Complex Multimode Process Monitoring Based on the Improved Canonical Variable Analysis

Complex industrial processes often switch among multiple operation modes, and the multimode data often show the characteristics of multi-center and obvious difference in variance. However, most multivariate statistical process monitoring methods assume the unimodal distribution of process data and cannot perform well in multimode process monitoring. Therefore, a novel process monitoring method using local neighborhood standardization strategy based on improved canonical variate analysis is proposed in this paper. Firstly, the local neighborhood standardization strategy is applied to transform the multimode data into unimodal data. Then, we adopt canonical variate analysis to extract features and obtain the projection matrix. Finally, Kullback-Leibler divergence is used to enlarge the difference between fault feature distribution and normal one. The advantages and validity of the proposed scheme are verified on the Tennessee Eastman process.