Understanding the Effect of Build Direction and Scanning Strategy on the Tensile Response of Additively Manufactured in 625 with Innovative Calibration Strategy

Abstract

Spatial variation of microstructure is an integral feature of almost all forms of metal additive manufacturing processes. This directional dependence makes part qualification extremely difficult. Using computational models with complex constitutive laws to bypass trial and error requires accurately identifying several model parameters with limited data. This article presents a systematic approach to using limited characterization data to identify crystal plasticity (CP) material law parameters and apply it to predict orientation and scan strategy dependence of the mechanical properties of laser powder bed fusion IN 625 tensile coupons. This work applies two methods, a higher-order proper generalized decomposition (HOPGD) and a novel interpolating neural network (INN), as surrogates of the full field model, and discusses their comparative performances. Furthermore, the article goes through the details of the adaptive sampling strategy to efficiently use offline databases and how to correctly approximate the microstructure representation from characterization data. The work demonstrates that the INN differentiable surrogate model requires a small set of offline data to effectively calibrate multiple CP parameters without resorting to expensive genetic algorithms for calibration. A computational model calibrated using INN can predict mechanical responses closer to the experiments compared to HOPGD. The authors have applied HOPGD to build a model to win the National Institute of Standards and Testing (NIST) Additive Manufacturing Benchmark Challenge 2022, and the same set of experiments are used to perform the comparison.