Though manufacturing is becoming increasingly technologically advanced, statistical destructive and nondestructive evaluation (NDE) are still the dominant methods for quality control and Industry/NDE 4.0 is still not fully realized. Ultrasonic inspection systems are increasingly used, but there is still a need for fast, automated, and accurate data interpretation. To this end, the IDIR has developed an approach for ultrasonic B-scan interpretation using deep learning (DL) which is a form of artificial intelligence (AI) using deep artificial neural networks to automatically learn from data. DL forms the state of the art in many problem domains in e.g. natural language processing and computer vision, hence it has become increasingly, and often successfully, applied in NDE. Our aim was to investigate a DL approach to automatic characterization of ultrasonic B-scans. We experimented on ultrasonic B-scans from resistance spot welding (RSW) because we could rapidly generate a large dataset of samples using this process. We developed and labelled a dataset of ultrasonic B-scans from RSWs of varying parameterizations, along with important metadata (e.g. sheet thicknesses, weld time, etc.), and subsequently trained DL models for object detection on the labelled samples. The resultant AI system conducts a morphological analysis of the weld geometry after the weld is completed. Using an object detection approach, we created models that exhibit high detection rates with extremely low false positive rates, while accurately measuring the position of the nugget within the welded stack. Our work shows the applicability of DL in real-time NDE data interpretation. Such AI-based systems can be combined with ultrasonic NDE to comprehensively, accurately, and practically instantly characterize 100% of parts without human intervention, representing a major step toward Industry/NDE 4.0 and zero-defect RSW.
(1) Ouellette, A., Karloff, A.C., Perez-Regalado, W., Chertov, A.M., Maev, R.Gr., Lichaa, P., 2013, Real-Time Ultrasonic Quality Control Monitoring in Resistance Spot Welding: Today and Tomorrow, Materials Evaluation, V. 71, No. 7.
(2) Maev, R. Gr., Chertov, A.M., Perez-Regalado, W., Karloff, A., Tchipilko, A., Lichaa, P., Clement, D., Phan, T., 2014, In-Line Inspection of Resistance Spot Welds for Sheet Metal Assembly, Welding Journal, pp. 58-62.
(3) Ultrasonic In-Process Monitoring And Feedback Of Resistance Spot Weld Quality, United States Patent 9,296,062 B2. Inventors: Maev, Roman Gr., Chertov, Andriy M., Paille, John M., Ewasyshyn, Frank J. Issued 29 March 2016.
(4) Goodfellow, I., Bengio, Y., Courville, A. Deep learning, MIT press, 2016.
(5) Cantero-Chinchilla, S., Wilcox, P.D., Croxford, A.J., 2021, Deep learning in automated ultrasonic NDE –developments, axioms and opportunities, arXiv Preprint. https://arxiv.org/abs/2112.06650
(6) Ren, S., He, K., Girshick, R., Sun, J., 2016, Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks, IEEE Transactions on Pattern Analysis and Machine Intelligence, 39(6), pp. 1137-1149.
(7) Liu, W., Anguelov, D., Erhan, D., Szegedy, C., Reed, S., Fu, C.Y., Berg, A.C., 2016, SSD: Single shot multibox detector. In European Conference on Computer Vision, pp. 21-37.
(8) Redmon, J., Farhadi, A., 2018, YOLOv3: An incremental improvement. arXiv preprint.
(9) Santiago, A., Solaque, L., Velasco, A., 2020, Deep Learning Algorithm for Object Detection with Depth Measurement in Precision Agriculture. In Proceedings of the 17th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2020), pp. 490-497.
(10) Dirir, A., Ignatious, H., Elsayed, H., Khan, M., Adib, M., Mahmoud, A., Al-Gunaid, M., 2021, An Advanced Deep Learning Approach for Multi-Object Counting in Urban Vehicular Environments. Future Internet, 13,
84 Page Views
0 PDF Downloads
0 Facebook Shares