Fault isolation during transient conditions on a heated two-tank system: a multiway PCA approach

Abstract

This study aimed to employ a Multiway Principal Component Analysis (MPCA) approach for fault detection and isolation during transient conditions on a benchmark heated two-tank system, which closely resembles a chemical process. The significance lies in the system’s similarity to chemical processes, characterised by controllable variables, noise, disturbances, and changes in operating modes. The data generated by the system can include both faulty system data and data under Normal Operating Conditions (NOC). Building on prior studies that focused on fault detection and isolation during steady state conditions, this research employed the MPCA approach to enhance the ability to detect and isolate faults during transient conditions. For the model’s training, essential to Fault Detection and Isolation (FDI), training data are generated from both a simulated model of the system and the practical system. The simulated model’s data are then validated against the practical system data. Following the model’s training, the MPCA approach, validated through relevant literature, is applied to detect and isolate faults using test data. This test data are generated from both the simulated model and the practical system. Performance metrics, including detection rate, false alarm rate, time until detection, and contribution plots, are employed to evaluate the effectiveness of the MPCA method. Results indicate that the MPCA method successfully detected and isolated faults using the SPE and T2 control charts for most faults, provided the right minimum number of experimental and simulated training runs under NOC. Due to extended run times in practical experiments, a combination of practical and simulated data were utilized to achieve effective fault detection and isolation. In conclusion, the research affirms the feasibility of detecting and isolating faults in the benchmark heated two-tank system using the MPCA method during transient conditions.