Research On Recognition Method Of Abnormal Noise Of Typical Components In Automobile Based On Acoustic Signal Analysis

Frequent abnormal noise inside the car have greatly affected the user’s driving experience.However,it is difficult to judge the source of abnormal noise subjectively,considering that there are many types of abnormal noise,the source is not fixed,and the characteristics are not obvious.Therefore,it is urgent to find an objective auxiliary algorithm to realize the rapid and effective recognition of automobile abnormal noise.Based on a research project of an enterprise,this paper selects the abnormal noises of several typical components in the car as the research objects.This subject analyzes the sound signal,and proposes a method for identifying abnormal noise of typical components in the car based on hybrid feature extraction and support vector machine(SVM)optimized by gray wolf optimization algorithm.A complete abnormal sound recognition system includes sound signal collection,preprocessing,feature parameter extraction,and classification.The main research work of the full text is as follows.Firstly,abnormal noise of typical components in the car are collected and a database of abnormal noise are established.Based on the analysis of the cause of the abnormal noise,the abnormal sound of the seat rail percussion,the abnormal sound of the seat skin friction,the abnormal sound of the seat belt retractor,the abnormal sound of the plastic parts of the glove box and the abnormal resonance of the armrest box are selected as the research object.The collection of five kinds of abnormal sound are carried out through the abnormal sound laboratory of parts and components,and the abnormal sound data set are compiled.Preliminary analysis of the abnormal noise signal characteristics shows that five abnormal noises are all typical non-stationary signals,and the energy is mainly concentrated in the range of 5000 Hz.Secondly,the preprocessing method of abnormal noise signal is studied.The focus is on the noise reduction processing method of abnormal sound signals,and a threshold adaptive wavelet packet noise reduction method is proposed.The noise reduction effect of the proposed method is analyzed by superimposing the road noise on the abnormal noise signal collected in the laboratory,using the signal-to-noise ratio and the root mean square error as evaluation indicators.Research shows that the improved wavelet packet threshold noise reduction algorithm can effectively reduce the road noise superimposed in the abnormal noise signal,and better retain the details of the original signal.Through the example analysis of normalization,pre-emphasis and frame-by-windowing processing methods,the error brought by the acquisition system is reduced and the foundation for the next feature extraction is laid.Thirdly,the feature extraction method of abnormal noise signal is researched.Short-term energy and Mel frequency cepstrum coefficients(MFCC)are extracted and analyzed from the time domain and frequency domain.In view of the problems of the standard MFCC with poor time-frequency localization performance and poor anti-noise performance,MFCC is optimized based on the Mel-scale wavelet packet decomposition method.The improved MFCC is mixed with short-term energy parameters to obtain the characteristic parameters of the input classification model.Through the analysis of the feature extraction results and the comparison of experimental simulations,it can be seen that the proposed hybrid feature parameters can well characterize the characteristics of abnormal noise signals,and have better recognition effect and anti-noise performance.Finally,based on the method of SVM optimized by GWO algorithm,the abnormal noise recognition of typical components in the car is realized.The penalty factor and kernel function parameters are optimized by the GWO algorithm,and then the GWOSVM model is constructed.The optimal kernel function type is determined by comparing the recognition performance of different kernel functions through simulation experiments.The recognition performance of different classification models are compared and analyzed to verify the effectiveness of the GWO-SVM model.The results show that the radial basis kernel function has obvious advantages both in recognition stability and recognition accuracy.Compared with other recognition models,the GWOSVM model has a greater improvement in the convergence speed and the recognition accuracy.Finally,a road test on a real vehicle has further verified the feasibility of the proposed research method applied to the identification of abnormal noise.