Machine learning-enabled predictions of as-built relative density and high-cycle fatigue life of Ti6Al4V alloy additively manufactured by laser powder bed fusion

Abstract

Ensuring high forming quality and performance of as-built components via laser powder bed fusion (LPBF) additive manufacturing (AM) necessitates, meticulous consideration of the material-process-structure-property relationships, particularly due to multiple interrelated AM process parameters exert complex non-linear effects on macro and micro-structures of as-built parts. The quest for optimal process parameters through iterative trial-and-error experiments incurs long time periods and substantial costs, while simultaneously suffering from a lack of precision in discerning the optimized parameters. To address these challenges, this study integrates machine learning (ML) and ML feature engineering techniques for predicting relative densities of LPBF-built Ti6Al4V alloy parts. The prediction performance of Bayesian network (BN) and Multilayer Perceptron (MLP) models are compared, concluding that the MLP prediction model has higher efficiency and accuracy. This prediction model can be inverted to derive optimized LPBF process parameters based on desired density values. The influences of AM defects characterized by micro-computed tomography (μ-CT) on high-cycle fatigue life of LPBF-built Ti6Al4V alloy was studied, for establishing a ML-based pathway for predicting the fatigue life with another MLP model. The efficient approaches offer valuable insights into developing high-precision and widely applicable mapping models that encompass the interplay between process, defects, and fatigue life.