Detecting Surface Defects in Products of The Industrial Manufacturing Process Using Machine Learning and Image Processing

Abstract

Detection of surface defects in planar surfaces in the industrial manufacturing process is an ongoing area of research. It has always been burdensome to guarantee the absolute flawlessness of surfaces of carbon and iron alloys, or any surface for that matter. Quality of appearance, or rather, the surface of industrial products like metal sheets is being held at a particularly high standard especially of late, to meet these standards and to ensure that customer requirements are met, use of computer vision based solutions are encouraged, these solutions essentially entail using a 2D or 3D defect detection technique like for instance edge detection, these algorithms combined with a machine learning model is what will detect and identify defects on the various planar surfaces. The model will be trained with a dataset of images that will contain damaged as well as undamaged surfaces, this is to make sure that the model can identify a defect and in the absence of a defect the model should recognize the surface as homogeneous. As long as the model is trained properly, these methods have been proven to be exceptionally robust, and as a solution they have been prevalent of late and over the past few years. CNN and neural networks based on CNN architectures were used to train the model in order to classify the defects.