Abstract: Accurate monitoring of tool wear is essential for ensuring machining efficiency and reducing production costs. To address the limitations of single-model approaches in feature extraction and the challenges of manual parameter tuning in tool wear monitoring, an adaptive Transformer-BiLSTM tool wear monitoring method optimized by the NSGA-II multi-objective algorithm was proposed. Firstly, milling force and vibration signals are preprocessed using the Hampel filter to effectively remove abnormal noise while preserving the intrinsic wear-related features. Subsequently, multidimensional feature sets representing tool wear are extracted from the time domain, frequency domain, and time-frequency domain. A hybrid Transformer-BiLSTM feature fusion model is then constructed, where the multi-head attention mechanism of the Transformer captures long-range global dependencies in the wear evolution process, while the bidirectional structure of BiLSTM enhances the characterization of temporal dynamics. Furthermore, to achieve adaptive parameter configuration, the NSGA-II algorithm is introduced to optimize key model parameters with mean squared error and model complexity as dual objectives, thus obtaining the optimal Pareto solution set automatically. Experimental results demonstrate that the proposed method significantly outperforms comparative models. The root mean square error, mean absolute error, and coefficient of determination reach 7.542 7 μm, 6.121 0 μm, and 0.958 8, respectively, indicating the model's high accuracy in predicting tool wear progression.