Integrated local Fisher discriminant analysis based fault classification

The actual industrial process data is often companied with complex local characteristics, which is not conducive to the extraction of sample features and the improvement of fault classification accuracy. To solve this problem, an integrated local Fisher discriminant analysis(ILFDA) model is proposed in this paper, which can mine the local structure characteristics of data from variable and sample dimensions simultaneously, thus fault classification accuracy is improved and the difficulty of modeling is reduced. Firstly, the complex system is partitioned based on the structure principle, so that the local information of data can be obtained from the variable dimension efficiently and the influence of irrelevant variables is excluded. Secondly, as for the local information from sample dimension, local Fisher discriminating analysis( LFDA) classification model is established in each sub-block, and corresponding weights are assigned to local models, so as to measure the influence of different sub-blocks on current fault more accurately. Finally, the classification performance weighting strategy is used to fuse the classification results in each sub-block. The simulation results on Tennessee Eastman (TE) process show that the proposed ILFDA method has better fault classification performance. © 2021, Editorial Department of Control Theory & Applications South China University of Technology. All right reserved.