Research On Vibration Signal Analysis And Fault Diagnosis Of Engine

In recent years,with the rapid development of engine technology,the structure of engine system becomes more complex,and traditional experience diagnosis method has been difficult to diagnose engine faults accurately.Domestic automobile manufacturers are facing the challenge of developing accurate and efficient engine fault diagnosis technology.This paper focuses on three types of typical fault problems of a automobile engine of a vehicle manufacturer.The purpose is to apply a scientific method to expound the connection between engine fault and vibration signal.This paper propose an engine fault diagnosis method for automobiles based on vibration signal analysis and machine learning,and form a meaningful case for similar problems.By researching engine misfire fault,engine valve group fault and engine flywheel group fault,the cause of the engine fault and the connection between the engine fault and engine vibration are confirmed,and the transmission path of engine vibration is analyzed.Then the analysis methods of engine vibration signal are introduced,including time domain analysis,frequency spectrum analysis,Variational Mode Decomposition(VMD)and signal parameter analysis.A engine vibration signal measurement experiment was designed and implemented,and it collected 160 sets of engine vibration signals in normal conditions and three fault conditions.VMD was used to process 160 sets of collected vibration signals,and the high-noise signals in VMD were filtered out.The other signals were taken for signal reconstruction,and the reconstructed signals were used for signal analysis.Then parameters of the reconstructed signals were calculated and analyzed,and it was found that rectified average value,peak-to-peak value,root mean square value,kurtosis,reconstructed signal fractal dimension,reconstructed signal fuzzy entropy,IMF1 center frequency and IMF5 fuzzy entropy in four different working conditions have the highest degree of discrimination.Using the above 8signal parameters to form the state feature vectors,and the state feature vectors were preprocessed and recorded.For the research of fault diagnosis method which based on machine learning,and the theories of Particle Swarm Optimization(PSO)and Least Square Support Vector Machine(LSSVM)are introduced firstly.Secondly,this paper establish an engine fault diagnosis model based on particle swarm optimization and least square support vector machine.Then the fault diagnosis simulation experiment is implemented,and the engine fault diagnosis model is trained and tested by inputting the set of state feature vectors.The test results show that the fault diagnosis method has a diagnosis accuracy of 97.5%,and its mean absolute percentage error(MAPE)is 1.25%,and its diagnosis time is 5.7s.It proved that the engine fault diagnosis method is feasible.Using the same set of state feature vectors to train and test the BP neural network fault diagnosis model and the random hyper-parameters least square support vector machine fault diagnosis model.The test result shows that their diagnostic accuracy is 95.0% and 90.0%,and their MAPE is 1.46% and 6.88%,and their diagnostic time is 3.6s and 2.5s,respectively.It can prove that the proposed fault diagnosis method has relative superiority by comparing fault diagnosis performance of different fault diagnosis models.