Sistema óptico para detecção de anomalias em dispositivos eletromecânicos de comutação

Abstract

Electromechanical switching devices are fundamental for the protection and control of electrical power systems, where failures can generate catastrophic consequences. This study proposes a novel approach to evaluate the operational status of switching equipment using optical sensors based on Bragg gratings (FBG). Compared to conventional systems for measuring vibration, the presented method is less susceptible to electromagnetic interference, making it promising for applications in aggressive environments, such as those found inside electrical machines. Four experimental configurations, which increased in complexity during the study’s development, were tested to validate the methodology. Firstly, activation parameters were manipulated in an extra-low voltage relay, enabling the identification of imposed conditions by analyzing the signals’ characteristics. In the second part, the evaluation methods were improved by using the wavelet scattering transform (WST) as the main algorithm for extracting information and the support vector machine (SVM) for predicting different degradation levels in a contactor’s internal components typically used in industrial plants. The selection of more relevant and less redundant features facilitated the evaluation of class proximity, achieving a maximum accuracy of 99.4%. However, when considering only additional measurements obtained after reassembling the contactor, performance dropped to 86.4%. The third stage assessed the feasibility of extrapolating the classification models to new datasets obtained by similar contactors and sensors. At this stage, transfer learning was applied to convolutional neural networks (CNN), and the WST algorithm with SVM was used for comparison. The results indicated that the variabilities caused by changing the contactor or sensor rendered generalization through supervised learning methods unfeasible. In such cases, anomaly detection algorithms are proposed as alternatives. Finally, in the last experimental setup, sensors were installed on the fixed contacts of a voltage regulator. The results obtained by the instrumentation demonstrate that FBG sensors, when positioned on critical components, exhibit increased sensitivity. Furthermore, the installation of multiple sensors provides information about the location of faulty points – a level of detail unattainable by conventional systems. The methods and algorithms developed for anomaly detection in this study can be extended in the future to high-voltage devices with greater relevance to electrical system, such as circuit breakers and on-load tap-changers.