Deep learning-based automated characterization of crosscut tests for coatings via image segmentation

Abstract

A manual scratch test to measure the scratch resistance of coatings applied to a certain substrate is usually used to test the adhesion of a coating. Despite its significant amount of subjectivity, the crosscut test is widely considered to be the most practical measuring method for adhesion strength with a good reliability. Intelligent software tools help to improve and optimize systems combining chemistry, engineering based on high-throughput formulation screening (HTFS) technologies and machine learning algorithms to open up novel solutions in material sciences. Nevertheless, automated testing often misses the link to quality control by the human eye that is sensitive in spotting and evaluating defects as it is the case in the crosscut test. In this paper, we present a method for the automated and objective characterization of coatings to drive and support Chemistry 4.0 solutions via semantic image segmentation using deep convolutional networks. The algorithm evaluated the adhesion strength based on the images of the crosscuts recognizing the delaminated area and the results were compared with the traditional classification rated by the human expert.