Detection and classification of defects of rolling origin on the ends of sleeve using a convolutional neural network

In the process of manufacturing seamless pipes, great attention is paid to the quality of products, which in turn consists of the quality of the workpiece, the quality of the sleeve and the quality of the draft pipe. At the same time, various shape defects can occur at the rear end of the sleeves, which can prevent the introduction of the mandrel and subsequent rolling of the sleeve in a continuous rolling mill. Also, shape defects on the rear end of the sleeve (“crown”, “earring”, “thorn”, etc.) can provoke an emergency stop. Timely informing the millers about the type of the butt end of the sleeve can contribute to the prompt decision-making on adjusting the settings of the piercing mill. Which in turn will lead to the prevention of the appearance of products with critical deviations in shape. One of the ways to obtain information about an object (pipe end) is possible with the use of digital cameras. Therefore, it is necessary to analyze the current possibilities of using digital cameras and digital image processing methods in order to determine the characteristics of interest.