Feature Extraction in Ball Mill Pulverizing System Based on Multi-sensor Data Fusion

The output capability of ball mill pulverizing system mainly represents the efficiency of system. In order to measure output capability, a novel feature extraction method based on multi-sensor data fusion is proposed. We use fast ICA algorithm to extract independent components from the field data, which includes six sensor measurements. The extracted components are mutually independent in statistic sense and assumed to be corresponded to certain physical processes. We propose a calculation method of reference value, which compares the variation of each independent component when the input variable changed. By the calculation of reference value, we are able to select feature vectors from the independent components and determine their sign. Experiment results show that the extracted feature vectors can represent the output capability well and give a good physical interpretation.