Research Of Motor Vehicle Engine Fault Diagnosis Algorithm And System Realization Based On Acoustic Signal

With the continuous increase in the number of motor vehicles nationwide,driving has become the most important way for people to go out.In order to travel safely,it is significant to diagnos motor vehicle engine failures.However,traditional fault diagnosis methods not only require professional prior knowledge,but also have low diagnosis efficiency.With the increasing application of deep learning in image,speech recognition and other fields,it provides a new idea for automatic diagnosis of motor vehicle engine faults.A fault diagnosis algorithm based on vehicle engine acoustic signal is studied,which uses deep convolutional networks.At the same time,a fault diagnosis system integrating collection,transmission and analysis is designed and implemented.The main research work and results are as follows:(1)A motor vehicle engine acoustic signal data set was constructed.The data set included the acoustic signals of normal and fault engines of 101 large-scale transport vehicles.The normal engine data included acoustic signals collected under two conditions of idling and acceleration.The data of the fault engine includes acoustic signals collected under two conditions of one cylinder is not working and two cylinders are not working.After pre-emphasis,framing,and windowing are performed on the acoustic signal,the Mel-scale Frequency Cepstral Coefficients(MFCC)feature of the acoustic signal is extracted through the Mel cepstrum transform;at the same time,in order to verify the network’s Robustness,through cutting,adding noise and tuning to enhance the acoustic signal data.(2)A motor vehicle engine fault diagnosis algorithm based on gated double convolutional neural network is presented.In order to extract the distinguishing characteristics of the engine acoustic signal fault in the Mel spectrogram and accurately realize the fault recognition,a gated double convolutional network is designed on the basis of Gated CNN.By adding a convolution module and a gating unit to control the feature information transmission process,The time-frequency structure information in the engine acoustic signal characteristics can be better retained,thereby effectively alleviating the over-fitting phenomenon and improving the recognition accuracy.Using the constructed motor vehicle engine acoustic signal data set for experiments,the recognition accuracy of normal and fault acoustic signals is up to 99.9%;the accuracy of identifying fault types under idling and acceleration is over 90%.By adding noise and audio tuning to the data set to identify the fault type,a higher recognition rate can still be obtained,which shows that the algorithm has good robustness.(3)Based on the proposed algorithm,a motor vehicle engine fault detection system based on acoustic signals is designed and implemented.The system includes collection terminal,database management platform,deep learning model and fault diagnosis module.The collection terminal collects the acoustic signal of the motor vehicle engine,transmits it to the server on the one hand,and extracts MFCC features on the other,uses the loaded deep learning model to diagnose engine faults,and displays the acoustic signal time-domain diagram,Frequency domain graph,Mel spectrum graph and recognition result.The function of the server-side database management platform is to perform basic operations such as adding,searching,and deleting collected data,and at the same time,the accepted data will be uniformly named for management.The deep learning model and the fault diagnosis module are divided into two parts.On the one hand,the deep learning model is constructed,and the received acoustic signal data is used for model training,and then the model is downloaded to the collection terminal;on the other hand,the real-time received acoustic signal is diagnosed through the deep learning model after training on the server side.