Physics-Enhanced Data-Driven Models for Predicting the Fatigue Response of Additively Manufactured Metals

Alberto Ciampaglia (Politecnico di Torino)

Abstract:

The accurate prediction of mechanical behavior in advanced materials, such as additively manufactured metals, is crucial for advanced structural engineering applications. This talk explores the integration of physics-informed data science methods with experimental data to enhance the reliability and robustness of mechanical response predictions. Specifically, the presentation will focus on the development of hybrid models for the assessment of the fatigue performance of additively manufactured metals produced under varying manufacturing conditions. By embedding physical laws into data-driven frameworks, these models achieve improved generalization and interpretability. Additionally, it highlights the implementation of probabilistic machine learning techniques to generate statistical distributions of material response, enabling the evaluation of prediction reliability and uncertainty. Case studies will highlight the practical implementation of these methods and discuss their potential to redefine engineering design and analysis. Finally, the talk discusses how these predictive models can be integrated into optimization frameworks, paving the way for advanced design processes tailored to the unique properties of additively manufactured materials.