Abstract:

During the hot rolling process, some defects may be formed in inner and outer levels on product. Eddy Current and Ultrasonic methods are used for the detection of internal defects, the penetrant and visual inspection methods are used for the outer defects on surface. For these methods careful and time-consuming process are required for the detection of product defects. These methods can only be applied to product after rolling process. But many flaws can be seen on surface without applying these kind of techniques. To detect surface defects of hot rail images have been obtained on line with the help of a CMOS camera. During the passage of each product, CMOS camera takes approximately 500 images. These images have to be processed quickly before arrival of the following product to mill. Reading and processing of a lots of images saved in a directory at the same time can be heavy workload.Some unrelated pixels have been seen around the detected defect on these images. A cleaning process with erosion was carried out to eliminate these unwanted pixels. Eroding process on the image used for limiting the areas marked on the grey images is the one of the most basic and important mathematical morphological operations. Morphological operation is used to identify the defected region on the image.In this study, fast cleaning process on image is done on GPU (Graphics Processing Unit) applying parallel processing by using OpenCV. CPU (Central Processing Unit) and GPU performance are compared afterwards. According to the results obtained, several images can be processed3 times faster by parallel processing.

Keywords: