Pipeline Flange Defect Detection based on Deep Learning

Abstract

Pipe flanges are an indispensable part of industrial piping equipment. Due to the large number of flanges and different installation positions, it will inevitably have a certain negative impact on the quality of manual inspections. Therefore, this paper proposes a pipeline flange visual inspection method based on the improved YOLO v3 algorithm to adjust the network structure: In order to reduce the impact of image shooting scale changes on the detection accuracy, the original network's multi-scale target detection is adjusted to 5 types Scale; At the same time, use the RAdam optimizer to replace the SGD optimizer to improve the training efficiency of the initial stage. Experimental results show that the improved YOLOv3 network can greatly improve the accuracy of flange image recognition and accelerate the convergence speed of the network training phase. It can make full use of the surveillance cameras and intelligent inspection robots in the environment to analyse the flange image, thereby Realize intelligent operation and maintenance.