email notificationUser or owner of vehicle can monitor current condition ofvehicle remotely In next section architecture of proposedVMMS is discussed in detail

3 Overview of VMMS Architecture

VMMS has three main layers as shown in Figure 1 In firstlayer data is generated An OBD2 scanner is connected withthe vehicle through OBD2 port A variety of these scannersare available in market ELM 327 Bluetooth (ELM 327) isbeing used in the proposed systemwhich is basically amicro-controllerThis scanner behaves like a bridge between vehicleand portable device that is mobile or laptop which supportsBluetooth There are tiny ICs (integrated circuits) generatedfor this communication between vehicle and portable deviceAll sensorrsquos data in form of DTC are generated when vehicleis on the move and sent to smart phone Data is continu-ously being generated and transmitted to smart phone viaBluetooth Actually smart phone communicates with ECUvia wireless connection which is used as a source to connectvehicle to back-end serveThe conceptual diagram of VMMSis shown in Figure 1

Figure 1 shows complete flow diagram of proposedVMMS In data processing layer first step is feature selectionin which data stream of DTC is filtered in feature selectionprocess using experts suggestion Then a PCA (PrincipleComponent Analysis) is applied on data set for featurereduction After that four classification algorithms are usedin classification phase including Decision Tree RandomForest 119870-NN and SVM Interesting combinations of DTCor relationships are learned and further processing is doneon server end These results are stored on server for furtherderivation which is used for fault prediction and remotemonitoring of vehicle

In remote monitoring layer the owner or concernedperson of vehicle canmonitor the current condition of vehicleremotely like fuel status speed and current position Thedriver or the owner of vehicle is notified about the failure ofany subsystem of vehicle through automatic notification

The main advantages of the proposed system are asfollows


Data generation

OBD II scanner

OBD IIinterpreter

Bluetoothinterface

Data processing

Feature selection PCA

Finding interesting pattern

ClassificationRF DT kNN SVM

DTC stream

Server

Remote monitoring

Emailnotification notification

Mobile

Figure 1 VMMS Architecture

(i) A nontechnical person can get into the failure ofany subsystem of vehicle as VMMS provides allinformation of vehicle status and condition on smartphone which is connected to vehicle

(ii) VMMS focuses on the real life solutions usingmachine learning techniques which can save moneyand time as well

(iii) Autonomous vehicle needs continuous stream ofsensory input of functioning part of vehicle VMMSprediction will guide autonomous vehicles decisionsystem from failure of any system part

(iv) Vehicle life time is increased as when ownerdriverknows all about internal conditions of systems thenhe can get some steps to get rid of any system failure

(v) Accident risk decreases with the awareness of struc-ture of systems

4 Data Generation

Vehicle four main systems are analyzed including ignitionsystem fuel system exhaust system and cooling system fordata generation Those systems are monitored by sophisti-cated computer controls Sensors provide fault information toEngine Control Unit (ECU) ECU is a microcomputer whichconsists of lot of electronic components and circuits includingmany semiconductor devices Its input device receives inputin the form of electric signals These signals come frommany sensors located at different positions in engine Itsprocessing unit compares input data with data stored inmemory The memory unit contains basic information abouthow to operate engine The output device pulses the electricsignals of the solenoid type injection valve ECU basicfunction in electric fuel injection system is to control pulserate through injector idle speed ignition timing and fuelpump The ECU adjusts quickly to change the conditionsby using programmed characteristics map stored in memory


unit Aim of ECU is to maximize engine power with lowestamount of exhaust emission and lowest fuel consumptionOBD2 scanner tools download on-board trouble codes bycommunicating with ECU to determine which sensor is nothelping (ECU) Then there is a serial bus communicationprotocol CAN (Controller Area Network) which is madeby Boch in 1980s It describes a standard for effective andtrustworthy communication between controller actuatorsensor and ECU

When vehicle is on drive OBD2 scanning tool is con-nected with vehicle and DTCs generated on run time arebeing transmitted to the smart phone ELM327 the scanningtool being used can display more than 1500 values of sensordata Toyota Corolla vehicles have been used in experimentsData of around 70 vehicles has been collected both normalcase when there is no fault and other case when faults werethere in different vehicle subsystems A stream of DTC isproduced by sensors with sampling frequency of 1Hz whenvehicle is on the move for each system under experimentEach reading is taken as one example Data set consists of150 examples DTC generated by sensors is considered as anattribute or feature The feature value is set 1 if that particularDTC is generated and set as 0 otherwiseThe output of systemor class label is also in binary form If system is on operationthen output is set as 0 which means that vehicle is in safecondition If fault occurred or system breaks down thenoutput class label is set as 1 So data set generated is completelybinary in nature

5 Feature Selection

Around 20 DTCs were produced in each reading Ninefeatures for each system have been selected carefully byexperts by mutual understanding Moreover PCA has beenapplied to prioritize the feature on the basis of variance whichis basically square of standard deviation (119904)

var (119883) = sum119899119894=1 (119883119894 minus 119883) (119883119894 minus 119883)119899 minus 1 (1)

where (2) shows standard deviation and (3) shows mean119883119904 = radicsum119899119894=1 (119883119894 minus 119883)2119899 minus 1 (2)

119883 = sum119894119894=1119883119894119899 (3)

The cut-off value of PCA is selected as 95 Since the highesteigenvalues represent the most relevant components a cut-off value 119862PCA is chosen the cut-off value 119862PCA is chosenempirically from the dataWe compared the selected featuredby experts with results of PCA Almost the same 9 featurescontained 95of variation as selected by experts rest of the 11features contained 5 of variationTherefore those 11 featureshave been excluded as they contained small informationregarding failure of a specific subsystem as shown in theFigure 2

1

10

100

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Rele

vanc

e of c

ompo

nent

s(in

per

cent

age o

f tot

al)

Components

Figure 2 Representative example of the sorted PCA eigenvalueThe119909-axis shows the values of the component as a percentage of the totalin log scale

Those classification algorithms are selected which per-form good on binary data that is with high accuracy andminimum error rate

Decision Tree Basic algorithm for Decision Tree [34] isapplied using Gini (gdi) Diversity Index (measure of nodeimpurity) for splitting criteria calculated by

(gdi) = 1 minussum119894

1199012 (119894) (4)

where 119901(119894) is probability that particular instance belongs toclass C119894 that is class 0 or class 1 Maximum number of splitsis set to 10 to restrict tree size

Support Vector Machine SVM is optimization problemclassifier [35] In binary class problem SVM draws a line thatseparates the instances of two classes and classify test instanceby maximizing the margin between both sides of line RadialBasis Function kernel (119870) is selected for similarity measure

119870(119909 1199091015840) = exp (minus120578 10038161003816100381610038161003816119909 minus 1199091015840100381610038161003816100381610038162) (5)

where is 120578 a parameter to handle nonlinear classification119870-Nearest Neighbor kNN [36] is applied whereEuclidean distance (119889)measure is used to calculate similarityor distance and number of neighbor (119896) is set to 5

119889 (119883 119884) = radic 119899sum119894=1

(119883119894 minus 119884119894)2 (6)

Random Forest [37]An ensemble learning method is appliedto improve the accuracy that is Random Forest DecisionTree algorithm is used the maximum number of learners isset to 20 Data is sampled 119899-times with replacement for eachsample a Decision Tree is learned So at the end there are119899 Decision Trees Test instance is classified on the basis ofmajority votes of learned classifiers (Decision Trees)

6 Results and Discussion

In this section all experimental results of classifiers are beingdiscussed in detail


Four classifiers are selected including Decision TreeSupport Vector Machine119870-NN and Bagging Tree (RandomForest) As the data set is binary in nature so these algo-rithms are selected as they perform good on binary dataThese algorithms are implemented on each system includingignition system exhaust system fuel system and coolingsystem The performance of each algorithm on particularsystem is evaluated on the basis of accuracy precision recalland1198651 scoremeasures 10-fold cross validation has beenusedAccuracy is the percentage of the total number of predictionsthat was found correct whereas precision is true positiveaccuracy (tpa) 1198651 score is accuracy which is measured usingprecision and recall and recall is true positive rate (tpr)According to [38 39] the precision 1198651 score accuracy andrecall are calculated as follows

Precision = TP(TP + FP)Recall = TP(TP + FN)1198651 Score = 2 lowast 119875 lowast 119877(119875 + 119877)Accuracy = TP + TN(TP + TN + FN + FP)

(7)

ROC curve is basically a graphical representation whichillustrates the performance of binary classifier It is actuallyused to compare the accuracy of different classifiers

Now detailed discussion of classification results on eachsystem is presented then the results of the proposed systemVMMS are compared with base papers [4 7]

61 Ignition System Ignition system is very important asevery system depends on it Decision Tree SVM KNN andRandom Forest (Bagging Tree) are applied All classifiers aretrained on 150 readings ROC precision 1198651 score accuracyand recall measures have been used to evaluate the classifierPerformance comparison between classifiers is performedusing ROC curve as shown in Figure 3

Performance comparison of all classifiers including Deci-sion Tree SVM 119870-NN and Random Forest is shown inFigure 3 Accuracy is measured by how much area is coveredby ROC curve As Figure 3 shows SVM covers highest areaunder the curve with 099 and Decision Tree covers lowestarea under the curve as 073119870-NNcovers 082area undercurve whereas FR covers 073 area under curve Resultsshow that SVM performs much better as compared to otherclassifiers Table 1 shows precision recall accuracy and F1measure values for each classifier

SVM beats other classifiers with 966 accuracy 094precision and 098 recall DT shows poor performance with725 accuracy 073 precision and 05 recall

62 Fuel System For failure prediction of fuel systemselected four classifiers are applied on data set includingDecision Tree SVM 119870-NN and Random Forest resultsshow that SVM performs much better as compared to otherclassifiers as shown in Table 2

True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020

RF_ROC AUC = 090SVM_ROC AUC = 099

DT_ROC AUC = 073KNN_ROC AUC = 082

Figure 3 ROC comparison (ignition system)

Table 1 Ignition system VMMS

Classifiers Precision Recall Accuracy 1198651 scoreDT 073 05 725 068SVM 094 098 966 096kNN 082 078 819 077RF 079 075 792 076

Table 2 Fuel system VMMS


SVM shows 987 accuracy with 098 precision andrecall DT shows 765 accuracy with 077 precision and 066recall In the same way119870-NN and RF performance measuresare given in Table 2

Similarly for performance comparison of these classifiersROC curve comparison is presented in Figure 4

SVM covers maximum area under curve which meansSVM shows highest accuracy where Decision Tree showspoor performance with 077 area under curve For fuelsystem 119870-NN and RF perform much better with area undercurve of 095 and 097 respectively

63 Exhaust System Performance of each classifier is judgedon the basis of accuracy precision recall and 1198651 score whichis shown in Table 3

Journal of Advanced Transportation 7Tr

ue p

ositi

ve ra

te

False positive rate

1

08

06

04

02

0

1080604020



Figure 4 ROC comparison (fuel system)

Table 3 Exhaust system VMMS


Table 4 Cooling system VMMS


SVM shows good performance with accuracy 097 119870-NN and RF show approximately the same accuracy with087 and 086 whereas DT shows 785

Accuracy comparison of all classifiers on the basis of ROCcurve is shown in Figure 5

In the case of exhaust system RF and SVM show highaccuracy with area under curve including 097 and 099119870-NN shows 090 area under curve and DT shows pooraccuracy with lowest area under curve with 076

64 Cooling System Same algorithms including DecisionTree SVM KNN and Random Forest have been appliedPrecision1198651measure recall and accuracy values for coolingsystem are shown in Table 4 which shows that SVMperformsmuch better as compared to other classifiers

In the case of cooling system SVM and 119870-NN showhighest accuracy with 966 and 946 respectively DTshows poor accuracy with 072

True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020



Figure 5 ROC comparison (exhaust system)

True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020



Figure 6 ROC comparison (cooling system)

ROC curves of each algorithm are shown in Figure 6 tocompare which classifier performs best

Figure 6 shows that SVM shows highest accuracy witharea under curve 099 whereas Decision Tree shows pooraccuracy with area under curve 077

Figures 3 4 5 and 6 show that SVMhits highest accuracyof approximately 98 whereas Decision Tree shows poor


0

20

40

60

80

100

120

Igni

tion

syste

m

Fuel

syste

m

Coo

ling

syste

m

Exha

ust s

yste

mDTSVM

KNNRF

Figure 7 Accuracy comparison

Table 5 Comparison with existing systems

Systems VMMS DT SVM kNN RFIgnition system 725 966 819 792Fuel system 765 985 946 900Exhaust system 785 98 899 886Cooling system 758 966 946 893

performance According to the survey of classifiers [40] themain reason for outstanding performance of SVM is thatdecision surface is drawn based on boundary cases SVMdoes not consider entire data SVM performs very goodon small observations In current situation data for 150experiments is collected That is why SVM achieved highestaccuracy In contrast to SVM Decision Tree considers alldata set and classify instances in step wise process A smallchange in node selection can results in big change in resultThe other flaw of Decision Tree is that it strongly depends ontraining data set so it often shows poor performance on testdata set So these are the reasons that can cause poor perfor-mance in the case of Decision Tree 119870-NN and RF neithershowed outstanding performance nor showed poor perfor-mance On average 119870-NN and RF showed good perform-ance

Now accuracy comparison with all systems being dis-cussed in this paper is presented in Table 5 which shows theaccuracy comparison of four classifiers including DecisionTreeC50 Support vector machine119870Nearest Neighbor andRandomForestBagging Tree on different systems of vehicles

The accuracy of these four classifiers is compared withproposed VMMS ignition system VMMS fuel systemVMMS exhaust system and VMMS cooling system graph-ically in Figure 7 which shows graphical interpretation ofaccuracy comparison

Graphical interpretation of Table 5 is shown in Figure 7In this section we discussed classification results of each

system of vehicle including ignition system fuel systemexhaust system and cooling system and accuracy comparison

of existing systems Next section contains conclusion andfuture work

7 Conclusion and Future Work

With the increasing trend of smart phones and wirelesscommunication it has become feasible to use these technolo-gies for real time solutions Despite limited resources thesetechnologies are being used along with machine learningapproaches to solve big problems in automotive industryA novel vehicle monitoring and fault predicting system ispresented in this paper including VMMS Four classifiersincludingDecision Tree SVM RF and119870-NNhave been usedfor fault prediction Four main systems of vehicles have beenconsidered The main objective of the proposed system is toreduce the fault frequency of systems in vehicle

Our first contribution to this paper is data generation andfeature selection Sensor data of Corolla carsmanufactured byToyota has been used We selected those distinctive sensorsthat can cause a system to break downThe second contribu-tion is to propose a user friendly vehicle fault prediction andvehicle remote health monitoring system

There are several ways to extend current results Anoticeable thing is to find the ways of refining classificationperformance Four classification algorithms have been usedDecision Tree Random Forest Support Vector Machine and119870-Nearest Neighbor The accuracy can also be discussedby applying other prediction techniques and working moreon data set All data that we have collected comes from 70vehicles of Toyota Corolla type In the future we intend towork on various types of vehicles to gather data We willcontinue our effort in this field to improve results and toexplore more systems in vehicles like CAN and so on

Acronyms

OBD On-board diagnosticDTC Diagnostic Trouble CodeROC Receiver Operating CharacteristicsAUC Area under the curveECU Engine Control UnitCAN Controller Area NetworkPCA Principle Component AnalysisDT Decision TreeSVM Support Vector Machine119870-NN 119870-Nearest NeighborRF Random ForestNA Not availableTP True positiveTN True negativeFP False positiveFN False negativeP PrecisionR Recall

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper


Acknowledgments

Special thanks are due to Mr Amjad Rafiq (MechanicalEngineer Toyota Islamabad) since a part of this work issupported by him

References

[1] R Prytz Machine Learning Methods for Vehicle PredictiveMaintenance Using Off-Board And On-Board Data DissertationHalmstad University Press 2014

[2] A R Ruddle A Galarza B Sedano I Unanue I Ibarra andL Low ldquoSafety and failure analysis of electrical powertrainfor fully electric vehicles and the development of a prognostichealth monitoring systemrdquo in Proceedings of the IET Hybridand Electric Vehicles Conference 2013 HEVC 2013 pp 1ndash6 IETLondon UK November 2013

[3] J Jeong N Kim W Lim Y Park S W Cha and M E JangldquoOptimization of power management among an engine batteryand ultra-capacitor for a series HEV a dynamic programmingapplicationrdquo International Journal of Automotive Technologyvol 18 no 5 pp 891ndash900 2017

[4] M A Taie M Diab and M Elhelw ldquoRemote prognosis diag-nosis and maintenance for automotive architecture based onleast squares support vector machine and multiple classifiersrdquoin Proceedings of the 2012 4th International Congress on UltraModern Telecommunications and Control Systems ICUMT 2012pp 128ndash134 Russia October 2012

[5] G Shi P Dong H Q Sun Y Liu and Y X Cheng ldquoAdaptivecontrol of the shifting process in automatic transmissionsrdquoInternational Journal of Automotive Technology vol 18 no 1 pp179ndash194 2017

[6] S Byttner T Rognvaldsson and M Svensson ldquoConsensus self-organized models for fault detection (COSMO)rdquo EngineeringApplications of Artificial Intelligence vol 24 no 5 pp 833ndash8392011

[7] R Prytz S Nowaczyk T Rognvaldsson and S Byttner ldquoAnal-ysis of truck compressor failures based on logged vehicle datardquoin Proceedings of the International Conference on Data Mining(DMIN) p 1 2013 The Steering Committee of The WorldCongress in Computer Science Computer Engineering andApplied Computing (WorldComp)

[8] R Prytz S Nowaczyk T Rognvaldsson and S Byttner ldquoPre-dicting the need for vehicle compressor repairs using mainte-nance records and logged vehicle datardquoEngineeringApplicationsof Artificial Intelligence vol 41 pp 139ndash150 2015

[9] E Naryal and P Kasliwal ldquoReal time vehicle health monitoringand driver information display system based on CAN andAndroidrdquo International Journal of Advance Foundation andResearch in Computer vol 1 no 11 pp 76ndash84 2014

[10] J Buddhakulsomsiri and A Zakarian ldquoSequential patternmining algorithm for automotive warranty datardquo Computers ampIndustrial Engineering vol 57 no 1 pp 137ndash147 2009

[11] A K Choudhary J A Harding andMK Tiwari ldquoDataminingin manufacturing a review based on the kind of knowledgerdquoJournal of Intelligent Manufacturing vol 20 no 5 pp 501ndash5212009

[12] R Prytz S Nowaczyk and S Byttner ldquoTowards relation discov-ery for diagnosticsrdquo in Proceedings of the 1st InternationalWork-shop on DataMining for Service andMaintenance KDD4Service2011mdashHeld in Conjunction with SIGKDDrsquo11 pp 23ndash27 USAAugust 2011

[13] A Alzghoul M Lofstrand and B Backe ldquoData stream fore-casting for system fault predictionrdquo Computers amp IndustrialEngineering vol 62 no 4 pp 972ndash978 2012

[14] K Napierala and J Stefanowski ldquoBRACID a comprehensiveapproach to learning rules from imbalanced datardquo Journal ofIntelligent Information Systems vol 39 no 2 pp 335ndash373 2012

[15] J A Rodger ldquoToward reducing failure risk in an integratedvehicle health maintenance system a fuzzy multi-sensor datafusion Kalman filter approach for IVHMSrdquo Expert Systems withApplications vol 39 no 10 pp 9821ndash9836 2012

[16] W Lu L D Han and C R Cherry ldquoEvaluation of vehicularcommunication networks in a car sharing systemrdquo InternationalJournal of Intelligent Transportation Systems Research vol 11 no3 pp 113ndash119 2013

[17] J Stefanowski ldquoOverlapping rare examples and class decom-position in learning classifiers from imbalanced datardquo SmartInnovation Systems and Technologies vol 13 pp 277ndash306 2013Emerging Paradigms in Machine Learning

[18] H Kargupta R Bhargava K Liu et al ldquoVEDAS a mobileand distributed data streammining system for real-time vehiclemonitoringrdquo in Proceedings of the 2004 SIAM InternationalConference onDataMining pp 300ndash311 Philadelphia PAUSA2004 Society for Industrial and Applied Mathematics

[19] G F Turker and A Kutlu ldquoEvaluation of car accident preven-tion systems through onboard diagnosticrdquo Global Journal onTechnology vol 7 pp 01ndash06 2015

[20] W He G Yan and L D Xu ldquoDeveloping vehicular data cloudservices in the IoT environmentrdquo IEEE Transactions on Indus-trial Informatics vol 10 no 2 pp 1587ndash1595 2014

[21] Q Sun ldquoSensor fusion for vehicle health monitoring anddegradation detectionrdquo in Proceedings of the 5th InternationalConference on Information Fusion FUSION2002 pp 1422ndash1427USA July 2002

[22] J Schmalzel F Figueroa J Morris S Mandayam and RPolikar ldquoAn architecture for intelligent systems based on smartsensorsrdquo IEEE Transactions on Instrumentation and Measure-ment vol 54 no 4 pp 1612ndash1616 2005

[23] T Murakami T Saigo Y Ohkura Y Okawa and T TaninagaldquoDevelopment of vehicle health monitoring system (VHMSWebCARE) for large-sized construction machinerdquo KomatsursquosTechnical Report vol 48 no 150 pp 15ndash21 2002

[24] H K Ng R H Chen and J L Speyer ldquoA vehicle health moni-toring system evaluated experimentally on a passenger vehiclerdquoIEEE Transactions on Control Systems Technology vol 14 no 5pp 854ndash870 2006

[25] K Choi J Luo K R Pattipati S M Namburu L Qiao and SChigusa ldquoData reduction techniques for intelligent fault diag-nosis in automotive systemsrdquo in Proceedings of the 2006 IEEEAUTOTESTCON - IEEE Systems Readiness Technology Confer-ence pp 66ndash72 USA September 2006

[26] I S Cole P A Corrigan W Ganther et al ldquoDevelopment ofa sensor-based learning approach to prognostics in intelligentvehicle health monitoringrdquo in Proceedings of the 2008 Interna-tional Conference on Prognostics and HealthManagement PHM2008 USA October 2008

[27] Y Zhang M Salman H S Subramania et al ldquoRemote vehiclestate of health monitoring and its application to vehicle no-startpredictionrdquo in Proceedings of the IEEE AUTOTESTCON 2009- Systems Readiness Technology Conference Mission AssuranceThrough Advanced ATE pp 88ndash93 USA September 2009

[28] S Chen and M Zhu ldquoDesign and realization of health moni-toring system on electric wheel truckrdquo in Proceedings of the 2010


International Symposium on Intelligence Information Processingand Trusted Computing IPTC 2010 pp 51ndash54 China October2010

[29] A L Manakov A A Igumnov and S A Kolarzh ldquoMonitoringtechnical state of transportation vehicles and productionmach-inesrdquo Journal ofMining Science vol 49 no 4 pp 630ndash636 2013

[30] R Ganesan and S Mydhile ldquoDesign and development ofremote vehicle health monitoring system using context awareweb servicesrdquo in Proceedings of the 2013 IEEE Conference onInformation and Communication Technologies ICT 2013 pp760ndash764 ind April 2013

[31] P Baraldi F Di Maio M Rigamonti et al ldquoA Procedure forPractical Prog- nostics and Health Monitoring of Fully ElectricVehicles for Enhanced Safety and Reliabilityrdquo in Proceedingsof the 5th IET Hybrid and Electric Vehicles Conference (HEVC2014) London UK 2014

[32] V J Hodge S OrsquoKeefe M Weeks and A Moulds ldquoWirelesssensor networks for condition monitoring in the railway indus-try a surveyrdquo IEEE Transactions on Intelligent TransportationSystems vol 16 no 3 pp 1088ndash1106 2015

[33] M S Babu T A Anirudh and P Jayashree ldquoFuzzy systembased vehicle health monitoring and performance calibrationrdquoin Proceedings of the 2016 International Conference on ElectricalElectronics and Optimization Techniques ICEEOT 2016 pp2549ndash2554 India March 2016

[34] L Breiman J H Friedman R A Olshen andC J StoneClassi-fication and Regression Trees Wadsworth Belmont Mass USA1984

[35] N Cristianini and J Shawe-Taylor An Introduction to SupportVector Machines Cambridge University Press CambridgeUK2000

[36] I Mani and I Zhang ldquokNN approach to unbalanced datadistributions a case study involving information extractionrdquo inProceedings ofWorkshop on Learning from Imbalanced Datasetsp 126 2003

[37] L Breiman ldquoRandom forestsrdquoMachine Learning vol 45 no 1pp 5ndash32 2001

[38] DM Powers ldquoEvaluation fromprecision recall andF-measureto ROC informednessrdquoMarkedness and Correlation 2011

[39] S Russell andPNorvig ldquoAmodern approachrdquo inArtificial Intel-ligence vol 25 Prentice Hall Englewood Cliffs NJ USA 1995

[40] S B Kotsiantis I D Zaharakis and P E Pintelas ldquoMachinelearning a review of classification and combining techniquesrdquoArtificial Intelligence Review vol 26 no 3 pp 159ndash190 2006

International Journal of

AerospaceEngineeringHindawiwwwhindawicom Volume 2018

RoboticsJournal of

Hindawiwwwhindawicom Volume 2018


Active and Passive Electronic Components

VLSI Design



Shock and Vibration


Civil EngineeringAdvances in

Acoustics and VibrationAdvances in



Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawiwwwhindawicom

Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Control Scienceand Engineering

Journal of



Journal ofEngineeringVolume 2018

SensorsJournal of



RotatingMachinery


Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018


Chemical EngineeringInternational Journal of Antennas and

Propagation




Navigation and Observation


Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia

Submit your manuscripts atwwwhindawicom


and prognostic processes Diagnostics is concerned withcurrent state of any subsystem whereas prognostic is relatedto the future state of subsystem [5]

There are serious challenges when we deal with prog-nostic maintenance Prognostic maintenance copes with on-board data Development cost of on-board diagnostic islimited in vehicle which results in limited number of sensorsThese sensors produce thousands of signals or data streamswhen vehicle is on the move These signals are continuouslysent to mobile or laptop attached with vehicle via wirelesscommunication It needs huge storage capacity which resultsin high cost Therefore such systems are not implementedStill there is tremendous increase in research on vehicle diag-nostic during last decade A detail discussion of predictivemaintenance of vehicle industry is presented by Prytz [1]in which comprehensive analysis of how machine learningapproaches are being used in automotive industry for faultprediction is also discussed ldquoConsensus self-organizedmod-els for fault detection (COSMO)rdquo was presented [6] in whichsensorrsquos data is used Heavy trucks and city bus were usedin experiments COSMO is presented to increase lifetimeof vehicle Another approach for vehiclersquos compressor faultprediction was reviewed [7] in which logged on-board datawas used Approach is demonstrated on Volvo trucks [8]an approach was presented to predict need for repair in aircompressor in buses and trucks Random Forest was appliedon logged on-board data for prediction

Due to increasing complexity of vehicles industryrsquosfocus has moved toward automated data analysis In themeantime with the minimal cost of wireless communicationand increasing trend of android applications it has becomefeasible to analyze on-board data for fault prediction Naryalet al [9] used android phone application on-board vehiclediagnostic system for vehiclersquos health monitoring wheredriver was notified in case of any alarming conditions

For each approach mentioned above there are manyoptions to perform diagnostic and prognostic maintenanceWe propose complete infrastructure of vehicle remote healthmonitoring and prognostic maintenance system using datadriven methodology which is based on real time data col-lected when vehicle is on move Toyota Corolla model 2010vehicles have been used for data generation Toyota Corollais series of cars developed by Corolla Corolla started tomanufacture these cars in 1966 and became best seller in 1997Remote health monitoring refers to monitoring the workingof different systems of vehicles remotely and prognostic refersto predicting fault in advance which is discussed in detailin Section 3 Our twofold contributions are fault predictionin four main systems of vehicle and providing real timevehicle monitoring and prognostic maintenance systemThispaper presents an approach using sensor data and machinelearning algorithms along with smart phone applications tocompletely analyze subsystems and provide fault predictionin subsystems One great advantage of the proposed systemcould be in latest project of vehicle industry including auto-nomous car where every system in car is automatic VMMScan help in automatic fault prediction of vehicle systems

The rest of the paper is structured as Section 2 con-tains some related work after that the proposed system

architecture is presented in Section 3 experiments are in Sec-tion 4 process of feature selection is presented in Section 5results are discussed in Section 6 and Section 7 presentsconclusion and future work

2 Related Work

Machine learning approaches are being used in vehicleindustry since last decade to improve vehicle up-time Theexisting technologies being used in vehicle industry includemachine learning approaches soft computing and on-boarddata analysis A new algorithm named ldquoSequential PatternMining Algorithmrdquo was proposed [10] in which proposedalgorithm learned patterns from warranty data of vehiclesand the learned patterns are converted to rule based expertsystem Choudhary et al presented an overview of applica-tion being used in manufacturing industry using machinelearning approaches including classification clustering andprediction [11] Then with development in manufacturing ofvehicle trendmoved fromdiagnostics of vehicles to prognos-tic using sensor data as large number of sensors were beingimplemented in vehicles Using these sensorrsquos data a newmodel for fault prediction ldquoConsensus Self-OrganizedModelfor Fault Predictionrdquo was presented Byttner et al [6] in whichmodel learned interesting relationships between sensorrsquos datain vehicles and also in small mechatronic systems Prytz etal [12] presented an unsupervised method to find relationsbetween sensorrsquos data in two rounds in first round goodrelations were found by MSE (Mean Square Error) then insecond round LASSO (Least Absolute Shrinkage and Selec-tion Operator) and least square error were used to determinemodel parameters Alzghoul et al [13] used linear regressionmethods for fault prediction rather than classification andresults showed that regression performed better than clas-sification Vehicle data bases containing vehiclersquos repairingrecords including sensorrsquos data are normally imbalanced Acomprehensive technique ldquoBRACIDrdquo (Bottom-up Inductionof Rules And Cases for Imbalanced Data) to deal withimbalanced data in order to learn rules was presented by[14] experiments and results showed that new algorithmperformed much better as compared to classifiers includingC45 PART RIPPER and CN2 in case of imbalanced dataAnother techniqueswas presented [8] to learn classifiers fromimbalanced data

Rodger [15] proposed a vehicle health maintenance sys-tem using Kalman model sensor data was used for faultprediction moreover engine abnormal behavior was alsoobserved by anomaly detection Then in the same yearanother vehicle maintenance system was proposed for diag-nostic and remote prognostic of vehicles based on leastsquares support vector machine classifier the system pro-moted the use of smart phones in automotive industryRemote maintenance of vehicle system is strongly based onvehicular communication network Lu et al [16] demon-strated vehicle ad hoc networks in detail and found param-eters for communication In preceding year [17] presented acomprehensive analysis of failure prediction of compressorSensor data was used and also challenges and problems havebeen discussed in detail Data was collected from logged














Data generation

OBD II scanner

OBD IIinterpreter

Bluetoothinterface

Data processing




DTC stream

Server

Remote monitoring


Mobile







4 Data Generation





5 Feature Selection




119883 = sum119894119894=1119883119894119899 (3)


1

10

100

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Rele

vanc

e of c

ompo

nent

s(in

per

cent

age o

f tot

al)

Components





1199012 (119894) (4)



119870(119909 1199091015840) = exp (minus120578 10038161003816100381610038161003816119909 minus 1199091015840100381610038161003816100381610038162) (5)


119889 (119883 119884) = radic 119899sum119894=1

(119883119894 minus 119884119894)2 (6)







(7)







True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020













ue p

ositi

ve ra

te

False positive rate

1

08

06

04

02

0

1080604020













True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020




True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020








0

20

40

60

80

100

120

Igni

tion

syste

m

Fuel

syste

m

Coo

ling

syste

m

Exha

ust s

yste

mDTSVM

KNNRF














Acronyms





Acknowledgments


References













































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia















Data generation

OBD II scanner

OBD IIinterpreter

Bluetoothinterface

Data processing




DTC stream

Server

Remote monitoring


Mobile







4 Data Generation





5 Feature Selection




119883 = sum119894119894=1119883119894119899 (3)


1

10

100

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Rele

vanc

e of c

ompo

nent

s(in

per

cent

age o

f tot

al)

Components





1199012 (119894) (4)



119870(119909 1199091015840) = exp (minus120578 10038161003816100381610038161003816119909 minus 1199091015840100381610038161003816100381610038162) (5)


119889 (119883 119884) = radic 119899sum119894=1

(119883119894 minus 119884119894)2 (6)







(7)







True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020













ue p

ositi

ve ra

te

False positive rate

1

08

06

04

02

0

1080604020













True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020




True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020








0

20

40

60

80

100

120

Igni

tion

syste

m

Fuel

syste

m

Coo

ling

syste

m

Exha

ust s

yste

mDTSVM

KNNRF














Acronyms





Acknowledgments


References













































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia





5 Feature Selection




119883 = sum119894119894=1119883119894119899 (3)


1

10

100

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20

Rele

vanc

e of c

ompo

nent

s(in

per

cent

age o

f tot

al)

Components





1199012 (119894) (4)



119870(119909 1199091015840) = exp (minus120578 10038161003816100381610038161003816119909 minus 1199091015840100381610038161003816100381610038162) (5)


119889 (119883 119884) = radic 119899sum119894=1

(119883119894 minus 119884119894)2 (6)







(7)







True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020













ue p

ositi

ve ra

te

False positive rate

1

08

06

04

02

0

1080604020













True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020




True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020








0

20

40

60

80

100

120

Igni

tion

syste

m

Fuel

syste

m

Coo

ling

syste

m

Exha

ust s

yste

mDTSVM

KNNRF














Acronyms





Acknowledgments


References













































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia





(7)







True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020













ue p

ositi

ve ra

te

False positive rate

1

08

06

04

02

0

1080604020













True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020




True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020








0

20

40

60

80

100

120

Igni

tion

syste

m

Fuel

syste

m

Coo

ling

syste

m

Exha

ust s

yste

mDTSVM

KNNRF














Acronyms





Acknowledgments


References













































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



ue p

ositi

ve ra

te

False positive rate

1

08

06

04

02

0

1080604020













True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020




True

pos

itive

rate

False positive rate

1

08

06

04

02

0

1080604020








0

20

40

60

80

100

120

Igni

tion

syste

m

Fuel

syste

m

Coo

ling

syste

m

Exha

ust s

yste

mDTSVM

KNNRF














Acronyms





Acknowledgments


References













































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



0

20

40

60

80

100

120

Igni

tion

syste

m

Fuel

syste

m

Coo

ling

syste

m

Exha

ust s

yste

mDTSVM

KNNRF














Acronyms





Acknowledgments


References













































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia



Acknowledgments


References













































RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia


















RoboticsJournal of




VLSI Design



Shock and Vibration







Journal of



Volume 2018



Volume 2018


Journal of




SensorsJournal of



RotatingMachinery





Propagation






Hindawi


Advances in

Multimedia


Vehicle Remote Health Monitoring and Prognostic...

Documents

Transcript of Vehicle Remote Health Monitoring and Prognostic...