华中科技大学学报（自然科学版）2025，Vol.53Issue(6)：7-14,8.DOI:10.13245/j.hust.250602
基于Transformer-CNN-GWO的多信息融合刀具磨损预测
Transformer-CNN-GWO based multi-information fusion tool wear prediction
摘要
Abstract
Aiming at the problems of poor single information effect of tool wear diagnosis and time-consuming manual adjustment of hyperparameters,a method of fusion of attention convolutional neural network(Transformer-CNN)and grey wolf optimizer(GWO)algorithm was proposed to realize high-precision prediction of tool wear.First,after the pre-processing of multi-sensor data,the Pearson correlation coefficient method and kernel principal component analysis were sequentially used for multi-domain feature extraction and feature fusion and dimensionality reduction to obtain the model input parameters with high predictive performance.Then,to enhance the model's capability to learn key features autonomously and parallel computation,the Transformer-CNN combinatorial model with M-N structure was constructed for tool wear prediction,and the gray wolf algorithm was used to optimize the hyperparameters of the M-N structure Transformer-CNN network to obtain the optimal M-N hyperparameter combination.Finally,the proposed tool wear prediction model was tested with the tool wear test data.The prediction accuracy is above 99.98%,and the error is much smaller than that of other models.Experimental results verified the accuracy and robustness of the proposed method.In addition,the necessity and effectiveness of the proposed method were further validated through a comparison with the optimized deep learning method.关键词
刀具磨损预测/注意力卷积神经网络/灰狼优化算法/特征融合/超参数优化Key words
tool wear prediction/attentional convolutional neural networks/grey wolf optimizer algorithm/feature fusion/hyperparameter optimization分类
计算机与自动化引用本文复制引用
赛希亚拉图,成佳闻,张超勇..基于Transformer-CNN-GWO的多信息融合刀具磨损预测[J].华中科技大学学报（自然科学版）,2025,53(6):7-14,8.基金项目
工业和信息化部高端数控机床与基础制造装备国家科技重大专项(2024ZD0707501) （2024ZD0707501）
中德重点研发计划资助项目(2022YFE0114200). （2022YFE0114200）
