Development of data-driven constitutive models for aerospace materials

Thesis

This study presents novel techniques to develop data-driven constitutive models. The adoption of data-based machine learning-driven models obtained from mechanical loading experiments allows for the accurate and computationally efficient prediction of the mechanical behaviour of materials eliminating the need for theoretical assumptions and potential constraints associated with traditional models.

The research is divided into four phases. In the first phase, uniaxial compression experimental data is used to develop surrogate models for the temperature and strain-rate dependent stress-strain response of a polymeric syntactic foam. In the second phase, the proposed techniques are applied to the history-dependent non-monotonic uniaxial response of commercially pure titanium. The third phase introduces a strategy to formulate data-driven constitutive models from random multiaxial experiments. The obtained surrogate constitutive models are capable of capturing the in-plane stress response of isotropic, elastic-plastic materials loaded by combined normal and shear stresses. The feasibility of this approach is evaluated by conducting virtual experiments by means of Finite Element (FE) simulations in which a hollow, cylindrical, thin-walled test specimen is subjected to random histories of axial displacement and rotation. Finally, in the fourth phase, the methodology developed in the third phase is applied to the real experimental combined normal and shear response of aluminium specimens.

To validate the surrogate models, their predictions are compared against experimental data not used in the training process. The results demonstrate a very good agreement between the measurements and the predictions of the data-driven surrogate models.

In conclusion, this research proposes an innovative approach to data analytics and materials constitutive modelling based on machine learning techniques, offering significant potential to enhance the accuracy and efficiency of predicting the mechanical behaviour of aerospace materials.

Authors: Tasdemir, B

• DOI: 10.5287/ora-m8qagy9gv
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: plasticity; machine learning; data-driven; surrogate model; material modelling; constitutive models
• Subjects: Supervised learning (Machine learning); Deep learning (Machine learning); Plasticity; Solid mechanics; Aerospace engineering; Materials science; Mechanical engineering