UDC 004.932
DETECTION OF CERAMIC COATING DEFECTS ON SMALL-SIZED PRODUCTS USING CONVOLUTIONAL NEURAL NETWORKS FOR INDUSTRIAL QUALITY CONTROL
S. V. Spitsyn, PhD, Leading Software Engineer, JSC Morinsystem-Agat-KIP, Associate Professor, RSREU, Ryazan, Russia;
orcid.org/0000-0001-6468-5043, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
S. S. Rumyancev, Head of the State Defense Order Support Department, JSC Morinsystem-Agat-KIP, Ryazan, Russia;
orcid.org/0009-0006-6102-3902, This email address is being protected from spambots. You need JavaScript enabled to view it.
The article considers the technology of using convolutional neural networks for automated detection and classification of defects on ceramic coatings of small-sized products in industrial quality control condi tions. The results of experiments confirming the effectiveness of proposed approach including metrics for assessing the quality of ML model and their numerical indicators are presented. The characteristics of video camera used to obtain images, training sizes; test and validation samples are also presented. The relevance of study is the necessity to improve the efficiency of quality control on industrial production scale where tra ditional methods of visual inspection are quite labor-intensive and give subjective results. The aim of the work is to increase accuracy and speed of quality control of small-sized ceramic products using advanced machine learning methods. During the study the following tasks were solved: collecting and marking of ceramic coatings images dataset obtained from high-resolution monochrome video camera via GigE Vision interface; training ML model to search for cracks, chips, stains and other types of ceramic defects, optimizing model hyperparame ters; evaluation of model quality using generally accepted classification and regression metrics, such as av erage accuracy rate mAP, precision, recall, and F1-score. The results obtained demonstrate fairly high effi ciency of the proposed technology for automating the quality control of small-sized ceramic products which allows us to significantly reduce inspection time compared to «manual» inspection.
Key words: : machine learning, defect detection, industrial quality control, computer vision, convolutional neural networks.