Research On Sound Quality Evaluation System For Nonstationary Vehicle Interior Noise Based On SVM

With the rapid development of social economy and the continuous improvement ofpeople’s living standards, vehicle has become an essential means of transportation in ourdaily life. The followed sound quality of vehicle interior noise has attracted more and moreattention with the development of automobile industry. To create an auditory-perception-based sound quality evaluation (SQE) system for quickly and accurately evaluating soundquality of non-stationary vehicle interior noises has become a research focus in the field ofvehicle NVH. On the basis of previous studies, a sound quality evaluation system based onsupport vector machine (SVM) aiming at both the stationary and non-stationary vehicleinterior noises is developed in this thesis.Firstly, some interior noise measurements according to GB and ISO standards for fourdomestic cars belonging to different brands is conducted with PULSE noise test system.Four kinds of stationary and non-stationary working conditions are included in this carinterior noise data collection experiment: idle speed, constant speed, accelerating andbraking. The collected noise signals are preprocessed and used for establishment of vehicleinterior noise database. Some objective evaluation parameters, such as A-weighted soundpressure level, loudness, sharpness, roughness, fluctuation strength and articulation indexof the processed sample data are calculated and used as system inputs. A jury test withsemantic differential method is conducted to the measured noise samples and the resultsare used as outputs. With the input and output data, finally, a SVM-based SQE model isestablished. Furthermore, a SQE system for vehicle interior noise is developed by GUIDEmodule of software MATLAB. The unused noise samples in the modeling processes areimported into the SQE system to verify its effectiveness by comparing the predicted andthe real values.The above studies suggest that, SVM method has strong nonlinear fitting ability and can be applied in SQE of non-stationary vehicle interior noise; the improved grid searchmethod can ensure search accuracy and reduce the search amount, thus improve the searchefficiency. Comparing the predicted results and the subjective evaluation values of theunused noise samples, one may conclude that the SVM-based SQE model is verified tohave a higher prediction accuracy than the multiple linear regression method and leastsquare support vector machine (LSSVM) method, which confirms the effectiveness andsuperiority of the proposed SVM model. The SVM-based SQE system for non-stationaryvehicle interior noise developed in this thesis can be directly applied in SQE of vehicleinterior noise, which is beneficial for acoustical design and improvement in the process ofvehicle engineering development.