Study On Hybrid Evaluation Model For Vehicle Noise Quality Based On Auditory Perception

Nowadays, people put forward an increasing demand of vehicle comfort. As one ofthe significant factors to comfort, vehicle noise has been paid more attention. Soundquality evaluation (SQE) of vehicle noise has become a very active research direction. Forsubjective SQE, it needs higher requirements on evaluators, the progresses are complex,and the results generally inconsistent. Though program of loudness, the most importantindex of objective SQE, has been organized by ISO, it cannot completely reflect people’sauditory perception. This thesis combined the finite element analysis (FEA) method andArtificial Neural Network (ANN) in SQE study and finally built a hybrid qualityevaluation model for vehicle noises.Firstly, models of external ear canal, tympanic membrane and middle ear are builtwith point clouds obtained by scanning sample ear. After models are integrated and meshed,the materials and boundary conditions are given to different components respectively. TheFEA model is completed successfully. By comparing the simulated results of displacementin umbilical eardrum and stapes foot with that of references, the built FEA model isverified. Vehicle noises as the stimuli of the FEA model, the displacement in stapes foot arecalculated by direct frequency response analysis. Secondly, some psychoacoustics indices,such as A-weighted sound pressure level (SPL), loudness and sharpness, are calculated byprogramming. The Graphical User Interface (GUI) for calculation is also programmed. Bycomparing the results calculated by software and program, the correctness of the programwritten by self is verified. With the program, A-weighted SPL, loudness and sharpness ofthe measured vehicle noises are calculated in range from the first specific frequency bandto the thirteenth specific frequency bank. Finally, based on one-third-octave filter bank, thefilter bank with filtering characteristics of human’s ear is designed. And with it, thedisplacements’ energy features are extracted. The filter bank’s output as the input, the psychoacoustic indices’ values as the output, a Radial Basis Function (RBF) neuralnetwork model is built. So far, the hybrid evaluation model is built successfully. Themodel’s error of predition is in reasonable range.The presented hybrid evaluation model in this thesis is an attempt for SQE method,which may be regard as a reference for building whole human ear model to study auditoryperception and SQE method.