Scholarlink

<p>Tool wear significantly impacts machining accuracy and surface quality. Traditional methods typically focus on establishing a direct mapping between indirect sensor signals and offline-measured tool wear states, often neglecting the influence of in-situ tool wear conditions during machining on future predictions. This study proposes a novel two-stage tool wear prediction framework utilizing multi-sensor signals, comprising two stages: tool condition monitoring (TCM) and tool wear forecasting (TWF). The framework integrates a residual convolutional neural network (ResCNN) for TCM and a bidirectional long short-term memory (BiLSTM) for TWF. First, raw sensor signals are processes using variational mode decomposition. The ResCNN then performs anomaly detection for TCM, feeding into BiLSTM to predict future wear. A data fusion method combining short-term and long-term predictions enhances both stability and accuracy. Evaluated on the IEEE PHM 2010 dataset, the proposed method achieves an MSE of 3.456 µm in TCM and 19.734 µm in TWF, demonstrating its effectiveness in precise tool wear prediction.</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

Milling tool wear monitoring and forecasting with indirect multi-sensor information based on ResCNN and BiLSTM

  • 期刊论文

  • 发表日期：

    2025年11月8日

查看全文
  • Xiaojian Liu,
  • Hanqi Yan,
  • Lemiao Qiu,
  • Yang Wang,
  • Zili Wang,
  • Yiming Zhang,
  • Shuyou Zhang,
  • Jianrong Tan

摘要

Tool wear significantly impacts machining accuracy and surface quality. Traditional methods typically focus on establishing a direct mapping between indirect sensor signals and offline-measured tool wear states, often neglecting the influence of in-situ tool wear conditions during machining on future predictions. This study proposes a novel two-stage tool wear prediction framework utilizing multi-sensor signals, comprising two stages: tool condition monitoring (TCM) and tool wear forecasting (TWF). The framework integrates a residual convolutional neural network (ResCNN) for TCM and a bidirectional long short-term memory (BiLSTM) for TWF. First, raw sensor signals are processes using variational mode decomposition. The ResCNN then performs anomaly detection for TCM, feeding into BiLSTM to predict future wear. A data fusion method combining short-term and long-term predictions enhances both stability and accuracy. Evaluated on the IEEE PHM 2010 dataset, the proposed method achieves an MSE of 3.456 µm in TCM and 19.734 µm in TWF, demonstrating its effectiveness in precise tool wear prediction.