Bearing remaining useful life prediction based on NGO-transformer-BiLSTM model

To improve the accuracy and robustness of bearing remaining useful life (RUL) prediction, a hybrid deep learning model combining Transformer and BiLSTM is proposed in this study. The model leverages the multi-head self-attention mechanism of the Transformer encoder to capture global temporal dependencies from sensor signals, while the BiLSTM component enhances the modeling of both local and long-term sequential features. In addition, the Northern Goshawk Optimization (NGO) algorithm is introduced to automatically optimize key hyperparameters through global search and adaptive refinement, thereby avoiding performance bottlenecks caused by manual tuning. Experimental evaluations were conducted under two distinct operating conditions. The results show that the predicted RUL curves closely follow the actual degradation trends, indicating strong temporal modeling capability and good generalization performance. Furthermore, ablation studies demonstrate that the integration of Transformer, BiLSTM, and NGO contributes significantly to the model’s performance, achieving lower RMSE and MAE compared to baseline approaches.