Remaining useful life prediction of DC contactor based on LSTM

As a crucial electronic component in DC systems, predicting the Remaining Useful Life (RUL) of DC contactors can significantly enhance the operational reliability of the systems they are part of. Current methods for RUL prediction, which are based on single data points or traditional machine learning, face issues such as the selection of features that are inconvenient to monitor, high application costs, and low accuracy. In response, this paper proposes a method for predicting the RUL of DC contactors using Long Short-Term Memory (LSTM) neural networks. A specific DC contactor is examined as a case study to demonstrate the feasibility of applying this method. The advantage of the proposed method lies in its requirement for only the collection of current signals throughout the full lifecycle of the DC contactor to predict its RUL, resulting in low application costs. Compared to RUL prediction methods based on traditional Back Propagation Neural Networks (BPNN), this method achieves higher accuracy. Moreover, by considering key structural parameters that affect the lifespan of DC contactors, the method provides guidance for contactor design and exhibits better generalization capabilities in the predictive model.