轨道扣件在运营过程中会出现松动甚至掉落、断裂等异常情况,不利于列车行驶稳定和安全,需要进行定期、及时的检查与维修.传统的人工巡检效率低,难以匹配我国轨道交通的快速发展,且对于部件松动等不易察觉的问题检测效果差.利用计算机视觉形成自动化的检测设备逐渐成为发展趋势,其中基于三角测量原理的线结构光技术因其成本低、精度高、速度快等优点得到广泛应用,且适合轨道检测场景.该技术核心设备为可以采集并分析线结构光进行三维重建的3D相机,基于成像原理设计可搭载于轨道检测车的扣件检测系统并进行现场试验,经过数据分析和处理可以分别得到高质量的图像数据和三维模型.针对图像数据利用目标检测的方法,构建数据集,搭载YOLO(You Only Look Once)v5深度学习模型,实现挡肩及扣件部件的快速识别,进行部件丢失检测;针对三维模型利用轨道扣件相对位置固定的特点,根据阈值筛选扣件数据并进一步得到弹条及螺栓等部件的坐标信息,通过边缘提取、平面拟合等方法计算位移量,进行部件松动检测.研究结果表明,检测系统可以采集高质量的扣件数据,扣件部件识别平均精准度达到99.0%,速度满足现场实时检测的要求,同时对于弹条和螺栓的松动量检测精度分别达到了1 mm和0.1 mm.该方法具有实际工程价值,可以大幅提升轨道巡检效率,对于扣件部件丢失、松动等严重问题可以及时预警指导修复,保障轨道安全服役性能.
Intelligent lost and loose detection of track fastener components based on 3D camera
During railway operations,track fasteners may develop defects such as loosening,falling off,or breaking,which can impair the stability and safety of train travel.Regular and prompt inspection and maintenance of track fasteners is necessary.However,traditional manual inspections are inefficient and cannot keep up with the rapid development of China's rail transit.Additionally,detecting defects such as loose components is challenging.Consequently,using computer vision to develop automated detection systems has become a trend.Among them,line structured-light technology based on the principle of triangulation is widely used due to its low cost,high precision,and fast speed,and is well-suited for rail inspection scenarios.The primary device utilized is a 3D camera that can capture and analyze line structured light information for 3D reconstruction.Based on the imaging principle,a fastener detection system was designed and installed on a rail inspection vehicle.High-quality image data and 3D model were obtained through data analysis and processing after field testing which is applied to verify inspecting performance.For image data,a dataset was created using object detection methods,and a YOLO(You Only Look Once)v5 deep learning model was used to achieve fast recognition of track shoulder and fastener components.For 3D model,the fastener data was filtered according to a threshold based on the fixed relative position.The coordinate information of components such as strips and bolts was further obtained.Loose components detection was performed through displacement calculating using methods such as plane fitting.The results indicate that the detection system can acquire high-quality fastener data.The detection average precision for fastener components loss can achieve 99.0%with a fast detection speed which means real-time on-site detection is feasible,and the accuracy of detecting looseness in strips and bolts reach 1mm and 0.1 mm,respectively.The proposed method holds significant practical engineering value,as it has the potential to substantially enhance track inspection efficiency.Notably,in instances of critical defects such as lost or loosening of fastener components,this approach can provide timely warnings and facilitate prompt repairs,ensuring optimal track safety performance.
万方数据
维普
