Mechanical behavior predictions of additively manufactured microstructures using functional Gaussian process surrogates
Published in npj Computational Materials, 2021
Relational linkages connecting process, structure, and properties are some of the most sought after goals in additive manufacturing (AM). This is desired especially because the microstructural grain morphologies of AM components can be vastly different than their conventionally manufactured counterparts. Furthermore, data collection at the microscale is costly. Consequently, this work describes and demonstrates a methodology to link microstructure morphology to mechanical properties using functional Gaussian process surrogate models in a directed graphical network capable of achieving near real-time property predictions with single digit error magnitudes when predicting full stress–strain histories of a given microstructure. This methodology is presented and demonstrated using computationally generated microstructures and results from crystal plasticity simulations on those microstructures. The surrogate model uses grain-level microstructural descriptors rather than whole microstructure descriptors so that properties of new, arbitrary microstructures can be predicted. The developed network has the potential to scale to predict mechanical properties of grain structures that would be infeasible to simulate using finite element methods.