Data driven method for predicting the effect of process parameters on the fatigue response of additive manufactured AlSi10Mg parts. (May 2023)

• Record Type: Journal Article
• Authors: Ciampaglia, A; Tridello, A.; Paolino, D.S.; Berto, F.
• Is Part Of: International journal of fatigue. Volume 170(2023)
• Publication Date: 2023-05
• Subjects: Fatigue -- Machine Learning -- Additive manufacturing -- Physic informed -- Neural networks; Materials -- Fatigue -- Periodicals; Materials -- Fatigue; Periodicals; 620.1122
• DOI: 10.1016/j.ijfatigue.2023.107500
• Languages: English
• ISSNs: 0142-1123

Abstract

Highlights: Additive Manufactured AlSi10Mg fatigue response estimated with Machine Learning. A Physics Informed Neural Network (PINN) inspired to Murakami's model is proposed. PINN performs better than simple Feed Forward Neural Networks (FFNN) SLM AlSi10Mg fatigue response is estimated from process parameters with PINN. Abstract: The fatigue response of Additive Manufacturing (AM) components is driven by manufacturing defects - whose size mainly depends on process parameters - and by the resulting microstructure - mainly affected by heat treatments and process parameters. In the paper, Machine Learning (ML) algorithms are applied to estimate the fatigue response from AM process parameters and heat treatment properties. Feed-forward neural networks (FFNN) and physics-informed neural network (PINN) algorithms are designed and validated on literature datasets of AM AlSi10Mg alloy, proving the effectiveness of physics-based ML approaches in predicting the fatigue response of AM parts. Leveraging PINN interpretability, the authors analyse the relationship between process parameters and fatigue response.