Scholarlink

<p>In the process of machining, the condition of tool wear directly determines both the quality of the workpiece and the production efficiency. To address the limitations of existing prediction methods, this study proposes a novel collaborative optimization approach integrating feature fusion and predictive modeling, namely a tool wear prediction method based on multi-domain feature fusion and the TabPFN–SimAM attention mechanism. Partial Least Squares (PLS) is employed to perform supervised fusion of multi-sensor features, thereby constructing a highly discriminative feature set. Subsequently, the SimAM attention mechanism is introduced to adaptively reweight the fused features, enhancing the representation of critical information. The Tabular Prior-data Fitted Network (TabPFN) serves as the core predictive model, and an Enhanced Zebra Optimization Algorithm (EZOA) is employed to perform combinational search for optimal solutions of the key parameters. Experimental results based on the PHM2010 dataset demonstrate that the proposed method achieves high prediction accuracy in tool wear estimation, with an average R² exceeding 0.99. Its performance significantly outperforms single-domain feature methods as well as several mainstream benchmark models, including SVR, XGBoost, BiLSTM, Transformer, and TabPFN. These results verify the comprehensive superiority of the proposed approach in terms of prediction accuracy and robustness.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Tool wear prediction based on multi-domain feature fusion and TabPFN-SimAM integration

  • 期刊论文

  • 发表日期：

    2026年5月8日

查看全文
  • Shuqiang Wang,
  • Zhiyang Li,
  • Mingyang Yang,
  • Gongshuai Cui

摘要

In the process of machining, the condition of tool wear directly determines both the quality of the workpiece and the production efficiency. To address the limitations of existing prediction methods, this study proposes a novel collaborative optimization approach integrating feature fusion and predictive modeling, namely a tool wear prediction method based on multi-domain feature fusion and the TabPFN–SimAM attention mechanism. Partial Least Squares (PLS) is employed to perform supervised fusion of multi-sensor features, thereby constructing a highly discriminative feature set. Subsequently, the SimAM attention mechanism is introduced to adaptively reweight the fused features, enhancing the representation of critical information. The Tabular Prior-data Fitted Network (TabPFN) serves as the core predictive model, and an Enhanced Zebra Optimization Algorithm (EZOA) is employed to perform combinational search for optimal solutions of the key parameters. Experimental results based on the PHM2010 dataset demonstrate that the proposed method achieves high prediction accuracy in tool wear estimation, with an average R² exceeding 0.99. Its performance significantly outperforms single-domain feature methods as well as several mainstream benchmark models, including SVR, XGBoost, BiLSTM, Transformer, and TabPFN. These results verify the comprehensive superiority of the proposed approach in terms of prediction accuracy and robustness.