Fault Detection and Isolation Based on Deep Learning for a Fresnel Collector Field

Abstract

With the advancement of new technologies, power systems are increasingly equipped with more sensors and actuators, heightening the risk of failure. This fact, together with the vulnerability of solar plants -not only to internal faults but also to the action of the sun, rain, wind, and animals, among others- gives rise to the need for detecting and identifying faults to deal with them. Methods that detect and diagnose faults play a crucial role in solar plants, allowing the systems to cope with them as soon as they occur and before they lead to large-scale problems. This work proposes using neural networks to detect and distinguish mirror and flow rate faults in a Fresnel plant. In addition, a defocusing stage is added to access hard-to-isolate faults, increasing the accuracy of 89.61% to 97.43%. These results contribute to the problem of isolability in thermal solar plants. The simulations for obtaining the neural networks and the results were conducted on a model of the Fresnel plant located at the Engineering School of Seville, Spain (ETSI).