Abstract: Data-driven models have been widely used to predict the fatigue life of additive manufactured materials because of their powerful capability in high-dimensional nonlinear modeling. However, owing to the high requirement of data support, these models are not economically feasible for many practical cases. To tackle this issue, a transfer learning-based model for fatigue life prediction of additive manufactured materials is proposed, through combining source and target domain data and introducing a clustering-based procedure for hyper parameter optimization. The proposed model can collaboratively model fatigue life using experimental data from different manufacturing conditions, effectively addressing the insufficiency of training data under a single condition and the inconsistency of data distributions under multiple conditions. Experimental results of 316L stainless steel processed by laser powder bed fusion are collected for model verification. The results show that, compared to several degradation models, the proposed model has better prediction performance and lower data requirement, exhibiting promising potential in predicting the fatigue life of additive manufactured materials.

Key words: additive manufacturing, fatigue life prediction, data-driven model, transfer learning, clustering