VisionLabs computer vision technologies will help Evraz save millions of rubles a year

2023: Introduction of intelligent video analytics from VisionLabs

Evraz has introduced VisionLabs intelligent video analytics at the West Siberian Metallurgical Plant to automate quality control. Using computer vision technologies, the system automatically detects surface defects on steel workpieces. This was announced on July 24, 2023 by VisionLabs. Since the beginning of the project, the implemented video analytics has already helped to save more than 20 million rubles at one rolling mill.

For productions high-quality rolled products, it is necessary in time to detect defects on steel blanks. Such control at Evraz was carried out visually by technological shop staff and quality management employees. Since steel blanks are stored tightly to each other, time in the rounds, employees did not have the opportunity to conduct an inspection from all sides. In addition, when loading into the furnace landing line, the speed of the workpieces is too high and is 2 m/s, which does not allow enough time for visual inspection. The introduction of a system for detecting and recognizing defects based on convolutional neural networks from VisionLabs helped to automate the process. The installation cameras and use of video analytics made it possible to analyze every centimeter of steel blanks.

The production process is structured as follows. The billet follows the loading roller table of the mill. Prior to weighing, billet passes surface defect control point, where video cameras are installed for fixing state of billet surface of each of four sides. When the workpiece enters the field of view of the cameras, neural networks detect it and check for defects.

To train video analytical algorithms, the VisionLabs team assembled a datacet consisting of several thousand photographs of blanks marked by type of defect. After that, the images were submitted to the input of the neural network, which was trained on the basis of the markup made. It was also necessary to take into account that the blanks move along the roller table at high speed. For instant fixation of even the smallest parts in the project, specialized machine vision cameras were used.

When a defect is detected in real time, an audible signal is sent to the operator, and the image and sequence number of the defective workpiece are displayed. This makes it possible to reject them in time and send them not to rental, but to process the defective area. The system classifies the detected defects for further analysis and calculates the number of accepted blanks. The collected statistics can be used to change the process plan of previous workshops.

The implemented video analytics from VisionLabs automatically detects more than 95% of the scrap, depending on the face of the workpiece and the type of defect. At the same time, it was possible to achieve a low rate of false positives - no more than once an hour with a total flow of about 1000 blanks per day. The solution helped to reduce the percentage of errors of technological personnel in identifying defects in workpieces, to level the impact of the human factor on the technological process, as well as to reduce financial losses and, due to the rational use of raw materials, to reduce the consumption factor of production.

Digital transformation is the main trend in metallurgy in Russia. Technologies improve production processes and improve the efficiency of enterprises. Evraz systematically implements AI solutions at its mills, and video analytics is one of the key areas. Automatic quality control of workpieces based on VisionLabs technologies increases the efficiency of the technological process and reduces the number of downtime, thereby helping to produce an additional volume of products. told Pavel Tsygankov, head of the mid-rate workshop, EVRAZ ZSMK JSC.

The application of computer vision in industry is interesting in that cases constantly arise in the industry. Often, the production processes of enterprises are special, so the most difficult thing in such projects is to understand at what stage it is necessary to embed video analytics, as well as collect data for training neural networks. The VisionLabs team coped with the tasks. The implemented system helped to reduce the downtime of the rolling mill, which, in turn, stabilizes the production of finished products, told Dmitry Markov, CEO of VisionLabs.