关键词: 磨机负荷(ML), 功率谱密度(PSD), 主元分析(PCA), 最小二乘支持向量机(LSSVM)

Abstract: It is difficult to detect the mill load (ML) efficiently in the ore milling process of a ball mill and the mill is often underloading, thus causing the difficulty in implementing the optimal control with energy saving and cost reduction. To solve the problems, how the vibration of mill shell is generated is analyzed, as well as the power spectrum density (PSD) of the vibration signals under different milling conditions and the correlation between the PSD and ML parameters within different frequency bands. Then, the way to model the soft sensoring is proposed introducing the three ML parameters, i.e., the mineral to ball volume ratio in a mill, pulp density and charge volume ratio. With the vibration acceleration signals of the mill shell in time domain transformed into frequency domain by fast Fourier transform (FFT), the principal component analysis (PCA) is used for dimensionality reduction and extraction of the principal components (PCs) of characteristic spectra within the low, medium and high frequency bands separately. Finally, the least square support vector machines (LSSVM) are used to implement the mapping between characteristic spectral PCs and ML parameters. Experimental results showed that the way by FFT-PCA-LSSVM can extract the spectral characteristics efficiently with higher prediction accuracy than other conventional methods.