Automatic Detection of Industrial Wire Rope Surface Damage Using Deep Learning-Based Visual Perception Technology

Surface wear, which is most likely to occur in early damage of wire ropes (WRs), is a serious threat to WR safety. Visual perception technology (VPT) can intuitively grasp the surface damage situation of WRs. However, efficient detection of WR surface damage from complex morphological characteristics using VPT has always been a challenging task, and there are various problems, such as nonstandard data, low automation/intelligence, and poor detection effect. To overcome the above difficulties, this study proposes a deep learning-based VPT framework, which relies on an image preprocessing (IP) scheme and a deep convolutional neural network (DCNN), called WR-IPDCNN. The IP scheme is designed to remove the influence of the image background and to normalize data, including posture adjustment and region of interest (ROI) extraction. The improved DCNN based on LeNet-5 is proposed to mine the newly established WR data set, which considers different working conditions. Experimental results demonstrate that the proposed framework can accurately extract the ROI area of WR images and achieve 95.55% detection accuracy, which is 12.44% higher than LeNet-5, and a significant improvement on the automation/intelligence level in this field.