DFB-TT

Objective:

Steel is a key structural material for fuel storage and transportation. However, its degradation due to hydrogen intake, also known as Hydrogen Embrittlement (HE), is a complex phenomenon involving several lengths and timescales. Developing efficient interoperable models that capture the influence of composition and thermal treatment on the susceptibility of steels to HE, and that can be deployed as a ‘digital twin’ of the manufacturing process remains a challenging problem undercutting the rational design of steel alloys for hydrogen storage and transport applications. In partnership with Tubacex, this proposal aims at implementing an “Integrated Computational Materials Engineering (ICME)” framework to develop such a digital twin, capable of predicting the crystallography and mechanical resistance of steels given their composition and thermal treatment schedule under specific operating conditions. The ICME framework, originally proposed in 2012, advocates for multiscale modelling with an engineering mindset. Thus, it is frequently implemented by combining density functional theory and/or thermodynamic modelling (e.g. the CALPHAD methodology) for phase diagram estimation, with finite element (FE) analysis for predicting macroscopic behavior. However, instead of releasing the resulting models directly into an industrial setting, they are often employed to adjust simpler subrogate models for real-time calculations. Our ICME implementation will harness the models and methodologies previously developed by BCAM in the context of the ICME-22 and ICME-23 Elkartek projects (lead by Tecnalia). Supplementing the most typical ICME implementations, we will incorporate accurate interatomic force fields and enhanced classical atomistic simulations to predict hydrogen diffusion and its influence on crack formation and propagation at the nanoscale (i.e., at the single crystal level and at the vicinity of defects such as grain boundaries or precipitates). To propagate this information into the macroscale, a Machine-learning (ML) based dynamic-importance sampling methodology will be developed and implemented. In this approach, an ML model will be used to dynamically sample the phase space explored by a macro model using, for instance, FE simulations, enabling automatic feedback from the micro to the macro scale. Our framework can be combined with the traditional CALPHAD scheme to predict, given a particular alloying composition and heat treatment schedule, what the final microstructure, mechanical properties, and susceptibility to HE of the resulting steel will be. Finally, ML subrogate models will be trained to provide the industrial user real-time reliable information of the expected properties of a steel alloy given the thermal treatment schedule. The subrogate models may subsequently be incorporated into a screening scheme for in-silico testing of a wide variety of alloy compositions.