{"title":"New Algorithm to Register Defects through Automated Non-Destructive Testing in Products Obtained Using Selective Laser Sintering","authors":"A. S. Sergeev","doi":"10.1134/S2075113326700772","DOIUrl":null,"url":null,"abstract":"<p>The issue of detecting defects is explored for products made using the selective laser sintering during automated ultrasonic testing. An adaptive algorithm is proposed to detect and localize defects in automated ultrasonic quality control. It is based on combined use of recurrent neural network methods with long-term short-term memory (LSTM). It allows localizing defects based on the amplitude of the ultrasonic signal exceeding a given threshold value. A convolutional neural network (YOLOv5 architecture) is used to classify defects based on the B-scan image. The algorithm automatically determines the type, size, and location of defects, including the depth. The developed algorithm is capable of quickly matching the software settings for any type and size of control objects, provided the matching is made using artificially generated defects.</p>","PeriodicalId":586,"journal":{"name":"Inorganic Materials: Applied Research","volume":"17 4","pages":"1141 - 1144"},"PeriodicalIF":0.4000,"publicationDate":"2026-08-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Inorganic Materials: Applied Research","FirstCategoryId":"1085","ListUrlMain":"https://link.springer.com/article/10.1134/S2075113326700772","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q4","JCRName":"MATERIALS SCIENCE, MULTIDISCIPLINARY","Score":null,"Total":0}
引用次数: 0
Abstract
The issue of detecting defects is explored for products made using the selective laser sintering during automated ultrasonic testing. An adaptive algorithm is proposed to detect and localize defects in automated ultrasonic quality control. It is based on combined use of recurrent neural network methods with long-term short-term memory (LSTM). It allows localizing defects based on the amplitude of the ultrasonic signal exceeding a given threshold value. A convolutional neural network (YOLOv5 architecture) is used to classify defects based on the B-scan image. The algorithm automatically determines the type, size, and location of defects, including the depth. The developed algorithm is capable of quickly matching the software settings for any type and size of control objects, provided the matching is made using artificially generated defects.
期刊介绍：
Inorganic Materials: Applied Research contains translations of research articles devoted to applied aspects of inorganic materials. Best articles are selected from four Russian periodicals: Materialovedenie, Perspektivnye Materialy, Fizika i Khimiya Obrabotki Materialov, and Voprosy Materialovedeniya and translated into English. The journal reports recent achievements in materials science: physical and chemical bases of materials science; effects of synergism in composite materials; computer simulations; creation of new materials (including carbon-based materials and ceramics, semiconductors, superconductors, composite materials, polymers, materials for nuclear engineering, materials for aircraft and space engineering, materials for quantum electronics, materials for electronics and optoelectronics, materials for nuclear and thermonuclear power engineering, radiation-hardened materials, materials for use in medicine, etc.); analytical techniques; structure–property relationships; nanostructures and nanotechnologies; advanced technologies; use of hydrogen in structural materials; and economic and environmental issues. The journal also considers engineering issues of materials processing with plasma, high-gradient crystallization, laser technology, and ultrasonic technology. Currently the journal does not accept direct submissions, but submissions to one of the source journals is possible.
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号
