Tool wear monitoring based on signal feature fusion and deep learning

To improve the accuracy of tool wear monitoring in machining, this paper proposes a method based on multi-sensor signal feature fusion and deep neural networks. The proposed method synchronously acquires multi-sensor signals, from which multi-domain features are first extracted. These features are then selected based on the Max-Relevance Min-Redundancy (mRMR) criterion and subsequently fused through weighted averaging to form a fused feature vector. Subsequently, this vector is transformed into an image dataset using the Recurrence Plot (RP) method. On this basis, a Multi-level Convolutional Gated Network model integrated with the Sparrow Search Algorithm (SSA-MLCGN) is designed for tool wear prediction. This model integrates convolutional layers with a Gated Recurrent Unit (GRU), utilizing the convolutional layers to extract features from the recurrence images while employing Dropout to mitigate overfitting. Concurrently, the Sparrow Search Algorithm (SSA) is introduced to optimize the hyperparameters of the GRU, thereby significantly enhancing the model’s prediction accuracy and generalization performance. Finally, ablation studies, single/multi-condition experiments, and noise robustness experiments were conducted on a public dataset, validating the high precision, strong generalization, and industrial applicability of the proposed method.