Ensemble Attention Bidirectional Recurrent Neural Network-based RUL Prediction Under Variable Working Condition

Predicting the remaining useful life (RUL) of mechanical equipment under variable operating conditions poses a significant challenge in prognostics. Existing methods often focus solely on monitoring data from the equipment itself, neglecting crucial operating condition data. This limitation leads to reduced prediction accuracy in fluctuating operating conditions. To address this gap, this paper proposes an Ensemble Attention Bidirectional Recurrent Neural Network (EABRNN). The EABRNN model processes multi-sensor monitoring and operational context data separately. It leverages parallel bidirectional Long Short-Term Memory (Bi-LSTM) and bidirectional Gated Recurrent Unit (Bi-GRU) architectures to extract comprehensive features. Furthermore, it integrates Temporal Pattern Attention to enhance remaining useful life prediction accuracy, particularly in variable aircraft engine operating conditions. By combining data integration and advanced model architecture, the EABRNN aims to improve the robustness and accuracy of RUL predictions in practical scenarios.