Search Results (2)

To address the challenges of high labeling costs and insufficient cross-condition generalization in training cutting tool wear state prediction models, this paper proposes a model optimization method based on knowledge distillation. By constructing a teacher model using a Bidirectional Gated Recurrent Unit (BiGRU-GRU) and a student model utilizing a Transformer architecture, we jointly employ KL divergence distillation, feature Euclidean distance distillation, and cross-entropy supervised training. A multi-objective joint loss function is designed to facilitate knowledge transfer. Using a self-collected dataset of aerospace TC18 titanium alloy cutting tools and the publicly available PHM2010 dataset, we conduct comparative experiments based on two different data partitioning strategies: tool grouping and mixed grouping. The results indicate that knowledge distillation significantly enhances the performance of the student model. The Transformer-BiGRU model is more robust to knowledge distillation. Positive results were obtained for the Transformer-BiGRU model in both tool grouping and mix grouping experiments. Additionally, comparative experiments show that the proposed method outperforms traditional methods (such as GS-XGBoost and Attention-CNN) in both grouping types, validating the effectiveness of knowledge distillation in transferring knowledge related to tool wear discrimination and addressing the issue of insufficient training data. The research demonstrates that combining knowledge distillation with the Transformer architecture can effectively enhance model generalization capabilities, providing theoretical support for monitoring the state of aerospace titanium alloy machining tools. However, the misclassification issue during the severe wear stage still requires further optimization based on physical principles. Full article

Aiming at challenges such as the high complexity of the network model, the large number of parameters, and the slow speed of training and testing in cross-view gait recognition, this paper proposes a solution: Multi-teacher Joint Knowledge Distillation (MJKD). The algorithm employs multiple complex teacher models to train gait images from a single view, extracting inter-class relationships that are then weighted and integrated into the set of inter-class relationships. These relationships guide the training of a lightweight student model, improving its gait feature extraction capability and recognition accuracy. To validate the effectiveness of the proposed Multi-teacher Joint Knowledge Distillation (MJKD), the paper performs experiments on the CASIA_B dataset using the ResNet network as the benchmark. The experimental results show that the student model trained by Multi-teacher Joint Knowledge Distillation (MJKD) achieves 98.24% recognition accuracy while significantly reducing the number of parameters and computational cost. Full article