Scholarlink

<p>Accurate prediction of disc cutter wear is essential for ensuring the efficiency and safety of tunnel boring machine (TBM) operations. Traditional physical models often rely on idealized assumptions and fail to capture the complex nonlinear interactions between geological conditions and cutter wear mechanisms, while purely data-driven models suffer from limited interpretability and instability under small-sample conditions. To address these challenges, this study proposes a Bayesian physics-integrated disc wear model (BPI-DWM) that integrates physical wear mechanisms with deep neural network techniques. The model embeds physics-informed loss functions, including theoretical wear residuals, L2 regularization, and physical constraint bounds, within the network to achieve a balance between physical consistency and data fitting accuracy. Bayesian optimization (BO) is employed to automatically tune key hyperparameters, improving convergence efficiency and model adaptability. Comparative experiments based on real-world TBM data demonstrate that the BPI-DWM outperforms traditional data-driven and physical models in terms of prediction accuracy, robustness, and generalization. Under the complete training dataset, the proposed model achieved an <i>R</i><sup>2</sup> of 0.9005, RMSE of 2.7282, MAE of 2.2224, and MAPE of 15.38%, showing superior predictive performance. Even under data-scarce conditions, the model maintains stable learning and prediction capability. Furthermore, SHAP analysis explicitly reveals the underlying wear mechanisms learned by the model: excavation distance, abrasive wear coefficient, and rock strength are identified as the main positive drivers of disc cutter wear, indicating the cumulative effect of tunneling distance and the intensified abrasive interaction between rock and cutter. In contrast, friction coefficient and penetration exhibit more stable but relatively weaker effects within the investigated operating range. Overall, BPI-DWM not only provides high-precision prediction but also captures physically consistent wear mechanisms, offering a reliable and interpretable tool for intelligent TBM operation and cutter maintenance planning.</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

BPI-DWM: A Physics- and Data-Driven Disc Cutter Wear Prediction Model Balancing Predictive Accuracy and Interpretability

  • 期刊论文

  • 发表日期：

    2026年7月24日

查看全文
  • Qian Zhang,
  • Yaoqi Nie,
  • Lijie Du,
  • Yujing Jiang

摘要

Accurate prediction of disc cutter wear is essential for ensuring the efficiency and safety of tunnel boring machine (TBM) operations. Traditional physical models often rely on idealized assumptions and fail to capture the complex nonlinear interactions between geological conditions and cutter wear mechanisms, while purely data-driven models suffer from limited interpretability and instability under small-sample conditions. To address these challenges, this study proposes a Bayesian physics-integrated disc wear model (BPI-DWM) that integrates physical wear mechanisms with deep neural network techniques. The model embeds physics-informed loss functions, including theoretical wear residuals, L2 regularization, and physical constraint bounds, within the network to achieve a balance between physical consistency and data fitting accuracy. Bayesian optimization (BO) is employed to automatically tune key hyperparameters, improving convergence efficiency and model adaptability. Comparative experiments based on real-world TBM data demonstrate that the BPI-DWM outperforms traditional data-driven and physical models in terms of prediction accuracy, robustness, and generalization. Under the complete training dataset, the proposed model achieved an R2 of 0.9005, RMSE of 2.7282, MAE of 2.2224, and MAPE of 15.38%, showing superior predictive performance. Even under data-scarce conditions, the model maintains stable learning and prediction capability. Furthermore, SHAP analysis explicitly reveals the underlying wear mechanisms learned by the model: excavation distance, abrasive wear coefficient, and rock strength are identified as the main positive drivers of disc cutter wear, indicating the cumulative effect of tunneling distance and the intensified abrasive interaction between rock and cutter. In contrast, friction coefficient and penetration exhibit more stable but relatively weaker effects within the investigated operating range. Overall, BPI-DWM not only provides high-precision prediction but also captures physically consistent wear mechanisms, offering a reliable and interpretable tool for intelligent TBM operation and cutter maintenance planning.