Soft sensing mill load in grinding process by time/frequency information fusion
融合时/频信息的磨矿过程磨机负荷软测量

Mill load (ML) is a key parameter of grinding process. Whether the status of ML and the parameters of ML can be accurately identified affects the quality and quantity of the product, and the safety of the grinding equipment. In practice, the ML status is monitored by the experience of the experienced operators. The ML parameters relate to ML and ML status directly, which is difficulty to be measured. To deal with these problems, a soft sensor strategy and an approach based on time/frequency information fusion are proposed. In this approach, at first the power spectrum of the shell vibration and acoustical signals are obtained. Then, the frequency spectrum features are selected by using adaptive genetic algorithm-partial least squares (AGA--PLS). These frequency spectrum features are fused with the current signal of the mill motor, constituting the PLS--based model for predicting the ML parameters. Finally the ML status is obtained by the ruler reasoning-based discrimination model. A grinding process experiment in the laboratory-scale ball mill validates the efficacy of the proposed soft sensor approach.