علي نكونام

The massive volume of data received from gas turbines and issues related to data storage, the unavailability of measurements in hot sections, and the presence of measurement noise are among the fundamental challenges in developing a fault diagnosis system for gas turbines. Therefore, to detect the gas path faults, a system must be developed that, above all, has the capability to confront noise, can operate with a minimum number of measurements, and also has the capability of retraining with limited datasets. To achieve these goals, in this research, a type of neural network known as the Deep Residual Compensation Extreme Learning Machine (DRCELM) has been employed for the first time to detect fouling and erosion. Then, in the next stage, it has been enhanced to Deep Residual Compensation Online Sequential Extreme Learning Machine (DRC-OSELM) with incremental learning capability. To implement the proposed idea, initially, the IGT25 gas turbine under study was thermodynamically modeled using the T-mats toolbox. Subsequently, performance degradation under fouling and erosion faults was simulated, considering data noise. In another part of the research, the optimal parameters for setting up the fault detection system were selected using the proposed VLGA-ELM algorithm. Subsequent to establishing fault diagnosis systems based on compensatory extreme learning machines and their deep models, and comparing them with conventional approaches, it was revealed that the compensatory extreme learning machine-based fault detection system outperforms other methods. It operates more robustly in the presence of data noise, and its learning speed is higher compared to non-compensatory extreme learning machine-based methods. The research findings indicate that the model achieves high accuracy of over 97% in noisy data. Additionally, an F1 score exceeding 98% was obtained, and the average false negative rate (FNR) and false positive rate (FPR) were both less than 2.5%. Finally, after conducting the eva‎luations, the developed fault detection system was integrated within an Internet of Things framework using the ThingSpeak platform.

سيستم تشخيص و شناسايي عيب , عيوب مسير گاز , توربين گاز صنعتي , ماشين يادگيري حدي با مكانيزم جبرانساز عميق , ماشين يادگيري حدي برخط با مكانيزم جبران ساز عميق , اينترنت اشيا

Fault detection and identification system , Gas path faults , Industrial gas turbine , Deep residual compensation extreme learning machine , Deep residual compensation online sequential extreme learning machine , Internet of things

Ali Nekoonam

Morteza Montazeri-Gh