Residual Signal-based Process Monitoring of Industrial Systems

Process monitoring is necessary to ensure the reliability and safety of the industrial systems for a long-term. In this paper, a residual signal is proposed for process monitoring. The data that represent normal operation are collected off-line, then they are used to build the monitoring approach. The proposed monitoring approach extracts the information stored in the last singular vector of the normal data covariance matrix. The system faults are first detected and then isolated by removing the crosstalk among system variables. The efficiency of the proposed approach is measured by two indices, false alarm rate (type I error) and missed fault detection rate (type II error). The proposed monitoring approach is applied to a chemical system, Continuous Stirred Tank Reactor (CSTR) which is popular as a benchmark for process motioning and control design problems. The results show the ability of the proposed approach to process monitoring in terms of reducing the false alarm and missed fault detection rates.