An image processing based product defect detection system for metal industry

Product defect detection based on digital systems increases production speed and reduces production cost. In the metal industry, especially in sheet metal cutting operations, the detection of defective products is usually done by the operators. Developing digital system based product defect detection systems for this sector will reduce operator related errors and enable more accurate product defect detection. In this thesis, a product defect detection system which is based on image processing has been developed for the metal sector. The development process is carried out in two stages. In the first stage, morphological operations and Hough transform are used in a personal computer to compare the defect detection performances. As a result of the experiments, it has been observed that both approaches have more than 80% accuracy rates. Morphological operations achieves 81% while circular Hough transform has 89% accuracy rates. In the second stage Raspberry Pi Model 3 B+ is selected as an embedded system and we consider the effect of the camera resolution on the performance by applying circular Hough transform which has higher accuracy rate in the first stage. As the product group, sheet metal plates which are produced frequently in the sector and containing circular holes have been selected. The information about the circular holes on the plates is extracted from the images taken from the camera using image processing methods. The obtained information is compared with the reference information and it is checked whether it is within the specified tolerances. The images taken with 10MP camera resolution are applied to the embedded system by using circular Hough transformation and 96.29% accuracy rate is obtained. As a result of the experiments, it has been seen that the developed system can be used in quality control applications when the accuracy rates are taken into consideration.