摘要
Generating experimentally realistic superalloy microstructures from simulated cooling histories remains challenging because simulated thermal histories and experimentally characterized microstructures are linked by an imperfect simulation-to-experiment mapping. Can generative prediction remain reliable when conditioning descriptors predicted from simulated thermal histories deviate from their experimentally measured counterparts? Here, we establish a descriptor-mediated, mode-aware conditional diffusion framework that maps simulated thermal histories to experimentally derived microstructural descriptors and constrains diffusion trajectories using data-driven statistical mode labels. Mode-aware supervision provides the largest improvement among the ablated modules, improving six quantitative metrics—including distribution MAPE, KS statistic, Wasserstein distance, two-point correlation errors (L1 and L2), and microstructure error—by an average of 47.4%. A residual-perturbation strategy further quantifies the propagation of simulation-to-experiment mapping errors. Although degradation follows a super-linear power law (b ≈ 1.39), practical descriptor-level mapping errors cause only 0.724% microstructure-consistency degradation. The proposed framework enables robust generative superalloy microstructure prediction under descriptor-level simulation-to-experiment mapping errors.
| 源语言 | 英语 |
|---|---|
| 期刊论文编号 | 117627 |
| 期刊 | Scripta Materialia |
| 卷 | 288 |
| DOI | |
| 出版状态 | 已出版 - 1 2月 2027 |
| 已对外发布 | 是 |
学术指纹
探究 'Robust generative superalloy microstructure prediction under simulation-to-experiment mapping errors via mode-aware conditional diffusion' 的科研主题。它们共同构成独一无二的学术指纹。引用此
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver