Hardware and software system for data collection, detection, classification of surface defects and visualization of results based on computer vision for large-diameter pipes

Technologically complex and potentially dangerous industries, which include ferrous metallurgy, have a number of tasks that can potentially be solved using artificial intelligence methods. One of these tasks is the problem of timely detection of surfacedefects in large-diameter pipes. A number of requirements are put forward for a computer vision model designed to facilitate the work of a flaw detector: timely detection, correct classification and a low number of false positives. The paper presents a methodology for developing a system for detecting and classifying surface defects based on computer vision for large-diameter pipes. As a result of the analysis of publications on the topic of defect recognition using computer vision methods and based on data obtained in real production conditions at the site of the final inspection of pipes of the finishing section of the Vyksa OMK plant, general approaches to software design, metrics for assessing detection and classification models and the sequence of personnel actions are formulated. The paper presents a comparative analysis of various approaches to building computer vision models. A scheme for calculating the resolution and number of cameras for efficient use of technical capacities is proposed. The software architecture together with the proposed approach to marking a training set allows for more productive and efficient collection of the training set. The developed methods and approaches are currently being tested and contribute to the accumulation of data for subsequent evaluation of their effectiveness.