Neural application Yolo network for defect detection on seamless rings

Currently, product quality control is the main focus for domestic enterprises. Timely product control allows to prevent potential losses, and also reduces the risk of claims from consumers. Unfortunately, due to the large volume of production and lack of qualified employees, not all domestic enterprises can provide one hundred percent product control. Foreign companies have long ago found a way out of this situation. Large enterprises such as “Porshe” and “BMW” use systems based on artificial intelligence to control products. For Russian enterprises, the introduction of artificial intelligence is relatively new, but popular direction. The purpose of the research is to study the possibilities of using neural networks to identify defective products. In the course of the study was studied AI.SEE technology, analyzed the production technology of seamless rings (highlighted the shortcomings of production), as well as the method of training neural network YOLO. Also, the following task was set and solved: to develop a model (based on computer vision) that will detect, and visualize the presence of a defect on seamless rings. The conducted research showed that the model trained on off-the-shelf YOLO scales can determine the presence of defects on ring forgings with a high accuracy of 80…90 %. But, it should be noted that, despite the high accuracy of defect detection, the model requires a number of modifications for more stable and qualitative work. The use of such a model can significantly improve the control processes in production. It is worth noting that the results obtained can be useful for companies engaged in the production of rings, and contribute to improving the efficiency and reliability of production processes, for further implementation of such a system.