A feature extraction enhancing-based LSTM integrated deep learning model for remaining useful life prediction

The prediction of Remaining Useful Life (RUL) for complex engineering systems is crucial for maintaining their safety and reliability. To address the issue of limited RUL prediction accuracy due to inadequate extraction of degradation features in these systems, this paper presents an innovative RUL prediction method that integrates residual dilated causal convolution and Long Short-Term Memory (LSTM) networks. This approach employs residual connections to stack multiple dilated causal convolutions, facilitating the deep extraction of degradation features and enhancing the feature learning process. Additionally, it incorporates LSTM structures to forge a linkage between the degradation features and RUL, thereby enabling accurate RUL prediction for complex engineering systems. To validate the efficacy of the proffered method, experimental validation was conducted using the NASA public dataset C-MAPSS, and a comparative analysis was performed against the results of existing mainstream methods. The experimental results demonstrate that the suggested approach significantly enhances the precision of RUL estimation for complex engineering systems.