Remaining Useful Life Prediction of Electrolytic Capacitors Incorporating Data-Driven and Model-Based Methods

Electrolytic capacitors are crucial but prone to failure components especially in power electronic systems. Health monitoring and prediction of such devices is an indispensable and promising work to guarantee a safe and reliable operation. Much research has been conducted on degradation modeling and remaining useful life (RUL) prediction of such devices. Nevertheless, most of the previous studies used exponential degradation model which is incapable of capturing the nonlinear characters during degradation. Additionally, model-based and data-driven methods were employed separately with an ignorance of complementary advantages of fusion methods. Concentrating on the mentioned limitations, this work is devoted to exploring an effective method for lifetime prediction of electrolytic capacitors. A second order state transition equation is used to model the degradation process. A fusion framework that incorporates model-based (particle filter) and data-driven (radial basis function neural network) methods supplemented by a mutation operation is developed. The developed framework is verified based on real data from aging experiments of electrolytic capacitors. It is indicated by the results that the proposed fusion method has desirable outcomes that outperforms each single method.