<p>Tool wear is a crucial factor in the milling process, directly impacting machining accuracy, part quality, and production costs. Accurate prediction of tool wear enables timely tool replacement, reducing downtime and enhancing product quality. However, traditional methods fall short of the high demands of smart manufacturing. To address this, a tool wear prediction method based on an extended long short-term memory (xLSTM) network with an efficient channel attention mechanism has been proposed. Time-frequency domain features are extracted using variational mode decomposition. Multi-domain features are derived from multi-sensor data, and Pearson decomposition is utilized to decrease the dimensionality in feature engineering. The xLSTM is used to improve the inherent defects of the original LSTM model in terms of extracting the stored signal, expanding the memory capacity, and allowing parallel operation, to achieve the model’s precision in predicting the extent of tool wear. The performance of the model was improved by introducing an efficient channel attention mechanism (ECA) to consider the weights of different feature mappings. In this research, the model was evaluated against the other four models (LSTM, transformer, LSTM-ECA, and xLSTM) in three sets of crossover experiments using four regression indexes, RMSE, MSE, MAE, and R², and the results were all optimal, which verified the superiority and generalization of the method.</p>