The aim of this postdoctoral project is to develop an effective supervised machine-learning approach for accelerating the prediction of the evolution of mechanical (elastic and viscoplastic) anisotropy associated with crystal preferred orientations in geodynamic models.
Such a tool is needed, because most microstructural evolutions in response to deformation produce anisotropic mechanical behaviors. This anisotropy changes the deformation patterns, because in an anisotropic medium, the principal strain directions might differ in orientation from the principal stress directions. It may also produce strain localization, since some domains or layers of the material will have lower (or higher) strengths than the average for a given solicitation. However, its calculation is computationally intensive and memory-costly.
Two main processes produce mechanical anisotropy during ductile deformation: (1) development of crystal preferred orientations (CPO) and (2) formation of a layering characterized by the alternation of weak and strong layers. We will initially focus on the first process. In a second step, we will treat the mechanical anisotropy associated with the development of rheological layer in a multiphase material. Our work in the last 10 years substantiates that CPO-induced viscoplastic anisotropy in the upper mantle changes the deformation patterns, enhancing the shear components, and has the potential to produce strain localization in continental plates [1,2]. Other groups have shown that CPO-induced viscoplastic anisotropy may change the convection patterns in subduction zones, delay the development of Rayleigh-Taylor instabilities at the base of the lithosphere, and resist changes in plate motions [3,4,5]. However, the available approaches to model the evolution of this anisotropy either consider only part of the full anisotropy tensor or are very memory- and time-consuming.
The aim of the postdoctoral project is therefore to develop an effective supervised machine-learning approach predicting the evolution of strain-controlled of viscoplastic anisotropy in the upper mantle. This work will be based on our group expertise in modeling CPO evolution and the associated elastic and viscoplastic anisotropy in polycrystalline materials [1,2,6-8] to build the training database.
In practice, the postdoctoral fellow will: ( 1) Acquire a solid understanding of the evolution of mechanical anisotropy during the viscoplastic deformation of the Earth’s mantle rocks and of the existing methods for its simulation (senior researchers in the RhEoVOLUTION team are experts on the topic); (2) Establish the best strategy for creating synthetic databases using the codes available within the RhEoVOLUTION team to simulate the evolution of crystallographic orientations and elastic and viscoplastic anisotropies of polycrystalline materials; (3) Define the machine learning (ML) approaches best suited to the problem (preliminary work by the RhEoVOLUTION team using LSTM networks to simulate the evolution of elastic anisotropy produced promising results); (4)Train and test the ML algorithms and perform error analysis; and (5) Implement the most successful algorithms as surrogates within geodynamic simulation codes.
Extension of this approach to the study of ice flows (in collaboration with researchers from the ERC RhEoVOLUTION team in Grenoble) is envisaged. Applications in Material Sciences, in particular metallurgy, are also possible.
We are looking for highly motivated candidates with strong numerical and methodological skills. A PhD degree in Geophysics, Material Sciences, Mechanics, Applied Mathematics, Computer Sciences or a closely related field is required at the time of appointment. Required skills: Proven expertise in numerical modelling and excellent programming skills (ideally Python and Fortran). Background in Crystallography, Deformation of crystalline materials, Solid/Fluid Mechanics. Ability and desire to work in a closely cooperating team but also independently. Proficiency in English and demonstrable communication skills. Experience with Deep Learning algorithms for regression tasks, ideally using time-series data and data analysis is not essential, but will be highly valued.
Starting date: The position is open from January 1st, 2023
___________________________________________
[1] Tommasi, A., et al., 2009, Structural reactivation in plate tectonics controlled by olivine crystal anisotropy. Nature Geoscience, 2: 423-427
[2] Mameri, L., Tommasi, A.; Signorelli, J., Hassani, R. 2020. Modelling olivine-induced viscous anisotropy in fossil mantle strike-slip shear zones and the resulting strain localization in the crust. Geophys. J. Intern., doi: 10.1093/gji/ggaa400
[3] Lev, E. and B.H. Hager, 2011, Anisotropic viscosity changes subduction zone thermal structure. Geochem. Geophys. Geosyst., 12, doi: 10.1029/2010GC003382
[4] Lev, E. and B.H. Hager, 2008, Rayleigh-Taylor instabilities with anisotropic lithospheric viscosity. Geophys. J. Int., 173: 806-814.
[5] Kiraly, A., Conrad, C., Hansen, L., 2020. Evolving viscous anisotropy in the upper mantle and its geodynamic implications. Geochem. Geophys. Geosyst., doi: 10.1029/2020GC009159
[6] Tommasi, A., et al., 2000, Viscoplastic self-consistent and equilibrium-based modeling of olivine lattice preferred orientations. Implications for upper mantle seismic anisotropy. J. Geophys. Res., 105: 7893-7908
[7] Mameri, L., et al., 2019, Predicting viscoplastic anisotropy in the upper mantle: a comparison between experiments and polycrystal plasticity models. Phys. Earth Planet. Int., 286: 69-80.
[8] Signorelli, J., Hassani, R., Tommasi, A., Mameri, L. 2021. An effective parameterization of texture-induced viscous anisotropy in orthotropic materials with application for modeling geodynamical flows. Journal of Theoretical, Computational and Applied Mechanics, https://arxiv.org/abs/2008.11494