Research On The Recognition System Of Abnormal Sound In The Train Based On Deep Learning

While the train is moving,the abnormal sound inside the vehicle can be used as the information sources that can reflect the abnormal state and hidden trouble of the vehicle equipment.This subject does signal detection and signal processing for the abnormal sound,does the intelligent identification of vehicle state by the deep learning method,and combine the learning method with the portable auscultation device for the "clairaudient & wisdom mind" system.This provides an intelligent testing equipment for train,which is of great significance on the safe operation,maintenance and management of the vehicle.For the noise signal of the vehicle in different states,the subject carries on the signal sampling,signal pretreatment,and signal amplification to build the sample library of audio data.According to the characteristics of human auditory perception,MFCC is used to extract the characteristic information of the data samples of the abnormal sound.Aiming at the complex mapping relationship between train noise characteristics and vehicle state types,a model of MFCC-1DMCNN is constructed to identify and classify the fault information contained in abnormal sound.The model parameters such as MFCC order,learning rate and batch size were determined by experiments and optimization.Finally,t-SNE algorithm and confusion matrix were used to analyze the model feature extraction ability,and verify the effectiveness of the MFCC-1D-MCNN model.Aiming at the problem of less sample feature input and insufficient adaptation at low sampling frequency,a model of ESMD-MIL-1DCNN is constructed to identify and classify the fault information contained in abnormal sound.The optimal input number of IMF components is determined by experiments,and the model has good robustness.Similarly,t-SNE algorithm and confusion matrix were used to analyze the model feature extraction ability,and verify the effectiveness of the ESMD-MIL-1DCNN model.The recognition and diagnosis system of abnormal sound applied in the running-train is designed.The hardware realizes the signal acquisition,storage,processing and data transmission,and the software platform realizes the operation,recognition and diagnosis function of MFCC-1DMCNN model.