Efficient Fault Isolation Method to Monitor Industrial Batch Processes

Abstract

Industrial batch processes are very popular manufacturing system with large number of process variables involved. Monitoring of batch processes using statistical process monitoring becomes very difficult in view of the complex correlations between the process variables. This paper focuses on a fault isolation based process monitoring method without prior information of fault where fault isolation problem is converted into a variable selection. Variable selection is a learning algorithm used here to solve the problem of selection and isolation of variables from a model. The method discussed here uses a sparse coefficient based dissimilarity analysis algorithm known as Sparse Dissimilarity Algorithm(SDISSIM) which checks a calculated D-index for identifying fault in the process. A sparse coefficient is tabulated to verify the process variables contributing to the fault and an absolute variance difference is calculated to select the variables for fault isolation. Finally SDISSIM method is explained by successful implementation in MATLAB with real time industrial process data.