Download file (979 KB)
Download as zip (903 KB)

Collection

Citation

Alternative title: Prediction of relationship between strength scatter and defect distribution in ceramics using a machine learning model

Description:

Ceramics are widely used as heat-resistant structural materials, such as thermal barrier coatings for gas turbines. However, they exhibit probabilistic fracture behavior due to the size distribution of defects such as pores and grain boundaries present within them. Experimental evaluation of this strength scatter in ceramic components is both time-intensive and costly. Our previous work successfully predicted strength scatter based on fracture mechanics from microstructural information, but this approach incurs enormous computational costs for large-scale applications. In this study, we develop a deep learning–based surrogate model that predicts the Weibull distribution parameters of ceramic bending strength directly from equivalent crack length distributions, enabling substantial reductions in computational cost without sacrificing predictive accuracy. Furthermore, we develop an inverse analysis framework that couples the surrogate model with particle swarm optimization to estimate defect distributions from reference strength data. The proposed method achieves high accuracy and efficiency.

Rights:

• In Copyright
  

Keyword: セラミックス, 破壊統計, サロゲートモデル, 機械学習

Date published: [2025年]

Publisher: 日本ガスタービン学会

Journal:

• 第53回日本ガスタービン学会定期講演会 講演論文集

Funding:

Manuscript type: Author's version (Submitted manuscript)

MDR DOI: https://doi.org/10.48505/nims.6396

First published URL: https://www.gtsj.or.jp/thesis/

Related item:

Other identifier(s):

Contact agent:

Updated at: 2026-07-10 12:00:08 +0900

Published on MDR: 2026-07-10 14:25:01 +0900