Akademik Veri Yönetim Sistemi
Engineering Applications of Artificial Intelligence, cilt.165, 2026 (SCI-Expanded, Scopus)
Operation of industrial exchangers under thermal stress and wide-span varying conditions increasingly complicates the process monitoring task. Under such circumstances, confusing effects of faults could be observed, misleading in turn the operators’ assessment. In this paper, a data-driven fault monitoring system is designed for a parallel flow thermal exchanger under consideration of wide-range operating conditions. The designed monitoring scheme consists of a data-based residual generation block, a Convolutional Neural Network (CNN)-based fault detection block and a diagnosis block, running consecutively to detect and identify faults of different types, including leakage events, temperature sensing malfunction and actuators failures. Comparable profiles of system temperatures were observed in healthy and some faulty modes, making the fault detection and diagnosis tasks rather impossible while leveraging raw sensors data. To overcome this issue, data-driven detection residuals are used together with a CNN classifier to put in evidence and identify the actual faults. The proposed monitoring scheme is tested under distinct healthy and faulty scenarios, showing superior performance over other machine learning and deep learning methods. Moreover, full satisfaction with explainable results reflecting the process physics was obtained under the critical water leakage events.