One of the most crucial building materials available today is steel, and making flat plates is a difficult procedure. To prevent flaws, steel sheets must go through a comprehensive inspection process before being shipped or delivered. One of the most important jobs for accurately evaluating the quality of a product is the identification and classification of surface imperfections in rolled metal. The objective of this effort is to create a real-time system for identifying and categorizing metal surface defects based on their images. The algorithm aims to increase process efficiency and production standards. This research project addresses the issue of flaw detection on steel surface plates using CNN and computer vision (CV) approaches. Different steel faults are identified and detected by comparing convolutional neural network (CNN) architectures. A comparative examination of various numbers of photographs for training and testing is the output of this work. Models and the selection of an algorithm intended for real-time defect search and classification. One CNN model can be used to produce a tool that significantly eases an individual’s labor.