Inverse fuzzy fault models for fault isolation and severity estimation in industrial pneumatic valves

Abstract

Fault detection is crucial in the chemical industry for identifying process problems, and determining the nature of the fault is essential for scheduling maintenance. This study focuses on the application of inverse fuzzy models to reconstruct faults for the purpose of detection, isolation, and classification. By inverting fuzzy models, the fault signal can be reconstructed, enabling identification of the fault source and its characteristics. To address the issue of undetected small abrupt faults, the wavelet transform is employed. This approach allows for the detection of incipient faults, while the classification is achieved by evaluating the response of the fault reconstruction. Fault isolation is accomplished by comparing the reconstructed faults. However, in the case of the pneumatic valve utilized, four out of the 19 simulated faults demonstrated poor isolation due to the similarity of their reconstructions using inverse fuzzy models. A comparison with similar applications in existing literature is also presented.