Fault Detection of Equipment Based on Weighted JS Divergence Partitioning Strategy
There are two main problems in the partitioning principal component analysis(PCA)fault monitoring method.One is that it is often difficult to determine the optimal variable block partition threshold.The other is that PCA fault monitoring is mainly aimed at data subject to normal distribution,while the actual industrial data often cannot fully meet the requirements of normal distribution.To address these issues,a fault detection method based on weighted JS divergence partitioning strategy is proposed.The normality of process variables is determined by observing their distribution patterns,and the data are divided in-to normal and non-normal variables.To solve the problem of difficult partition of variable blocks,a weighted JS divergence par-titioning strategy is adopted,which expands variables into normal weighted and non-normal weighted blocks.PCA and inde-pendent compent analysis(ICA)detection models are then established to monitor normal and non-normal blocks.After obtai-ning the detection results of each block,Bayesian fusion inference method is used to fuse monitoring results to obtain the global fault monitoring results.The effectiveness and feasibility of this method are verified through the application of fault data on the silk production equipment of a certain cigarette factory.
万方数据
维普
