Research on the degradation state division of DC contactors based on LSTM and Pelt algorithm

Achieving self-awareness of the state of DC contactors is of great significance for improving the intelligence and reliability of the systems in which they operate. This paper explores the phased division of the degradation states of DC contactors, considering their multiple features and time-dependent data characteristics. Based on the current signals collected over the full life cycle of DC contactors, we first established a life model of DC contactors using the Long Short-Term Memory (LSTM) neural network. On this basis, considering the volatility and signal interference in the original data, the Pelt algorithm was used to segment the degradation state time series output by the life model into different phases. Analyzing the life test results of multiple experimental contactors using this method revealed that the degradation process of DC contactors exhibits distinct three-stage characteristics: the aging stage, the stable stage, and the failure stage. This makes the proposed method superior to common life prediction methods in reflecting the current working state of the contactor, which has practical significance for the maintenance and early warning of switching devices and lays a theoretical foundation for degradation state assessment.