Machine Learning approach for Defect Identification in Machinery parts

Abstract

The prominent factor affecting the quality of metals are the various kind of defects present on their surfaces. Identifying these defects and taking remedial measures to overcome the defects is of paramount importance to maintain quality. Manual inspection of defects is a tedious process and may sometimes be inaccurate. The objective of this paper is to study various classification techniques and their performance in identifying rust from the metal surfaces. Auto color correlogram has been used on the images for feature extraction. We have evaluated the performance of 13 different classification techniques and they have been compared on the basis of their accuracy and error rates. Accuracy in the range of 95% - 97% was obtained by classification techniques like Bagging, LogitBoost and ensemble method such as Random Forest, whereas J48 gave the least error rate.