Deconvoluting thermomechanical effects in X-ray diffraction data using machine learning

Lim, R.E.; Shang, S.-L.; Chuang, C.; Phan, T.Q.; Liu, Z.-K.; Pagan, D.C.

doi:10.1107/S2053273325000403

Acta Crystallographica Section A
Acta Crystallographica
Section A FOUNDATIONS AND ADVANCES

IUCr IT WDC

home archive editors for authors for readers submit subscribe open access

view article

Figure 1
Overview of the simulations used to train Gaussian process regression (GPR) models for X-ray diffraction data analysis. The trained models are subsequently transferred to the target (experimental) domain to separate thermal and mechanical strain effects from experimental X-ray diffraction data. Temperature fields $[T({\bf r},t)]$ and temperature-dependent properties [Young's modulus E(T), Poisson's ratio $[\nu(T)]$ and coefficient of thermal expansion $[\alpha(T)]$ ] are used as input for thermomechanical modeling to predict evolving thermal $[\varepsilon^{\rm T}({\bf r},t)]$ and elastic strains $[{\boldvarepsilon}^{\rm E}({\bf r},t)]$ . These strains are used for input into X-ray diffraction simulations and GPR model training. The trained GPR models are then used for analysis of experimental data.