Condition Assessment of AC Contactors Using Optical Fiber Sensors and Deep Learning

Electromechanical contactors are core equipment of the electrical circuits environment, operating mainly in industrial areas for load control. As they have a lifespan and are susceptible to faults, predictive maintenance based on measurement systems is crucial to avoid accidents and profit loss. This paper presents a method of instrumentation and analysis of AC contactors under different operational conditions. Optical Fiber Bragg Grating (FBG) sensors were used as sensor elements. The experimental setup relied on two ABB AX40 AC contactors, each instrumented using a single FBG sensor externally. The devices were subjected to three switching conditions: the contactor with no faults, worn load contact, and end-life load contact. The events were submitted to different variability conditions to better generalize the system for field application. Continuous Wavelet Transform (CWT) generated scalograms from the acquired signals to be submitted to different Convolution Neural Networks (CNNs) models. The data set was split between the training, CNN validation, and repeatability test groups. This third group is a set of conditions not trained with the classifier to verify the generalization. The results presented in this paper show a high F1 score of around 0.9 for the CNN architectures in the training group and validation groups. However the repeatability test group had lower results, being around 0.3. This paper shows the high applicability of FBG as single sensor element and CNN to monitoring and classifying faults in electrical switching devices, with further studies aiming for field application.