PSI - Issue 75

Kris Hectors et al. / Procedia Structural Integrity 75 (2025) 102–110 Hectors et al. / Structural Integrity Procedia (2025)

110

9

Fig. 15: Scatter plot of the predicted values and the SCF values obtained through FEA, highlighted for specimen N35. 5. Conclusions This work focused on leveraging machine learning for the prediction of stress concentration factors ( ) in V notched cylindrical specimens, utilizing geometric data acquired through 3D scanning. A comprehensive database was generated through scanning of 175 unique specimens. Finite element analyses were performed on as-machined models derived from the scanned notch profiles. A bottom-up meshing strategy was adopted to obtain high-quality hexahedral meshes. A critical outcome of this study is the demonstrated superiority of predictions derived from FEA incorporating the as-machined geometry compared to those based on nominal or simplified measured parameters (e.g., notch opening angle, root radius). This underscores the necessity of accounting for geometric deviations inherent in manufacturing processes. Subsequent development of predictive models showed that data preprocessing is essential for strong performance, particularly with limited data sets. In this study, feature selection using recursive feature elimination resulted in only marginal performance gains. Voting regression and gradient boosting regression algorithms yielded the most accurate predictions among the considered ML models. Conclusively, this work establishes the efficacy of combining 3D scanning, FEA, and machine learning to accurately estimate stress concentrations reflective of real-world component geometry. This study confirms the potential of this data-driven approach for enhancing the fidelity of structural integrity assessments. Future work should focus on addressing the main limitations of this work. Extending the work to different notch geometries and a larger range of -values. Furthermore, considering 3D notch geometry rather than simplifying the notch to a 2D profile. R EFERENCES Hectors, K., & De Waele, W. (2023). Fatigue block loading experiment database [Dataset]. 10.17605/OSF.IO/6Y5SD Hectors, K., Vanspeybrouck, D., Plets, J., Bouckaert, Q., & De Waele, W. (2023). Open-Access Experiment Dataset for Fatigue Damage Accumulation and Life Prediction Models. Metals , 13 (3), 621. https://doi.org/10.3390/met13030621 Nath, D., Ankit, Neog, D. R., & Gautam, S. S. (2024). Application of Machine Learning and Deep Learning in Finite Element Analysis: A Comprehensive Review. Archives of Computational Methods in Engineering , 31 (5), 2945 – 2984. https://doi.org/10.1007/s11831-024-10063-0 Niederwanger, A., Warner, D. H., & Lener, G. (2020). The utility of laser scanning welds for improving fatigue assessment. International Journal of Fatigue , 140 . https://doi.org/10.1016/j.ijfatigue.2020.105810 Noda, N., & Takase, Y. (2006). Stress concentration formula useful for all notch shape in a round bar (comparison between torsion, tension and bending). International Journal of Fatigue , 28 (2), 151 – 163. https://doi.org/10.1016/j.ijfatigue.2005.04.015 Radaj, D., Sonsino, C. M., & Fricke, W. (2006). Fatigue Assessement of Welded Joints by Local Approaches (M. Grant, Ed.). Woodhead Publishing. Sonsino, C. M., Fricke, W., De Bruyne, F., Hoppe, A., Ahmadi, A., & Zhang, G. (2012). Notch stress concepts for the fatigue assessment of welded joints — Background and applications. International Journal of Fatigue , 34 (1), 2 – 16. https://doi.org/10.1016/j.ijfatigue.2010.04.011