Feature Extraction and Selection for Identifying Faults in Contactors Using Fiber Bragg Grating

Switching devices are used in a wide application field to control and protect electrical systems. Failures in such equipment cause a loss of reliability in electrical facilities, which can lead to catastrophic consequences. The main advantage of using optical sensors is their immunity to the electromagnetic field, allowing installation in unfeasible locations compared to other technologies presented in related works. Consequently, the proposed approach consists of a new application employing fiber Bragg grating (FBG) to measure dynamic strain signals while switching a low-voltage contactor and develop a signal processing algorithm to extract and select features for classification using supervised learning methods. The models were trained and validated with different measurement sets, dividing them into intermediate and critical wear-out stages. The test procedures were carried out in a controlled manner replacing the contactor's main internal components. Two feature extraction methods were evaluated. The first calculates the power spectral density (PSD) and the switching time, while the second considers the coefficients generated by the wavelet scattering transform (WST). With maximum relevance and minimum redundancy (mRMR) and the support vector machine (SVM) algorithms, it was possible to identify components states, obtaining an accuracy of 99.4% for cross validation, 100% for validation dataset, and 86.4% for the new test dataset. The results demonstrate that the proposed system can recognize critical faults and is promising to be applied in other types of commutation equipment in future applications striving to increase the complexity of the evaluated devices.