Mechanical Frequency Spectrum Sub-band Feature Estimation Based on Multiple Types Correlation Analysis

Mill load is one of the key factors that affects the operation optimization and control of grinding process. It is difficult to detect accurately in real time. The mill load relates to multiple channel mechanical signals, such as the mill shell vibration signals, the shaft seat front and rear vibration signals, the acoustic signals near to the shell surface and under the grinding area, etc. How to evaluate these sub-band feature of the mechanical spectrum of the above channels is a difficult problem. It can help to clarify the ball mill grinding mechanism and the mechanical signal generation mechanism. To solve these problems, this paper proposes a multi-channel mechanical frequency spectrum sub-band feature evaluation method based on multiple types correlation analysis. Firstly, the multi-channel mechanical signals are transformed from time domain to frequency domain to obtain the frequency spectrum data, which are divided into several sub-bands to extract features. Then, the normalized correlation coefficients between the multi-channel sub-band feature and the mill load are calculated. Thirdly, based on the partial least square algorithm, the variable projection importance (VIP) value of sub-band features are also calculated. The VIPs are normalized same as that of correlation coefficient. Finally, a new index is calculated by using normalized correlation coefficient and VIP value and the model's prediction performance. These sub-band features of single channel and all channel mechanical signals are measured based on the new defined combined evaluation index. The multi-channel mechanical signals of a laboratory-scale ball mill are used to verify the effectiveness of the proposed method.