NSF Award
2412395
Compared to traditional manufacturing processes such as machining, laser powder bed fusion (LPBF)-based metal additive manufacturing (AM) offers an opportunity for making complex metal components with design freedom, short development time, and environmental sustainability. However, the LPBF fabricated components often suffer from severe fatigue scattering problems, that is, the fatigue life of a component produced by LPBF under similar process conditions exhibits a very large variation. Fatigue scattering imposes a significant challenge to using an LPBF process for fabricating load-bearing and highly reliable components. This significantly limits the applicability of metal AM processes. To address this limitation, the objective of this project is to establish a physics-informed machine learning (PIML) framework, which integrates the physical knowledge of fatigue and the measured data to enable accurate and transparent predictions of fatigue life and its variation. Based on the PIML framework, process design optimization can be achieved to mitigate the fatigue scattering. The new knowledge and modeling methods obtained from this project will bring disruptive impacts on the AM industry by providing an enabling predictive tool for the fatigue life and scattering of printed materials. The scattering mitigation strategy facilitates printing consistent high-quality components in batch or mass production. This project will also contribute to workforce training by promoting interdisciplinary research at the intersection of AM, fatigue mechanics, and machine learning and provide unique training opportunities and learning testbeds for students.<br/><br/>To achieve the project objective, baseline fatigue samples as-printed via LPBF, and the post-processed (i.e., hot isotropic pressing) alloys, including SS316L and Ti-6Al-4V alloys will be fabricated. The sample quality including surface finish, geometrical defects, residual stress, and microstructure will be characterized. Then high-frequency and load-varying resonance-based fatigue testing (up to 20 million cycles/day) will be conducted to obtain the data of fatigue initiation, development, and fracture behaviors. With the experimentally obtained fatigue data, the PIML framework is established to integrate the governing phenomenological laws or physics-driven fatigue laws and uncertainty quantification with data-driven deep neural networks to enable the accurate, data-efficient, and interpretable prediction of fatigue life, scattering band, and dynamic fatigue behavior. A scattering mitigation strategy will be established as well using Bayesian optimization based on the established process-quality-fatigue (P-Q-F) model. If successful, this project can generate new understanding in the P-Q-F relationship of LPBF, which will advance the metal AM industry.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
