LAUSR

Abstract The exergy-based fault detection method has not yet been applied to a complex industrial system that adequately represents a dynamically changing process. One such system, the Tennessee Eastman process, is commonly used as a benchmark for fault detection methods. In this paper, an exergy-based fault detection approach is applied to the Tennessee Eastman process. This is done to investigate the feasibility of using this approach when confronted with noisy sensor data and control loops masking faulty behaviour. An exergy characterisation was performed on stream data obtained from the Tennessee Eastman process. The exergy characterisation included a new approach to calculate the standard chemical exergy of unknown components. For fault detection, threshold limits were determined for the exergy characterisation when normal operating conditions are assumed. The threshold limits were calculated following the upper and lower control limit determination of the Shewhart control chart. The results showed that this method could quantify both the physical state as well as the chemical features of the process and that 17 out of the 20 considered faults could be detected. This shows that the exergy-based method could be adequately applied to the Tennessee Eastman process.