Surface Defect Detection in Steel Plates Using Machine Vision

Abstract

Surface defect detection and classification in small size steel plates using machine vision inspection has been researched and presented in this paper. The steel plates that are used in the automobile drive chains for power transmissions are used as the test parts. The components of chains undergo the manufacturing steps such as punching, heat treatment and polishing before the final assembly. Due to a high temperature and loading conditions, sometimes the plates get distorted. The research project was set out to detect, classify and sort non-defective and defective plates into separate bins using machine vision inspection. A database of non-defective and defective parts was generated using the vision acquisition program. The number of images in the database were increased using data augmentation techniques. Various vision inspection techniques such template matching based object classification implemented using NI Vision Builder Software and Convolutional Neural Networks (CNN) implemented in MATLAB have been trained and tested. The vision algorithms communicated pass or fail decision through the serial communication and controlled the sorting system using a microcontroller. Depending on the decision, the parts were sorted into accepted (non-defective parts) or rejected (defective parts) bins. The project was successful in achieving the milestones defined in the research objective. The results with the two inspection methods, Object classification and CNN based method, have been reported. A classification accuracy of 96 % was achieved using the object classification method, while the CNN based method resulted in the 100 % classification accuracy. The developed algorithms can be implemented on a high-speed smart camera and the system can be used for realtime online inspection.