Adaptive Remaining Useful Life Prediction of Parallel DC ‐Link Capacitor Banks Considering Thermally Coupled Degradation

The remaining useful life (RUL) of DC‐link capacitor banks is critical to the reliability of power electronic converters. This paper proposes an adaptive RUL prediction method for parallel DC‐link capacitor banks. The method explicitly considers thermally coupled degradation among capacitor units. First, a discrete degradation‐increment model with nearest‐neighbor coupling is developed to describe intra‐bank degradation interactions. Then, a structure function is used to map capacitor‐level health states to the bank‐level capacitance. The end‐of‐life criterion of the capacitor bank is defined according to the energy‐storage and ripple‐suppression constraints of the DC‐link stage. To handle limited training data, a simplified empirical‐Bayes procedure is developed for offline parameter identification. A shared nonlinear coefficient is estimated from the training samples, and sample‐specific parameters are identified by constrained ridge regression. The identified parameter ensemble is used to construct Gaussian priors. During operation, these priors are recursively updated using online monitoring data. Experimental results show that the proposed method can accurately capture the coupled degradation evolution of capacitor banks and provide reliable RUL predictions.