RUL Prediction of DC Contactor Using CNN-LSTM With Channel Attention and Fusion of Dual Aggregated Features

Abstract

Predicting the remaining useful life (RUL) of DC contactors is crucial for the reliability of aerospace, rail transit, and electric vehicle systems. Challenges arise due to high-dimensional operational data, difficulty fusing spatial-temporal features, and noisy environments. This paper proposes a novel deep learning model called DAF-CA-CNN-LSTM, which integrates Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), a Channel Attention (CA) mechanism, and a Dual Aggregated Features (DAF) strategy. The CNN extracts local spatial features, while the LSTM captures long-term temporal dependencies. The CA mechanism enhances focus on critical degradation features by adaptively weighting feature channels. The DAF strategy enriches feature representation by combining aggregated encoded features with original inputs. An experimental system was developed to collect operational data under specific conditions. Key features were extracted, preprocessed, and used to train and evaluate the model. Results show that the DAF-CA-CNN-LSTM model significantly outperforms traditional LSTM and CNN-LSTM models in RUL prediction, achieving higher accuracy and robustness in complex, noisy environments. This model effectively addresses DC contactor life prediction challenges, offering a promising tool for improving maintenance strategies and operational reliability.

Keywords

• Channel attention mechanism
• CNN-LSTM deep learning model
• DC contactor remaining useful life prediction
• dual aggregated features strategy