MicroEvoEval: A Systematic Evaluation Framework for Image-Based Microstructure Evolution Prediction

TL;DR

MicroEvoEval introduces a standardized, multi-faceted benchmark for image-based microstructure evolution prediction, addressing the need for reliable long-term evaluation and physical fidelity beyond traditional pixel-wise metrics. It defines four PDE-driven tasks (plane-wave propagation, grain growth, spinodal decomposition, and dendritic solidification), builds short- and long-horizon datasets from high-fidelity simulations, and evaluates 14 models using a metric suite that includes RMSE, SSIM, L-ETAP, L-EAPSR, and inference time. Key findings show that short-term accuracy poorly predicts long-term stability, and modern architectures like VMamba achieve superior long-term performance and efficiency, underscoring the value of advanced architectural priors in physics-guided surrogacy. The benchmark’s insights point to future directions in incorporating physical priors and developing foundational models for microstructure evolution, with significant implications for accelerated materials design.

Abstract

Simulating microstructure evolution (MicroEvo) is vital for materials design but demands high numerical accuracy, efficiency, and physical fidelity. Although recent studies on deep learning (DL) offer a promising alternative to traditional solvers, the field lacks standardized benchmarks. Existing studies are flawed due to a lack of comparing specialized MicroEvo DL models with state-of-the-art spatio-temporal architectures, an overemphasis on numerical accuracy over physical fidelity, and a failure to analyze error propagation over time. To address these gaps, we introduce MicroEvoEval, the first comprehensive benchmark for image-based microstructure evolution prediction. We evaluate 14 models, encompassing both domain-specific and general-purpose architectures, across four representative MicroEvo tasks with datasets specifically structured for both short- and long-term assessment. Our multi-faceted evaluation framework goes beyond numerical accuracy and computational cost, incorporating a curated set of structure-preserving metrics to assess physical fidelity. Our extensive evaluations yield several key insights. Notably, we find that modern architectures (e.g., VMamba), not only achieve superior long-term stability and physical fidelity but also operate with an order-of-magnitude greater computational efficiency. The results highlight the necessity of holistic evaluation and identify these modern architectures as a highly promising direction for developing efficient and reliable surrogate models in data-driven materials science.

Figures (19)

• Figure 1: Schematic of the MicroEvoEval benchmark for microstructure evolution prediction.
• Figure 2: Case study on dendritic solidification.
• Figure 3: Case study on spinodal decomposition.
• Figure 4: The performance-cost of Plane-wave propagation.
• Figure 5: The performance-cost of Grain growth.