Research output: Contribution to conference › Abstract › peer-review
Machine Learning in Defectoscopy: DBSCAN for Localization, Agglomerative Clustering for Typification. / Григорьева, Анастасия Викторовна; Литвинов, Юрий Викторович.
2025. 58-63 Abstract from 2025 International Russian Smart Industry Conference (SmartIndustryCon).Research output: Contribution to conference › Abstract › peer-review
}
TY - CONF
T1 - Machine Learning in Defectoscopy: DBSCAN for Localization, Agglomerative Clustering for Typification
AU - Григорьева, Анастасия Викторовна
AU - Литвинов, Юрий Викторович
PY - 2025/3/24
Y1 - 2025/3/24
N2 - Acoustic emission (AE) is one of the most relevant and cost-effective non-destructive testing methods for identifying defects in industrial objects. This paper addresses two primary tasks in defectoscopy: locating hazardous AE zones and classifying signals according to the presumed defect type. Through our research, we have determined and experimentally confirmed that the most suitable machine learning method for localization is a modified DBSCAN, while Agglomerative Clustering is optimal for typification. This paper details the specific tuning and modifications applied to these methods for AE. The resulting models were successfully tested on real-world data. The accuracy of the localization results was assessed using statistical analysis and consultations with AE experts, while typification was validated probabilistically and tested on sample data.
AB - Acoustic emission (AE) is one of the most relevant and cost-effective non-destructive testing methods for identifying defects in industrial objects. This paper addresses two primary tasks in defectoscopy: locating hazardous AE zones and classifying signals according to the presumed defect type. Through our research, we have determined and experimentally confirmed that the most suitable machine learning method for localization is a modified DBSCAN, while Agglomerative Clustering is optimal for typification. This paper details the specific tuning and modifications applied to these methods for AE. The resulting models were successfully tested on real-world data. The accuracy of the localization results was assessed using statistical analysis and consultations with AE experts, while typification was validated probabilistically and tested on sample data.
U2 - 10.1109/smartindustrycon65166.2025.10986085
DO - 10.1109/smartindustrycon65166.2025.10986085
M3 - Abstract
SP - 58
EP - 63
T2 - 2025 International Russian Smart Industry Conference (SmartIndustryCon)
Y2 - 24 March 2025 through 28 March 2025
ER -
ID: 145533404