Remaining Useful Life Prediction of Cutting Tools Based on a Depthwise Separable TCN-BiLSTM Model with Temporal Attention

The prediction of remaining tool life is critical in practical manufacturing, as it enables early identification of tool wear conditions, preventing downtime and quality issues caused by tool failure. This, in turn, helps reduce costs and improve production efficiency. To address the limitations of traditional models in modeling complex temporal dependencies and capturing critical moment information, a hybrid deep learning model based on a TCN-BiLSTM-Attention architecture was proposed. The model first enhanced low-level features from raw multivariate signals using an initial feature extraction module. Local temporal patterns were then extracted by an improved Temporal Convolutional Network (TCN), followed by a Bidirectional Long Short-Term Memory (BiLSTM) network to capture global sequential dependencies. An attention mechanism was further introduced to strengthen the model’s ability to focus on and select critical features over time, enabling deeper fusion and modeling of complex temporal dynamics. Finally, a multi-layer regression network was applied to predict the remaining tool life. Experimental validation on the PHM 2010 dataset demonstrates that the proposed model outperforms other comparative models. Among the three evaluation metrics (RMSE, MAE, and R2), the optimal results are 0.0184, 0.0127, and 0.9959 respectively, which fully proves that the model exhibits excellent prediction performance. Ablation studies further confirm the contribution of each module to the overall performance improvement.