Scholarlink

<p>In practical industrial applications of CNC machine tools, conventional fault diagnosis methods often face challenges such as scarce fault samples and insufficient model generalisation capability. To address these issues, this paper proposes a digital twin fault diagnosis method that integrates convolutional attention gated recurrent units (CA-GRU) with physics-informed neural networks (PINN). Firstly, a multi-body dynamics model of the ball screw and its support bearings was established based on Hertz contact theory, serving as the physical constraint term for the PINN. Secondly, a CA-GRU module was designed to enhance the model’s capability to capture transient impact features in vibration signals through the synergistic effect of convolutional layers and attention mechanisms. Thirdly, a dynamic loss function with learnable weight parameters was constructed to achieve an adaptive balance between data-driven loss and physical consistency loss. Finally, a bidirectional feedback mechanism was established between the physical system and the digital twin, ensuring real-time system state tracking and dynamic parameter adjustment. Experimental results demonstrate that the relative error of fault time-domain period for support bearing inner race faults is controlled within 0.5%, while the relative error of fault characteristic frequency is below 0.9%. Compared with other methods, the proposed method achieves an F1-score of 96.5% with high fault simulation accuracy. This approach provides a novel technical pathway for intelligent operation and maintenance as well as predictive maintenance of CNC machine tools.</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

Fault diagnosis model of digital twin technology for CNC machine tool feed system based on CA-GRU-PINN

  • 期刊论文

  • 发表日期：

    2025年10月26日

查看全文
  • Xu Zhang,
  • Xie Xiaozheng

摘要

In practical industrial applications of CNC machine tools, conventional fault diagnosis methods often face challenges such as scarce fault samples and insufficient model generalisation capability. To address these issues, this paper proposes a digital twin fault diagnosis method that integrates convolutional attention gated recurrent units (CA-GRU) with physics-informed neural networks (PINN). Firstly, a multi-body dynamics model of the ball screw and its support bearings was established based on Hertz contact theory, serving as the physical constraint term for the PINN. Secondly, a CA-GRU module was designed to enhance the model’s capability to capture transient impact features in vibration signals through the synergistic effect of convolutional layers and attention mechanisms. Thirdly, a dynamic loss function with learnable weight parameters was constructed to achieve an adaptive balance between data-driven loss and physical consistency loss. Finally, a bidirectional feedback mechanism was established between the physical system and the digital twin, ensuring real-time system state tracking and dynamic parameter adjustment. Experimental results demonstrate that the relative error of fault time-domain period for support bearing inner race faults is controlled within 0.5%, while the relative error of fault characteristic frequency is below 0.9%. Compared with other methods, the proposed method achieves an F1-score of 96.5% with high fault simulation accuracy. This approach provides a novel technical pathway for intelligent operation and maintenance as well as predictive maintenance of CNC machine tools.