Scholarlink

<p>Tool wear state recognition (TWSR) is a process which provides essential input for predictive maintenance (PM) systems, helping operators optimize tool usage, schedule timely replacements, and maintain machining quality and efficiency. However, reliable TWSR faces challenges such as subjective labeling of wear stages, inconsistencies in tool wear across cutters, and computational inefficiencies that make models unsuitable for real-time applications. This study introduces a novel tool wear state division method, based on higher-order derivatives to reduce labeling subjectivity. Additionally, a robust time-domain feature is introduced, along with other domain-specific features identified through Levene’s test for homogeneity to mitigate data distribution discrepancies (DDD) across milling cutters. A lightweight artificial intelligence (AI) model based on AttentionLSTM is developed to reduce computational complexity while maintaining high accuracy. Experiments using the public PHM2010 dataset and validation on three additional public datasets provide further support generalizability and robustness of the proposed approach. The AttentionLSTM model achieved accuracy above 98% with a prediction time under 5 ms, making it suitable for real-time applications and integration into digital twin systems. These findings demonstrate a reliable and robust solution for TWSR, with the potential to be scalable and efficient, advancing the capabilities of data-driven PM.</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 State Recognition in CNC Milling using Reliable Labels, Robust Domain Features, and Lightweight AI Models

  • 期刊论文

  • 发表日期：

    2025年6月11日

查看全文
  • Deep Patel,
  • Sreekumar Muthuswamy

摘要

Tool wear state recognition (TWSR) is a process which provides essential input for predictive maintenance (PM) systems, helping operators optimize tool usage, schedule timely replacements, and maintain machining quality and efficiency. However, reliable TWSR faces challenges such as subjective labeling of wear stages, inconsistencies in tool wear across cutters, and computational inefficiencies that make models unsuitable for real-time applications. This study introduces a novel tool wear state division method, based on higher-order derivatives to reduce labeling subjectivity. Additionally, a robust time-domain feature is introduced, along with other domain-specific features identified through Levene’s test for homogeneity to mitigate data distribution discrepancies (DDD) across milling cutters. A lightweight artificial intelligence (AI) model based on AttentionLSTM is developed to reduce computational complexity while maintaining high accuracy. Experiments using the public PHM2010 dataset and validation on three additional public datasets provide further support generalizability and robustness of the proposed approach. The AttentionLSTM model achieved accuracy above 98% with a prediction time under 5 ms, making it suitable for real-time applications and integration into digital twin systems. These findings demonstrate a reliable and robust solution for TWSR, with the potential to be scalable and efficient, advancing the capabilities of data-driven PM.