Research On The Door Sound Quality Prediction Of Passenger Vehicles Based On SVM

With the development of the automobile industry,people are not satisfied with the traditional performances only,such as safety,fuel economy and handling stability of automobiles,but have begun to pay more and more attention to their vibration,noise and harshness(NVH).It is regarded as the door closing sound of H in NVH as one of the first car sound feelings that the user is in contact with,which directly affects the user's perception of the car's texture,and therefore attracts much attention.Therefore,it is necessary to strictly control the sound pressure level and improve its sound quality.This paper takes the closing sound quality of passenger cars as the research object,and uses a combination of subjective and objective evaluation methods to establish a genetic algorithm-based support vector machine(SVM)sound quality prediction model.By comparing the closing sound before and after,the analysis of the poor sound quality door and the prediction of sound quality were carried out.The main research work is as follows:First,subjective and objective evaluation tests were performed on the closing sounds of 22 different brands of passenger cars of different grades.22 passenger cars were selected,and the closing sound samples were collected in a semi-anechoic room.The closing speeds were respectively 1.0m/s,1.2m/s,and 1.4m/s.66 effective closing sound samples were obtained.In HEAD Artemis,the objective quantities such as loudness,sharpness,roughness,language intelligibility,and A sound level are calculated,and subjective evaluation tests are performed using the pairwise comparison method with preference as an index.Correlation analysis of subjective and objective quantities using SPSS shows that only loudness,sharpness,roughness,and A weighted SPL are related to preference.Secondly,a SVM sound quality prediction model based on genetic algorithm is established.Support vector regression machine(SVR)is used to establish the relationship model between the subjective and objective parameters.To improve model accuracy.The genetic algorithm is used to optimize the parameters.The libsvm toolbox is loaded in MATLAB.Loudness,sharpness,roughness,and A weighted SPL are used as inputs,and preference is used as output,training and testing the subjective and objective results.The squared correlation coefficient R~2 of the training result was0.9399,which indicates that the prediction accuracy is high.The squared correlation coefficient of the test result was 0.8832,which indicates that generalization ability of the prediction model is good.Then,the bench-marking analysis and improvement of a passenger car door were carried out.The frequency spectrum analysis based on wavelet decomposition was used to analyze the test data of the door to be improved and the standard door.The improvement direction of the door was determined to increase the A-level and roughness of the door closing sound,reduce its loudness and sharpness The objective parameters of the door are extracted,and the results show that:A sound level and roughness increase,and the loudness and sharpness decrease.The results show that only the muffler volume in the 500Hz~1000Hz and over 2000Hz bands somewhat reduced,the other bands have basically reached the improvement target,which shows that the muffler effect of the improved muffler is pretty good.Finally,the SVR prediction model based on genetic algorithm was used to predict the sound quality of the improved passenger car.In order to verify whether the sound quality of the door of the improved car is improved,the model is used to predict it.The results show that the sound quality is improved under the three closing speed conditions,and the preference value increases the most at 1.2m/s by 14.18,the preference value increased the least at 1.4m/s,by only 8.08,indicating that the improved structure did improve the quality of the closing sound.At the same time,subjective evaluation tests were carried out on the improved structure.The comparison of the predicted and experimental values showed that the squared correlation coefficient R~2=0.969,and the prediction error was less than 5%at all three closing speeds,indicating that the Closing sound quality improved after the improvement.The research in this paper shows that the SVR sound quality prediction model based on genetic algorithm can predict the closing sound quality of passenger cars well,indicating that the prediction model has engineering application value.