Session: 12-05-02: Data-Enabled Predictive Modeling, Scientific Machine Learning, and Uncertainty Quantification in Computational Mechanics

Paper Number: 99285

99285 - Data-Driven Creep Simulation Based on Gaussian Process Regression for 9% Cr Steel 

Metal creep is one of time-dependent phenomena that can occur in mechanical structures operating at an elevated temperature. Since it has most considerable influence on the material lifespan reduction in such as thermal power plants pressure vessels or pipes, it must be studied appositely in design step to secure long-term structure integrity. By the creep phenomenon, despite the applied stresses in a material is below the yield strength, it will provoke the irreversible damages inside the material over time, and microstructural voids caused by creep damage are coalesced to nucleate the cracks and eventually lead to a creep rupture. The creep behavior can be divided into three stages according to time: primary creep, which shows a gradual decrease in strain rate after a load is applied until it reaches to steady-state; secondary creep, which shows a steady-state constant strain rate; tertiary creep, which shows exponentially increasing strain rate and leading to failure. Therefore, the creep behavior is a nonlinear permanent deformation, and this nonlinearity is the main reason of difficulty in simulating the creep behavior of a material in a Finite Element (FE) analysis. In convention, various parametric models were devised and used to simulate such creep behavior in FE analysis. Though the most common parametric model, Norton creep, has great advantage on its simple formula, it has a limitation that it only confined to predict the secondary creep region. And, although the Kachanov-Rabotnov model can predict secondary and tertiary creep and the Dyson model can predict all three stages of creep behavior, the more the creep behavior region is embodied in the model means the more complicated formula is needed and the number of required parameters is increased. In addition, there is a significant amount of scatter in the creep test results for reasons of intrinsically inherent defects in the material, thus a probabilistic or statistical analysis is recommended for a creep life evaluation, whereas the parametric models have the disadvantage of showing only deterministic prediction results. Therefore, in this study, our purpose is to simulate the metal creep phenomenon by utilizing a data-driven method. As one of useful statistical modelling method, the Gaussian Process Regression (GPR) is used to establish a creep model. Several creep tests were conducted at a temperature of 550 ℃ on modified 9Cr-1Mo steel, which is one of prime candidate material for elevated temperature pressure vessels or pipes, and the obtained test results were used as a training data. For manifesting the creep behavior, GPR was applied by dividing the creep phenomenon as three parts: prediction of creep rupture time according to an applied stress, prediction of creep strain corresponding to a normalized creep time and prediction of creep strain rate corresponding to a normalized creep time. As a result, obtained optimized hyperparameters and covariance matrices were adopted in a creep user subroutine and in the FE analysis tool to simulate the creep behavior. In conclusion, the simulation results show a good accordance with the creep test results. And the effectiveness of data-driven creep simulation was evaluated by comparing to the results of the conventional parametric creep model simulations.

Presenting Author: Uijeong Ro Sungkyunkwan university

Presenting Author Biography: Uijeong Ro is a PhD student in the School of Mechanical Engineering at SKKU. He has been a researcher in Material Strength & Computational Bioengineering Laboratory in SKKU since 2018. His current research is concerned with metal creep and fatigue phenomena. He has worked on high temperature material testing projects with domestic research institution (NRF). His Ph.D. research aims to establish a theoretical model to design creep and fatigue behavior for various high temperature materials and analyze their related lifetimes.

Authors:

Uijeong Ro Sungkyunkwan university
Sangyeop Kim Sungkyunkwan University
Yonghwi Kim Sungkyunkwan University
Taeksang Lee Myongji University
Moon Ki Kim Sungkyunkwan University