Scholarlink

<p>Deep learning has achieved great success in cutter condition monitoring by using excellent feature mining methods. However, most existing methods treat anomaly detection and remaining useful life (RUL) prediction as independent tasks and ignore the relevant information in different tasks, resulting in a serious waste of equipment resources. Therefore, this paper proposes a multi-task collaborative monitoring method for anomaly detection and RUL prediction. First, vibration and cutting force signals are used as different but related views to construct graph-associated data structures. Second, a multi-task network mines shared spatial-temporal dependency features and completes subtasks. Third, the multi-regularization collaborative strategy is introduced to achieve collaborative optimization between tasks. Finally, the experimental results of the two cases clearly show that the proposed method has an excellent monitoring effect and feature extraction ability.</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

Collaborative monitoring method for cutter anomaly detection and RUL prediction based on multi-task learning

  • 期刊论文

  • 发表日期：

    2025年3月4日

查看全文
  • Xufeng Shao,
  • Xiaoyin Nie,
  • Hui Shi,
  • Zhicheng Zhao,
  • Gaohua Chen,
  • Gang Xie

摘要

Deep learning has achieved great success in cutter condition monitoring by using excellent feature mining methods. However, most existing methods treat anomaly detection and remaining useful life (RUL) prediction as independent tasks and ignore the relevant information in different tasks, resulting in a serious waste of equipment resources. Therefore, this paper proposes a multi-task collaborative monitoring method for anomaly detection and RUL prediction. First, vibration and cutting force signals are used as different but related views to construct graph-associated data structures. Second, a multi-task network mines shared spatial-temporal dependency features and completes subtasks. Third, the multi-regularization collaborative strategy is introduced to achieve collaborative optimization between tasks. Finally, the experimental results of the two cases clearly show that the proposed method has an excellent monitoring effect and feature extraction ability.