A 3-Layered Nonlinear Process Monitoring Strategy with a Novel Fault Diagnosis Approach

Abstract

Abstract The article proposes the development of a layered process monitoring strategy based on Multi- Block Kernel Principal Component Analysis (MBKPCA). MBKPCA aids in the development of a distributed process monitoring strategy by taking into account the nonlinear relationships existing amongst the measured characteristics. A distributed process monitoring strategy stratifies the proposed process into a multi-layered structure comprising of blocks, sub-blocks etc. In this article an MBKPCA based monitoring strategy was devised for a Wire Rod Manufacturing Facility (WRMF) of an Integrated Steel Plant (ISP). The proposed monitoring strategy stratified the entire process into 3 layers, with the first layer comprising the manufacturing stages, the next layer comprising the sub-stages and the third layer comprising the characteristics to be monitored within the respective sub-stages. The detection of the fault was carried out with the aid of Kernel Principal Component Analysis (KPCA) score based Hotelling T2 chart. Fault detection was followed by Fault Diagnosis, for which new Fault Diagnostic Statistics were proposed which took into account the contribution of the main and the auxiliary characteristics. The study also proposed the concept of Cumulative Percent Contribution Ratio (CPCR) to limit the number of parameters (stages/sub-stages/characteristics) that needs to be retained in fault diagnosis.