Surface defect detection and polishing parameter optimization using image processing for G3141 cold rolled steel

Traditionally the surface quality inspection especially for metal polishing purpose is perform by human inspectors. Defect detection is a method of nondestructive testing of material and products to detect defects. This study consists of two parts where the first part is applying vision system to detect and measure surface defects that have been characterized to some level of surface roughness. Specimen of G3141 cold rolled steel is used in this research as it represent the actual material applied in local automotive manufacturer. Gray image of scratch defect on metal surface is detected and information about mean gray pixel value (Ga) is interpreted and converted to surface roughness (Ra) measurement. In this study a new technique is developed where the Ga only read on the specific scratch line without considering the whole image. To realize this, automatic cropping algorithm is developed to detect the region of interest and interpret the Ga value. This techniques will enables the polishing to be done at specific scratch defect area without necessary to develop polishing path throughout the whole surface which is time consuming. Second part is to obtain the optimum polishing parameter by using artificial intelligence technique which is able to predict the grit size, polishing time and polishing force parameter to remove the scratch by polishing process. For the purpose of this study, multiple ANFIS or MANFIS have been selected to predict optimum parameter for polishing parameters. Polishing parameter data can be generated by using MANFIS to predict optimum polishing parameters such as grit size, polishing time and polishing force in order to perform polishing process. However due to lack of study done in the field of flat and dry polishing, the polishing parameter data have to be developed. The polishing parameter data for flat and dry polishing is performed by using robotic polishing arm and the experiment runs design by using full factorial design. Results show that the defect detection algorithm able to detect defect only on the scratch area and able to read the Ga value at detected scratch line and transform it to surface roughness measurement at considerably good level of accuracy compared with manual method. Results from MANFIS have shown that the system is able to predict up to 95% accuracy which is considerably high. The overall results from both parts of this research would inspire further advancements to achieve robust machine vision based surface measurement systems for industrial robotic processes specifically in polishing process.

Preview Text FK 2016 164 IR.pdf Download (1MB) | Preview

Actions (login required)

View Item