基于样本熵与 ELM-Adaboost 的悬架减振器异响声品质预测

摘要

车辆悬架减振器异响严重削弱了车内声品质，针对该异响问题设计并开展了不同路况条件下的整车道路试验，对采集到的车内噪声信号分别计算 A 计权声压级与心理声学客观参量（响度、尖锐度、语音清晰度、抖动度和粗糙度）以提取减振器异响特征信息，并将其与主观评价进行了相关分析。另一方面，再引入小波包分解与样本熵的概念，对减振器异响特征信息进行了有效地提取，同时提出基于 Adaboost 的极限学习机（ELM）算法，建立了减振器异响声品质预测改进模型，并将其与支持向量机（Support Vector Machine，SVM）、广义神经网络（Generalized Regression Neural Network， GRNN）进行对比。研究结果表明：传统的 A 计权声压与心理声学指标不能有效地用于减振器异响声品质预测而结合小波包样本熵提取的异响特征与 ELM-Adaboost 算法能有效地对减振器异响声品质进行预测，并且效果优于 SVM与 GRNN。%The abnormal noise from an automobile suspension shock absorber weakens its interior sound metric seriously.Aiming at this problem,a complete automobile road test was conducted on different road surfaces to collect shock absorber noises and calculate the A-weighted sound pressure level and the psycho-acoustic sound metrics of the interior noise,such as,loudness,sharpness,articulation index,fluctuation strength and roughness in order to investigate the correlation between these objective parameters and subjective evaluation.On the other hand,the concepts of wavelet packet decomposition and sample entropy were introduced to extract the characteristics of the abnormal noise of the shock absorber.An improved model of ELM-Adaboost was built to predict the sound metric of the shock absorber's abnormal noise.This algorithm was compared with the support vector machine (SVM)and the generalized regression neural network (GRNN).The results showed that the traditional A-weighted sound pressure level and the psychoacoustic indices cannot be used to evaluate the shock absorber's abnormal noise sound metric effectively,but the proposed model combining wavelet packet,sample entropy and ELM-Adaboost algorithm can predict the sound metric of shock absorber noise efficiently,its root-mean-square error is lower than those of SVMand GRNN.