Soft Sensor method of mill load for grinding process based on GA-PLS from spectral data using feature selection

Mill load (ML) is a key parameter of grinding process, and whether the state of ML (low load, optimal load, over load) and the operate parameters (pulp density, material to ball mass ratio, load volume charge ratio) can be accurate identified affects the quality&quantity of the production and safety of the devices. In practice, the state of ML is monitored by the experience of the operator, and the state of the operate parameters of the mill can not be measured. A soft sensor method of the ML based on Genetic Algorithm-Partial Least Squares (GA-PLS) from spectral data of the mill shell vibration and acoustical signal using feature selection is presented. First the vibration and acoustical signals are transformed from time domain to frequency domain, and then GA-PLS is used to select the feature spectral variable respectively, and finally the selected feature spectral variable fused with the current of mill motor, three PLS1 models are developed to predict the operate parameters. A grinding process of laboratory scale experiment study shows that the proposed soft-sensor method for grinding process produces better predictive performance than traditional Principal Component Regression(PCR) and PLS method based on full spectral data.