Research On Psychoacoustic Parameters Of Automobile Exhaust Noise

With people’s continuous requirements for improving the riding comfort,psychoacoustics is gradually applied to vehicle sound quality analysis.The research abroad laid the theoretical foundation of psychoacoustics,while few researches were down inland.Many researchers and automakers tend to use the existing commercial software for vehicle sound quality prediction.However,software users rarely know the principles of psychoacoustic models due to commercial confidentiality,which result in inconvenience and difficulties to correctly evaluate the vehicle sound quality.This thesis firstly introduces some basic theories of psychoacoustics and the evaluation indicators for exhaust sound quality,then the assessment index definitions are given for physical acoustics and psychoacoustics,as well as the loudness,sharpness,roughness and fluctuation as the main psychoacoustic metrics on the basis of selection principles.Secondly,an intensive study has been spent on the main psychoacoustic assessment models and a program been coded to perform the calculation,including Zwicker steady and time-varying loudness model,Moore steady and time-varying loudness model,Arues sharpness model,Zwicker sharpness model and Daniel & Weber roughness model.Based on Arues roughness model,a new fluctuation model is then proposed,using the external and middle ear transfer function according to ANSI S3.4-2007,and 75 filter channels are used to calculate specific fluctuations on the scale of ERB.The new model simulates ear filtering characteristics more accurately.Thereafter,exhaust noise samples are acquired for one prototype vehicle and four competing vehicles under idling,steady speed cruise and full acceleration state,as well as the library of sound samples from the development of exhaust system.The annoyance is divided into 10 levels in semantic differential method and 23 evaluation subjects have been chosen for subjective evaluation.The calculation results have been examined by correlation coefficient method to eliminate the incorrect results.The test results show that the most of average Spearman coefficients were higher than 0.7.The well-selected samples are verified to be usable for annoyance assessment of exhaust noises.Finally,according to the results from subjective and objective evaluations,BP neural network models are established to evaluate the annoyance of steady-state and unsteady exhaust noise,where the LS algorithm is selected as the network training algorithm,the logsig function as the neuron transfer function by cut-and-trial method.The node numbers are also determined for hidden layer.The calculation results show that the prediction errors of the models are respectively less than 10% and 15% with good stability which means it can be used for predicting sound quality.In summary,this thesis applies the main psychoacoustic objective parameters to evaluate the annoyance of exhaust noises thus quantificate subjective sound quality characteristics,and to some extent it eliminates the inconvenience where the subjective evaluation can not be repeated.