Scholarlink

<p>In the domain of wear prediction for cutting tools, traditional methods have primarily yielded point estimates, which may inadequately address the uncertainties inherent in manufacturing processes where precision and reliability are paramount. To confront this challenge, this paper proposed a Bayesian temporal convolutional neural network with a multi-head attention mechanism (BTCN-MHAT) model that effectively quantifies epistemic uncertainty by integrating Bayesian weight modeling. This innovative approach not only enhances the accuracy of point prediction but also provides 95 % confidence intervals for wear distributions, significantly improving the reliability of the predictions. The BTCN-MHAT model achieves a reduction in root mean square error (RMSE) and mean absolute error (MAE) by 42.16 % and 41.15 %, respectively, on tool 6 when compared to TCN model. Additionally, it offers more robust confidence intervals than the quantile regression method. This substantial improvement in epistemic uncertainty quantification positions the BTCN-MHAT as a highly effective tool for tool wear prediction, offering critical decision support and enhancing operational efficiency within the manufacturing industry.</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

Bayesian-deep-learning-based model with uncertainty quantification for cutting tool wear prediction

  • 期刊论文

  • 发表日期：

    2026年4月9日

查看全文
  • Chenhua Pang,
  • Chuang Chen,
  • Shubo Jiang

摘要

In the domain of wear prediction for cutting tools, traditional methods have primarily yielded point estimates, which may inadequately address the uncertainties inherent in manufacturing processes where precision and reliability are paramount. To confront this challenge, this paper proposed a Bayesian temporal convolutional neural network with a multi-head attention mechanism (BTCN-MHAT) model that effectively quantifies epistemic uncertainty by integrating Bayesian weight modeling. This innovative approach not only enhances the accuracy of point prediction but also provides 95 % confidence intervals for wear distributions, significantly improving the reliability of the predictions. The BTCN-MHAT model achieves a reduction in root mean square error (RMSE) and mean absolute error (MAE) by 42.16 % and 41.15 %, respectively, on tool 6 when compared to TCN model. Additionally, it offers more robust confidence intervals than the quantile regression method. This substantial improvement in epistemic uncertainty quantification positions the BTCN-MHAT as a highly effective tool for tool wear prediction, offering critical decision support and enhancing operational efficiency within the manufacturing industry.