Improved Faster R-CNN algorithm for defect detection in powertrain assembly line

Abstract

Inspection is a vital part of quality control in manufacturing systems. Some edges are difficult to extract because of the brightness and contrast caused by lighting equipment. To solve this problem, an improved Faster R-CNN algorithm is proposed. Based on the original framework, the anchors are optimized by clustering and RoI Pooling is replaced by RoI Align to improve the detection capability of sand inclusion defects. Training and testing on the image dataset of engine surface, the algorithm is verified to give the type and location of the defects with high accuracy and efficiency.