基于热红外图像的带式输送机托辊故障检测研究

Research on fault detection of belt conveyor roller based on thermal infrared image

  • 摘要: 针对目前带式输送机巡检机器人搭载红外采集设备移动受限,存在不能实时进行数据采集、数据处理并上传至监控终端,无法完成远程故障检测,续航能力不足等问题,提出了一种基于热红外图像的带式输送机托辊故障检测方法。带式输送机巡检机器人搭载托辊故障检测器及红外热像仪,红外热像仪将采集的托辊热红外图像序列与温度数据传输给托辊故障检测器进行托辊故障检测,检测结果由托辊故障检测器内置的WH−L101无线传输模块发送给上位机。提出了一种带式输送机托辊故障检测算法:利用YOLOv5s目标检测算法提取托辊热红外图像的感兴趣区域(ROI),采用维纳滤波和自适应中值滤波算法对ROI图像进行滤波,利用自适应直方图均衡化和图像锐化算法对滤波后的ROI图像进行增强,采用基于形态学的Otsu图像分割算法对增强后的ROI图像进行图像分割,得到待检测的托辊图像,利用Harris角点检测算法提取托辊图像特征,获得托辊位置信息,提取相应位置的温度信息,并采用基于相对温差法的托辊故障检测算法判定托辊故障。实验结果表明:① YOLOv5s网络模型提取托辊ROI的目标检测结果平均准确率为99.12%。② 提出的托辊故障检测算法对托辊故障(无故障、轴承锈蚀、托辊卡转、筒体磨穿)检测的平均准确率为97.625%,帧率为16 帧/s。③ 将检测结果通过无线传输模块传送至上位机,可显示故障类型及关键区域温度,并进行报警。

     

    Abstract: Currently, the inspection robot for belt conveyors equipped with infrared acquisition devices is limited in movement. There are problems such as inability to collect data, process data, upload data to monitoring terminals in real-time and complete remote fault detection, insufficient endurance and so on. A fault detection method of belt conveyor roller based on thermal infrared images has been proposed. The belt conveyor inspection robot is equipped with a roller fault detector and an infrared thermal imager. The infrared thermal imager transmits the collected roller thermal infrared image sequence and temperature data to the roller fault detector for roller fault detection. The WH-L101 wireless transmission module in the roller fault detector is used to send the detection results to the upper computer. A belt conveyor roller fault detection algorithm is proposed. The algorithm uses the YOLOv5s object detection algorithm to extract the region of interest (ROI) of the roller thermal infrared image. The image of the ROI is filtered using Wiener filtering and adaptive median filtering algorithms. The filtered ROI image is enhanced by using adaptive histogram equalization and image sharpening algorithms. The Otsu image segmentation algorithm based on morphology is used to segment the enhanced ROI image, obtaining the roller image to be detected. The Harris corner detection algorithm is used to extract the features of the roller image, and obtain the position information of the roller. The temperature information of the corresponding position is extracted, and a roller fault detection algorithm based on the relative temperature difference method is used to determine the idler fault. The experimental results show: ① The average accuracy of object detection in the roller ROI extracted by YOLOv5s network model is 99.12%. ② The proposed roller fault detection algorithm has an average accuracy of 97.625% and a frame rate of 16 frames per second for detecting roller faults (no faults, bearing rust, roller jamming, and cylinder wear). ③ The detection results are transmitted to the upper computer through a wireless transmission module, which can display the fault type and key area temperature, and provide an alarm.

     

/

返回文章
返回