Automobile Engine Monitoring And Fault Diagnosis Based On Data-driven

In modern times，In addition to enjoying the convenience brought by the car, the car’seconomy, comfort, safety and environmental protection are higher demanded than before. As theheart of the car, the engine is the most central parts of the car, moreover, the quality of the car isdetermined by the performance of the engine. Due to the development of electrical andelectronic technology, cybernetics and artificial intelligence, the engine is constituted bycontrollers, actuators, sensors and others, what’s more, all of these parts are formed by themechanical, hydraulic, electrical and electronic mixture of complex systems. Although complexsystem greatly improves the car’s power, comfort and economy efficiency, the reliability of thesystem is reduced because of the complex combination of institutions. The chain reaction offault propagation is also followed. The above goals are truly achieved owing to theimplementation of the OBD (On-Board Diagnosis). The engine running reasonably withworking conditions is ensured, and harmful gas emissions and the fuel consumption aresubstantially reduced to meet emission standards. In view of the above two points, the study ofEFI engine fault diagnosis is a very meaningful thing.After reading a lot of literature and viewing the existing technical information about the car,the engine fault diagnosis is mainly based on the model and data in theory. The approach, whichthe actual sensor signal is compared with pre-tested good theory MAP, is adopted to proceed thefault diagnosis in actual. The complex modeling and big disturbing effect exist in themodel-based approach, a lot of problems in the actual online application are also being. Thedata-based approach has now considered as one of the most hot research direction, becausemultiple sensor information can be made full use of by this approach, and it can be moreintelligent,achieved more convenient and easy than the other methods.Based on the results of the above analysis and previous studies, a complete method ofengine fault diagnosis which is based on data-driven is proposed. Its nature is that the faultdiagnosis is realised by classification based on data characteristics. Fault feature extraction, dataclassification algorithms and fault prediction mechanism are contained in this paper as the threekey points. The data-based diagnosis method is based on the fault feature extraction, and thesuccess of the diagnosis method is determined by whether the features are effective or not. Themost important of the three sensors in the engine and a reconstruction estimation model ofcylinder pressure are used as the source of the data in this article, furthermore, the multiple time-domain characteristics and the crucial engine parameters are extracted as the fault featureset. Because data redundancy of these sensors feature set is adverse to the latter classification,the PCA (Principal Component Analysis) is put to use to reduce the dimensionality of thefeature set. The classification of the sample data is achieved through SVM classificationalgorithm to realise fault diagnosis, and the fault detection, location and identification are cometrue by the accurate sample data classification. That SVM is the best algorithm in the existingdata classification algorithms is is proved by a lot of simulation, whose generalization abilityand robustness are superior to other methods. The throttle and rotational speed signal are usedfor cross-correlation analysis by the failure prediction, what’s more, the severity of failure aremeasured by the size of the correlation coefficient. Because the throttle is deemed as the inputsignal of the engine closed-loop system, speed is also regarded as an output, so the anyexceptions of engine are reactive on the speed. A system of the engine fault diagnosis mode,which is the most important elements of this article, is constituted by the above three parts. Theproposed method is proved to be feasible and effective through the software simulation results,which can be used as the engine on-line fault diagnosis method.