Abstract: A method for detecting the raw material level in ball-mills based on audio signal processing was presented. By using multi-scale wavelet transform and the multi-channel decomposition technique based on wavelet packet in audio signal processing, the feature frequency bands which can reflect the height and changes of the level of the raw material in ball-mills were identified and extracted. Then the audio signal energy in above-mentioned frequency bands was treated as a characteristic quantity which can reflect the changes of the material level. According to the audio signal energy of lower feature frequency of different material levels, the curve-fitting was done and the curve of characteristic quantity of the audio signal related to the material level was obtained. Thus, the non-linear mapping between the characteristic quantity of the audio signal and the material level was realized.

Key words: vibration and wave, ball mill, audio signal, wavelet analysis, curve fitting

摘要： 将音频信号分析用于球磨机料位检测。利用多尺度小波变换和基于小波包的多频道分解技术，从识别和提取出能反映料位高低及变化的特征频率段，以该频率段的信号能量作为反映料位变化的特征量。根据不同料位下特征频率段的音频信号能量进行曲线拟合，得到音频信号特征量与料位的关系曲线，实现音频信号特征量和料位之间的非线性映像。

关键词: 振动与波, 球磨机, 音频信号, 小波分析, 曲线拟合