一种基于心理声学品质和调制频率的车窗电机检测方法
A method of window motor detection based on psychological acoustic？？quality and modulation frequency

为保证车窗电机出厂前声品质和性能达到要求，提出了基于心理声学参量和调制频率的车窗电机检测方法。基于传统的心理学客观参量模型计算正常电机和故障电机声音样本的响度、粗糙度、尖锐度，基于修正的模型计算其尖锐度值，并通过主客观评价试验分析正常电机和故障电机的响度、粗糙度、尖锐度、修正后尖锐度与主观感受反映的相关性；以响度、修正后尖锐度作为特征向量，将电机分为正常电机和异常噪声电机，在此基础上为了诊断异常噪声电机的故障类型，加入物理参量调制频率作为预测车窗电机故障类型的特征量；最后，构建附加动量法优化的BP神经网络分类器对电机进行分类，通过试验验证优化的神经网络分类器。研究结果表明：正常电机与故障电机的响度和修正后尖锐度值存在明显差别，响度和修正后尖锐度与人的主观心理一致性较好，一致性系数达0.8以上；碳刷？不幌蚱魅毕莸牡缁？噪声频率在80~100 Hz，蜗杆？渤萋秩毕莸牡缁？噪声频率在20~40 Hz，而正常电机的噪声频率在100 Hz以上，调制频率可作为检测电机故障类型的特征量；优化的神经网络分类器对车窗电机的分类准确率达90%以上，且与传统BP神经网络分类器相比其准确率更高和耗时更少。
In order to ensure the sound quality and performance of the window motor before leaving the factory, the method of window motor detection based on the psychoacoustic parameters and modulation frequency was put forward. Based on the traditional psychology objective model of, the loudness, roughness and sharpness of the sound samples of the normal motor and fault motor were calculated. The sharpness was calculated based on the modified model，and the correlation between the loudness, roughness, sharpness, corrected sharpness and subjective perception of the normal motor and fault motor was analyzed through subjective and objective evaluation experiments. The motor was divided into normal motor and abnormal noise motor by using the loudness and corrected sharpness as feature vectors. On this basis, the modulation frequency of physical parameters was added as the characteristic quantity to predict the fault type of the vehicle window motor in order to diagnose the fault type of the abnormal noise motor. Finally, the BP neural network classifier optimized by the additional momentum method was constructed to classify the motor, and the optimized neural network classifier was verified by experiments. The results show that there are obvious differences between the loudness and the corrected sharpness of the normal motor and fault motor. The loudness and corrected sharpness have good consistency with people？？s subjective psychology, and the consistency coefficient is more than 0.8. The frequency of motor noise of carbon brush？？commutator defect is 80 to 100 Hz, the motor noise of worm？？gear defect is 20 to 40 Hz, while the noise of normal motor is above 100 Hz. The modulation frequency can be used as the characteristic quantity to detect the fault type of motor. The classification accuracy of the optimized neural network classifier for vehicle window motor is more than 90%, and it has higher accuracy and less time consumption compared with the traditional BP neural network classifier. 5 tabs, 10 figs, 20 refs