Synthesis of the artificial intelligence and model-based and statistical algorithms in the classification of the metal surface defects

MODEL ALGORITHMS FOR SURFACE DEFECTS

Abstract

Steel is an essential material used in many industries, particularly in architecture, aerospace and automotive industry and has been by far one of the most important components in manufacturing. Possible defects in the steelmaking process could significantly affect the quality and service life of the final product. With the aim of ensuring a timely response in steel production, in this paper we propose a model for the classification, defect region detection and visualization of spatial defects. The model is based on the synthesis of CNN, snake algorithms and algorithms for generating spatial defects based on images. The CNN is trained using images from the NEU Surface Defect database, and model evaluation is performed on unknown samples that were not used for training. CNN produced a high percentage of accuracy (in general and in individual cases). After the classification, a spatial representation of the damage is generated and the segmentation of the defect on the material is performed. The application of this model in modern industry could significantly increase performance and quality of high-risk manufacturing jobs, prevent unnecessary losses and allow making timely decisions about future steps in a more insightful way.