Research On Fault Diagnosis Of Automobile Engine For The Lack Of Cylinder And Wear Of Bushing Based On Vibration Signal

With the increase of car ownership,the maintenance tasks of faulty vehicles are also increasing.Therefore,higher requirements are put forward for the fault diagnosis and maintenance of automobiles.Researches show that the engine of automobile has the higher failure rate in the whole vehicle,and the research on the fault diagnosis technology of automobile engines has become the focus of attention at home and abroad in recent years.In this paper,the main common faults of automobile engine are set as the research objects,and we have realized the complete process including signal acquisition,feature analysis as well as the determination of fault-pattern.Especially,thorough comparison and analysis are carried out about different methods of feature extraction and pattern identification.Firstly,in line with the principle of fault diagnosis based on vibration signals,the fault signal acquisition system of automobile engines is set up.The system mainly includes piezoelectric sensor,signal magnified module,data acquisition card and upper computer software.The software includes the data acquisition control program based on LabVIEW environment and the data processing program based on MATLAB.In the experiment,vibration signals of the engine under four different working conditions were observed in real time and collected through the system that had been established.Specific conditions include: normal condition of test car No.1,single-cylinder misfire fault of test car No.1,normal condition of test car No.2 and serious bearing bush abrasion of test car No.2.And eventually,both the signal acquisition parameters and the signals obtained are all saved to the computer.In order to distinguish the four different states of tested engines,it is necessary to extract features of the signals and identify the fault pattern.First of all,use the following two methods to finish the process of feature extraction: time-domain kurtosis fused with energy entropy gained by wavelet packet and energy ratio gained by empirical mode decomposition respectively to analyze the original data.Thus,eigenvectors that can reflect the features of different signals under the four working conditions were obtained.Then,establish different classification models: support vector machine(SVM),K-Nearest Neighbor(KNN)classifier and extreme learningmachine(ELM)network.The final classification results of each classification model were compared and evaluated according to the accuracy rate,training time and confusion matrix of the test results.The results show that the above classification models can identify the four different working conditions of automobile engine including normal and single-cylinder misfire fault of test car No.1 as well as normal and bearing bush abrasion of test car No.2.Both feature extraction method and fault identification algorithm are two essential factors that determine the final effect of fault diagnosis system.Judging from the classification accuracy and the training time,IMF energy ratios obtained by empirical mode decomposition combined with extreme learning machine network has better test results than the other algorithms,which possesses the highest classification accuracy and the shortest training time.