变速器啸叫声品质的RBF神经网络预测与权重分析

Prediction of the whine noise quality of transmission and its weight analysis based on RBF neural network

以变速器啸叫为研究对象,提出使用径向基(RBF)神经网络的方法来确定变速器声品质评价中客观评价参数对主观评价结果的影响权重。采集变速器不同位置的声音信号作为试验样本,用等级评分法对111个样本信号进行了主观评价试验,同时计算11个声品质客观评价参数。以客观评价参数计算结果为神经网络输入,声品质主观评价结果为输出,引入径向基神经网络建立了变速器声品质预测模型。以预测模型为基础,利用各网络层间连接权值,计算变速器声品质客观评价参数对主观评价结果的影响权重。研究结果表明:变速器啸叫声品质主要受SIL-4、总响度和随时间响度三个客观参量的影响。

A method for calculating the weight to consider the impact of objective psychoacoustic metrics on subjective evaluation results was proposed by using RBF neural network. Gear whine signals were collected at different locations and used as the evaluating samples. Subjective sound quality evaluation testings for 111 noise samples were conducted. Meanwhile, eleven sound quality objective parameters were calculated. By using the objective parameters as inputs and the subjective values as outputs, a RBF neural network was adopted to establish a model for gear whine sound quality prediction. The network connection coefficients of the prediction model were used to calculate the impact weight of objective parameters on the results of subjective evaluation. The calculation results show that the SIL-4, the sharpness and loudness over time are three key psychoacoustic parameters to affect the gear whine sound quality. © 2017, Editorial Office of Journal of Vibration and Shock. All right reserved.