Upcoming Event: PhD Dissertation Defense

A graph neural network for solidification microstructure in additive manufacturing

Yigong Qin, Ph.D. Candidate, Oden Institute

Abstract

We propose GrainGNN,  a surrogate model for the evolution of polycrystalline grain structure under rapid solidification conditions in metal additive manufacturing. High fidelity simulations of solidification microstructures are typically performed using multicomponent partial differential equations (PDEs) with moving interfaces. The inherent randomness of the PDE initial conditions (grain seeds) necessitates ensemble simulations to predict microstructure statistics, e.g., grain size, aspect ratio, and crystallographic orientation.   Currently such ensemble simulations are prohibitively expensive and surrogates are necessary.

In GrainGNN, we use a dynamic graph to represent interface motion and topological changes due to grain coarsening. We use a reduced representation of the microstructure using hand-crafted features; we combine pattern finding and altering graph algorithms with two neural networks, a classifier (for topological changes) and a regressor (for interface motion). Both networks have an encoder-decoder architecture; the encoder has a multi-layer transformer long-short-term-memory architecture; the decoder is a single layer perceptron.

We evaluate GrainGNN by comparing it to high-fidelity phase field simulations for in-distribution and out-of-distribution grain configurations for solidification under laser powder bed fusion conditions.  GrainGNN results in 80%-90% pointwise accuracy; and nearly identical distributions of scalar quantities of interest (QoI)  between phase field and GrainGNN simulations compared using Kolmogorov-Smirnov test. GrainGNN's inference speedup (PyTorch on single x86 CPU)  over a high-fidelity phase field simulation (CUDA on a single NVIDIA A100 GPU) is  150-2000X for 100-initial grain problem. Further, using GrainGNN, we model the formation of 11,600 grains in 220 seconds on a single CPU core.

We further extend GrainGNN to predict grain formation in moving melt pools with non-planar geometry. We use a graph to represent the grain structure of the curved melt pool interface. The movement of the grain boundary is modeled by a combination of two components. One is associated with the translation of the melt pool which we model with an axial symmetric shape. The other is associated with grain competition predicted by GrainGNN. We introduce new network features to capture the local geometry information. We also introduce microstructure-to-graph mapping to automatically extract grain features and graph-to-microstructure reconstruction. We train GrainGNN using 500 simulations with cone-shaped melt pools. The accuracy of the network prediction is evaluated by comparing it to phase field simulations with setups different from training, including unseen thermal parameters, longer scan tracks, and different melt pool shapes.

Biography

Yigong Qin is a PhD candidate in the Department of Mechanical Engineering and working in the PADAS research group directed by Dr. George Biros. His research focuses on scientific machine learning and high-performance computing with applications in simulating solidification in additive manufacturing. He received his BS degree in Theoretical and Applied Mechanics from the University of Science and Technology of China in 2019.