10182 Abstract
|Table of Contents|

Creep Life Prediction of Ni-Based Single Crystal Superalloys by Physical Metallurgy Information Guided Machine Learning(PDF)

MATERIALS CHINA[ISSN:1674-3962/CN:61-1473/TG]

Issue:
2023年第09期
Page:
722-731
Research Field:
Publishing date:

Info

Title:
Creep Life Prediction of Ni-Based Single Crystal Superalloys by Physical Metallurgy Information Guided Machine Learning
Author(s):
FU Jiabo1 WANG Chenchong1 MATEO Carlos Gracia2 CARABALLO Isaac Toda2CABALLERO Francisca Garcia2 YU Hao1
1. The State Key Laboratory of Rolling and Automation, Northeastern University, Shenyang 110819, China 2. Department of Physical Metallurgy, National Centre for Metallurgical Research (CENIM-CSIC),Avda. Gregorio del Amo, Madrid 28040, Spain
Keywords:
Ni-based single crystal superalloy machine learning creep life high temperature and low stress
CLC:

PACS:
TP181;TG132
DOI:
10.7502/j.issn.1674-3962.202212004
DocumentCode:

Abstract:
Creep life is a key material parameter affecting the service life and mechanical properties of Ni-based single crystal superalloys. Therefore, how to predict the creep life of alloys accurately and effectively is critically important for engineering. To address this issue, a physical metallurgy (PM)-guided machine learning (ML) model is developed. Firstly based on literature research, a high quality creep dataset of single crystal at high temperature and low stress is established. Under the guidance of the principles of physical metallurgy, three dimensional physical metallurgy information (volume fraction of γ′ phase Vf,lattice misfit δ, diffusion coefficient DL) are added to the original dataset as extra dimensions to guide the training process. Additionally, the correlation analysis and importance evaluation of the original data features are made based on the Pearson correlation coefficient and the mean accuracy decrease (MDA) value of a random forest model respectively. As a result, the dataset is basically consistent with the physical metallurgy mechanism, and the threedimensional physical metallurgy information is of great significance for the creep life prediction. The creep life of the alloy is predicted on the dataset after data mining based on the machine learning method, and different machine learning models are evaluated according to the squared correlation coefficient (R2), mean absolute error (MAE) and the degree of overfitting. Finally, the optimal model is determined as support vector regression (SVR) model. The relationship between the composition, process, physical metallurgical parameters and creep rupture life of Ni-based single crystal superalloy under high temperature and low stress is successfully established, which can effectively predict the creep life and is expected to serve the reverse design of the alloy.

References

Memo

Memo:
Last Update: 2023-08-28