Bottom-hat filtering for Defect Detection with CNN Classification on Car Wiper Arm

Abstract

Quality control is an essential process for production as it ensures that product quality is maintained or improved throughout the manufacturing processes. An important component in all modern automobiles is the windscreen wiper which is used to remove rain and snow, ice and even debris from the windscreen. It generally consists of a metal arm, pivoting at one end and with a long rubber blade attached to the other. In the mass production of this windscreen wiper system, consisting of the wiper blade and the wiper arm, their quality is a major factor of competitiveness given that defects on either components will bring negative effect on its market value. As the volume produced is high, it is very difficult to monitor the quality manually. Nevertheless, the current practice is to conduct manual inspection, which leads to high production cost and other quality issues. We have developed an automated defect inspection system which can be implemented in the manufacturing process of car wiper arms using a combination of various image processing techniques together with convolutional neural network (CNN). The goal of this investigation is to detect and to accurately classify wiper arm defects in a very short time. This can improve the quality of the wiper arms shipped to customers and reduce its cost of manufacturing.