Scholarlink

<p>Timely and effective acquisition of micro-vibrations of rotating machinery can diagnose the fault status of the equipment. Visual target detection schemes have the advantages of contactless, real-time and high accuracy. However, detecting micro-amplitude vibration of high-speed rotating machinery requires high accuracy. Therefore, in this paper, a high-speed camera is used as an acquisition medium to collect end-face vibration information from a rotor vibration test bench, and a target detection model YOLO-MVD is proposed to recognize micro-vibrations of rotating machinery to accurately detect rotor displacements. Firstly, we used EfficientViT backbone to replace YOLOv8n’s backbone, and secondly, we introduced Ghost Net’s lightweight thinking to deal with the C2F module, and then embedded in the neck network CBAM attention mechanism to improve the network feature extraction ability, and finally the WIoU v3 loss function to improve network performance. The algorithm is trained on the self-constructed rotor end-face image dataset, comparing multiple visual detection algorithms all show excellent performance, the accuracy rate reached 95.78%, the recall rate reached 90.34%, the average precision mean value reached 95.63%, and the performance is outstanding in terms of lightweighting and detection speed, with the model size compressed to 2.94&#xa0;m and the frame rate increased to 114 FPS. and the performance is outstanding in comparing the laser Doppler vibrometer validation experiments in the frequency domain peak accuracy error range of 0.001. The research results verify the feasibility of the visual inspection program for rotating machinery, and provide a reference program for research in the field of practical industrial vibration detection.</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

Micro vibration detection algorithm for rotating machinery based on visual target detection

  • 期刊论文

  • 发表日期：

    2025年2月10日

查看全文
  • Sheng Liu,
  • Gulbahar Tohti,
  • Mamtimin Geni,
  • Hualong He,
  • Zhiqiang Wu,
  • Caiwen Liao

摘要

Timely and effective acquisition of micro-vibrations of rotating machinery can diagnose the fault status of the equipment. Visual target detection schemes have the advantages of contactless, real-time and high accuracy. However, detecting micro-amplitude vibration of high-speed rotating machinery requires high accuracy. Therefore, in this paper, a high-speed camera is used as an acquisition medium to collect end-face vibration information from a rotor vibration test bench, and a target detection model YOLO-MVD is proposed to recognize micro-vibrations of rotating machinery to accurately detect rotor displacements. Firstly, we used EfficientViT backbone to replace YOLOv8n’s backbone, and secondly, we introduced Ghost Net’s lightweight thinking to deal with the C2F module, and then embedded in the neck network CBAM attention mechanism to improve the network feature extraction ability, and finally the WIoU v3 loss function to improve network performance. The algorithm is trained on the self-constructed rotor end-face image dataset, comparing multiple visual detection algorithms all show excellent performance, the accuracy rate reached 95.78%, the recall rate reached 90.34%, the average precision mean value reached 95.63%, and the performance is outstanding in terms of lightweighting and detection speed, with the model size compressed to 2.94 m and the frame rate increased to 114 FPS. and the performance is outstanding in comparing the laser Doppler vibrometer validation experiments in the frequency domain peak accuracy error range of 0.001. The research results verify the feasibility of the visual inspection program for rotating machinery, and provide a reference program for research in the field of practical industrial vibration detection.