Experimental Platform for Feature Selection of Signal of Ball Mill

Ball mill load monitoring and rational parameters setting are important to ensure the ball mill long-term stable operation. Although vibration and acoustic signal of shell contain plenty of information about mill load, it is difficult to select the feature of them in time domain. Due to the high dimensionality and colinearity, models based on frequency spectrum are complex and with a low generalization and the irrelevant spectral variables deteriorate their quality. This paper use Synergy Interval Partial Least-Squares Regression(SiPLS) to select the feature frequency bands of vibration and acoustical, which are directly relevant to the parameters of ball mill load, and build effective prediction models. The experimental platform combines the strengths of MATLAB with the benefits of C#.net to implement the functions of frequency feature selection, feature modeling and load parameters prediction. Test results show that the platform selects the frequency spectrum feature effectively, and the generalization of mill load models are improved.