基于机器视觉的输煤皮带故障识别与定位研究

Research on fault recognition and location of coal conveying belt based on machine vision

摘要:

针对输煤皮带故障传统检测漏检率高、实时性差的问题，开展复杂工业环境下其故障视觉识别与定位研究。采用改进 YOLOv8 算法，结合尺度变换、亮度调整等数据增强扩展样本；优化网络结构，引入注意力机制与轻量化设计提升特征提取能力和实时性；通过多尺度融合、BiFPN优化及跨层级对齐增强多尺度检测能力，设计动态损失函数改善识别均衡性，结合UNet实现精细化分割。实验结果表明，故障识别mAP达93.8%，定位误差降至0.8mm（亚毫米级），平均检测帧率超42FPS；强光、高噪声等复杂环境中误检率低于 6.2%、漏检率低于5.5%，满足输煤皮带故障实时、精准检测的工业应用需求。

Abstract:

Aiming at the problems of high missed detection rate and poor real-time performance in traditional detection of coal conveyor belt faults, research on visual recognition and localization of such faults in complex industrial environments was conducted. An improved YOLOv8 algorithm was adopted, and data augmentation methods like scale transformation and brightness adjustment were combined to expand the sample size. The network structure was optimized: attention mechanisms and lightweight design were introduced to enhance feature extraction capability and real-time performance; multi-scale fusion, BiFPN optimization and cross-level alignment were used to improve multi-scale detection ability; a dynamic loss function was designed to enhance recognition balance; and UNet was integrated to achieve refined segmentation. Experimental results show that the mean Average Precision (mAP) of fault recognition reaches 93.8%, the localization error is reduced to 0.8mm (sub-millimeter level), and the average detection frame rate exceeds 42 FPS. In complex environments such as strong light and high noise, the false detection rate is less than 6.2% and the missed detection rate is less than 5.5%, which meets the industrial application requirements for real-time and accurate detection of coal conveyor belt faults.