Engine Fault Detection Based On Deep Learning

Various sounds are produced during the operation of the car.Maintenance technicians diagnose faults based on the sound signals generated during the operation of the car.This traditional initial auscultation judgment method will produce greater results due to the subjective judgment of the technicians.The diagnosis gap.The thesis takes the normal sound signal and abnormal sound signal of the car as the research object,converts the sound signal into spectrogram,signal graph and spectrogram,and classifies and recognizes the type of car fault based on deep learning,compared with traditional manual detection Method and fault classification based on acoustic characteristics analysis.This method has a relatively better classification effect.While improving the accuracy of detection,analysis and diagnosis,it can also make automobile fault detection more intuitive and quantitative.The experimental research data in this article is a large number of sound signals.Because the collected car sound signals contain noise,these sound signals need to be denoised.Wavelet threshold denoising will filter out some useful sound data signals,and also The quality of the sound data signal after denoising is reduced to a certain extent,so the noise reduction method adopts a Wiener-wavelet threshold denoising method.Wiener filtering has the characteristic of minimum mean square error,and most of the noise in the sound data signal The Wiener filter can be effectively eliminated,and then the final sound signal is obtained under the effect of the wavelet threshold secondary denoising.This noise reduction processing method can complete the noise reduction effect on the sound data signal,and on the other hand,it can also ensure the sound The data signal has a certain integrity.For the classification and identification of different abnormal sounds produced by car failures,it is first necessary to select the common car failure types.Through the investigation and statistics of different types of multiple car failure types in the car repair shop,the failure types to be studied are selected respectively.It is the abnormal sound at the front of the engine,the abnormal sound of the valve lifter and the abnormal sound of the oil pan.Then the VGG-16 convolutional neural network is used to classify and recognize the spectrogram,signal graph and spectrogram of the sound segment.At the same time,the support vector machine method is used to classify the abnormal sound of the car according to the acoustic characteristics of the abnormal sound of the car.The experimental results show that: in the classification and recognition of different abnormal sounds produced by car faults,the classification effect produced by the use of convolutional neural networks will be relatively better,which shows that the introduction of convolutional nerves in the classification and recognition of abnormal sounds of car faults The feasibility of the network has opened up a new path for the classification and identification of abnormal noises of automobile faults.