Scholarlink

<p>Abrasive belt grinding is widely employed in the precision machining of complex parts and difficult-to-machine materials, such as blades and integral disks. Accurate online prediction of material removal rate (MRR) is a prerequisite for ensuring workpiece accuracy. This study proposes a small-sample monitoring approach for the MRR of abrasive belt grinding, utilising multi-sensor data signal image fusion and deep regression networks. First, the visual (spark image), auditory (grinding sound), and tactile (vibration and force) signals—each closely related to the MRR during the grinding process—are collected. Employing the symmetric point pattern (SDP) method, the sound, vibration, and force time series are converted into visual images suitable for processing by deep regression networks. Subsequently, the spark image and the converted SDP image are fused via the pyramid Laplace (LPIF) method, enabling the integration of multi-sensor data into a single image that contains more representative MRR information. The deep regression network based on R-ResNet50 is developed to improve the image feature extraction and reduce the model dependence on large training datasets. Leveraging the strong visual extraction capability of the network, the model is trained on fused images corresponding to different MRRs. The results demonstrate that the input image, constructed via the integration of the SDP and LPIF methods, efficiently captures the rich, multimodal information associated with MRR. Furthermore, the proposed deep regression network enables small-sample MRR modelling, achieving an accuracy R2 of 0.9906 and root mean square error (RMSE) below 0.0095.</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

Small-sample monitoring approach for abrasive belt grinding material removal rate employing multi-sensor data signal image fusion and deep regression networks

  • 期刊论文

  • 发表日期：

    2026年1月4日

查看全文
  • Nina Wang,
  • Wanjing Pang,
  • Lijuan Ren,
  • Wenhai He,
  • Jingtong Zhang,
  • Guangpeng Zhang

摘要

Abrasive belt grinding is widely employed in the precision machining of complex parts and difficult-to-machine materials, such as blades and integral disks. Accurate online prediction of material removal rate (MRR) is a prerequisite for ensuring workpiece accuracy. This study proposes a small-sample monitoring approach for the MRR of abrasive belt grinding, utilising multi-sensor data signal image fusion and deep regression networks. First, the visual (spark image), auditory (grinding sound), and tactile (vibration and force) signals—each closely related to the MRR during the grinding process—are collected. Employing the symmetric point pattern (SDP) method, the sound, vibration, and force time series are converted into visual images suitable for processing by deep regression networks. Subsequently, the spark image and the converted SDP image are fused via the pyramid Laplace (LPIF) method, enabling the integration of multi-sensor data into a single image that contains more representative MRR information. The deep regression network based on R-ResNet50 is developed to improve the image feature extraction and reduce the model dependence on large training datasets. Leveraging the strong visual extraction capability of the network, the model is trained on fused images corresponding to different MRRs. The results demonstrate that the input image, constructed via the integration of the SDP and LPIF methods, efficiently captures the rich, multimodal information associated with MRR. Furthermore, the proposed deep regression network enables small-sample MRR modelling, achieving an accuracy R2 of 0.9906 and root mean square error (RMSE) below 0.0095.