Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

High-fidelity Grain Growth Modeling: Leveraging Deep Learning for Fast Computations

Pungponhavoan Tep, Marc Bernacki

TL;DR

This work addresses the bottleneck of high-fidelity grain-growth simulations by replacing costly PDE-based models with a spatio-temporal neural surrogate. A hybrid Autoencoder–ConvLSTM architecture learns to encode microstructure into a latent representation and forecast future evolution while preserving grain boundary topology through a boundary-preserving loss. Using a TRM-generated dataset, the approach achieves up to 89-fold speedups with high fidelity (SSIM around 0.867 and mean grain size error about 0.07%), and robustly captures boundary topology, morphology, and size distributions. The resulting surrogate enables rapid microstructure predictions for design and manufacturing, offering a practical, scalable tool for materials science workflows.

Abstract

Grain growth simulation is crucial for predicting metallic material microstructure evolution during annealing and resulting final mechanical properties, but traditional partial differential equation-based methods are computationally expensive, creating bottlenecks in materials design and manufacturing. In this work, we introduce a machine learning framework that combines a Convolutional Long Short-Term Memory networks with an Autoencoder to efficiently predict grain growth evolution. Our approach captures both spatial and temporal aspects of grain evolution while encoding high-dimensional grain structure data into a compact latent space for pattern learning, enhanced by a novel composite loss function combining Mean Squared Error, Structural Similarity Index Measurement, and Boundary Preservation to maintain structural integrity of grain boundary topology of the prediction. Results demonstrated that our machine learning approach accelerates grain growth prediction by up to \SI{89}{\times} faster, reducing computation time from \SI{10}{\minute} to approximately \SI{10}{\second} while maintaining high-fidelity predictions. The best model (S-30-30) achieving a structural similarity score of \SI{86.71}{\percent} and mean grain size error of just \SI{0.07}{\percent}. All models accurately captured grain boundary topology, morphology, and size distributions. This approach enables rapid microstructural prediction for applications where conventional simulations are prohibitively time-consuming, potentially accelerating innovation in materials science and manufacturing.

High-fidelity Grain Growth Modeling: Leveraging Deep Learning for Fast Computations

TL;DR

This work addresses the bottleneck of high-fidelity grain-growth simulations by replacing costly PDE-based models with a spatio-temporal neural surrogate. A hybrid Autoencoder–ConvLSTM architecture learns to encode microstructure into a latent representation and forecast future evolution while preserving grain boundary topology through a boundary-preserving loss. Using a TRM-generated dataset, the approach achieves up to 89-fold speedups with high fidelity (SSIM around 0.867 and mean grain size error about 0.07%), and robustly captures boundary topology, morphology, and size distributions. The resulting surrogate enables rapid microstructure predictions for design and manufacturing, offering a practical, scalable tool for materials science workflows.

Abstract

Grain growth simulation is crucial for predicting metallic material microstructure evolution during annealing and resulting final mechanical properties, but traditional partial differential equation-based methods are computationally expensive, creating bottlenecks in materials design and manufacturing. In this work, we introduce a machine learning framework that combines a Convolutional Long Short-Term Memory networks with an Autoencoder to efficiently predict grain growth evolution. Our approach captures both spatial and temporal aspects of grain evolution while encoding high-dimensional grain structure data into a compact latent space for pattern learning, enhanced by a novel composite loss function combining Mean Squared Error, Structural Similarity Index Measurement, and Boundary Preservation to maintain structural integrity of grain boundary topology of the prediction. Results demonstrated that our machine learning approach accelerates grain growth prediction by up to \SI{89}{\times} faster, reducing computation time from \SI{10}{\minute} to approximately \SI{10}{\second} while maintaining high-fidelity predictions. The best model (S-30-30) achieving a structural similarity score of \SI{86.71}{\percent} and mean grain size error of just \SI{0.07}{\percent}. All models accurately captured grain boundary topology, morphology, and size distributions. This approach enables rapid microstructural prediction for applications where conventional simulations are prohibitively time-consuming, potentially accelerating innovation in materials science and manufacturing.
Paper Structure (11 sections, 5 equations, 7 figures, 3 tables)

This paper contains 11 sections, 5 equations, 7 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Dataset
  4. Model Architecture
  5. Evaluation Framework
  6. Loss Function
  7. Training Hyperparameters
  8. Evaluation Metrics
  9. Results
  10. Discussion
  11. Conclusions

Figures (7)

  • Figure 1: Temporal evolution of a $5mm\times5mm$ microstructure. From top to bottom and left to right: $t=0min$ (initial state), $t=15min$, $t=30min$, $t=45min$, and $t=1h$. The color code corresponds to the equivalent circle radius (ECR) in mm. The ECR of a grain is estimated as the radius of the circle having the same area as the considered grain.
  • Figure 2: Neural network architecture for predicting subsequence grain evolution, consisting of an encoder-decoder autoencoder with residual blocks integrated with ConvLSTM layers.
  • Figure 3: Visualization of a prediction case (a) Microstructure state at $t=0h$ generated by LavoGen, serving as input to the conventional TRM simulation and the ML models. (b) Microstructure state at $t=1h$ predicted by the conventional TRM simulation, containing approximately 300 grains evolved from an initial 800 grains, serving as the ground truth reference for evaluating ML model predictions.
  • Figure 4: Heatmap visualization of prediction accuracy across different model configurations. (a) S-10-10 model. (b) S-20-20 model. (c) S-30-30 model. The common color scale indicates the magnitude of prediction error.
  • Figure 5: Grain size distribution analysis comparing Models S-10-10, S-20-20, and S-30-30 against TRM simulation. (a) Normalized frequency grain size distribution comparing model predictions with TRM simulation. (b) Distribution error by size range showing absolute errors in normalized frequency distribution for each model. (c) Surface-weighted ECR distribution showing the percentage of surface area contributed by different grain sizes for each model compared to TRM simulation. (d) Surface-weighted ECR distribution error by size range demonstrating absolute errors when grains are weighted by surface area.
  • ...and 2 more figures