Calibration of empirical, semi-empirical, and physics-based material models for the prediction of creep and tensile behaviour of Alloy 617

Material models are powerful analytical tools for predicting material behaviour under various loading conditions. These models vary in complexity, progressing from empirical models that rely solely on fitting experimental data, to semi-empirical models that incorporate simplified physical concepts, and finally to physics-based models that explicitly represent the underlying deformation and degradation processes. While increasing model sophistication can provide improved predictive accuracy, broader applicability, and richer mechanistic insight, it also introduces larger and more interdependent parameter spaces, higher computational demands, and greater calibration difficulty. To investigate the practical trade-offs between these modelling frameworks, this thesis examines their application in predicting the creep and tensile behaviour of Alloy 617 across three studies of increasing model fidelity. The first study calibrates twelve empirical models and applies symbolic regression (SR) to discover constitutive expressions for elevated-temperature creep behaviour. In general, the models were able to reasonably predict long-term creep behaviour at multiple temperatures using only short-term experimental creep data, with the SR models additionally capturing mechanism-shifted behaviour. The second study presents the calibration of two semi-empirical elastic-viscoplastic (EVP) models using a multi-objective three-stage calibration workflow to capture elevated-temperature creep and tensile behaviour. While the EVP models reproduced creep behaviour reasonably well, accurately capturing tensile behaviour required additional model complexity, resulting in higher computational cost and calibration effort. The third study calibrates three physics-based crystal plasticity finite element method (CPFEM) models using a multi-objective surrogate-assisted calibration workflow to capture multiscale tensile behaviour. The CPFEM models accurately captured the stress–strain response and texture evolution but showed reduced accuracy in predicting individual grain rotation. The results from all three studies are then synthesised to assess the predictive accuracy, robustness, generalisability, interpretability, and calibration difficulty of the different material modelling frameworks. Overall, the empirical models were the simplest and most efficient but provided limited mechanistic insight. The SR models improved adaptability and automated discovery but required careful validation to ensure physically meaningful results. Additionally, the semi-empirical models enhanced generalisability and interpretability but required greater calibration effort. Finally, the physics-based models provided the highest fidelity and mechanistic understanding but required substantial computational resources and detailed microstructural data. Together, these insights provide a principled basis for selecting suitable material models that best align with engineering priorities and available resources.

Grantor dc:publisher

UNSW, Sydney

Year dc:date

2026

Author dc:creator

• Choi, Janzen

• material model
• calibration
• parameter optimisation
• crystal plasticity finite element method
• CPFEM
• finite element method
• elastic-viscoplastic
• EVP
• anzsrc-for: 4016 Materials engineering
• anzsrc-for: 4602 Artificial intelligence
• anzsrc-for: 4017 Mechanical engineering

Statement dc:rights

• open access
• CC BY 4.0
• free_to_read

Licence

• https://purl.org/coar/access_right/c_abf2
• https://creativecommons.org/licenses/by/4.0/

Language dc:language

en

Identifier

https://doi.org/10.26190/unsworks/32130

OAI identifier oai:identifier

oai:unsworks.library.unsw.edu.au:1959.4/107249