33 / 2025-05-10 18:28:46

A Tool Wear Condition Recognition Model Based on GBDT-BiGRU

Milling processing, Signal processing, GBDT, GRU, Feature selection, Tool wear.

Tool wear condition monitoring is essential for enhancing machining efficiency and ensuring product quality. Conventional approaches are limited in predictive accuracy and generalization when coping with complex nonlinearities and temporal dependencies. To address these deficiencies, a novel monitoring model that integrates Gradient Boosting Decision Tree (GBDT) with Bidirectional Gated Recurrent Unit (BiGRU) is proposed. The architecture encompasses three sequential modules: data preprocessing, GBDT-based feature selection, and BiGRU-based training. After signal acquisition and preprocessing, GBDT is employed to quantify feature importance, through which key wear-relevant features are retained and dimensionality is reduced, thereby improving computational efficiency. Subsequently, the BiGRU network is utilized to capture temporal evolutions of the selected features and to classify wear states accurately, leading to markedly improved monitoring precision and generalization capability. Validation on both the public PHM 2010 dataset and the in-house EX dataset demonstrates training accuracies of 98.9 % and 97.8 %, respectively, confirming the validity and robustness of the proposed framework.