Previsão e detecção de falhas em capacitores eletrolíticos

Abstract

Capacitors are electronic components that vary widely in characteristics over their useful life. After being subjected to several charging and discharging cycles, the capacitors show a loss in their capacitance and with that they start operating outside nominal characteristics. Prognosis and Health Monitoring Techniques (PHM) can be used to monitor the evolution of capacitor degradation and estimate their remaining useful life (RUL). The present work aims to monitor the degradation and predict the remaining useful life of capacitors through the use of artificial neural networks and Gaussian process regression (GPR) using the NASA repository database. Different network architectures are evaluated to estimate the remaining useful life: MLP (Multilayer Perceptron), RBF (Radial Basis Function) and ELM (Extreme Learning Machine). The coefficient of determination (R2) and the mean square error (MSE) are used as performance indicators to compare GPR and the different network architectures. The accuracy of the RUL estimates are compared in terms of the relative accuracy (RA) and the α - λ performance, which are indicators proposed in the literature for PHM algorithms.