EXplainable MACHINe learning: A closed-loop feedback approach for metal hydrides design and discovery

Fencing climate change means looking for alternatives to fossil fuels and enhancing renewables energies (RE). In this direction, renewable H2 can be produced by electrolysis of H2O using RE. As an efficient energy vector, it can be stably stored in the solid state in metal hydrides (MH) reaching high volumetric densities. However, the development and characterization of new materials is time consuming and costly. EX-MACHINA aims to implement a rapid and robust tool for new material design through explainable machine learning (ML), focusing on the validation of a protocol and the modelling of thermodynamic (TD) properties of MH at low and high pressure for stationary H2 storage applications. The project will look at new low-cost alloys with outstanding storage performance and hydrogenation properties. The innovative closed-loop feedback approach will allow gaining fundamental knowledge from advanced statistical, characterization and theoretical methods starting from a large data mining to predict materials and models that satisfy key scientific and technological criteria. The novel combination of ML, Calphad and advanced experiments in a closed loop feedback aims to: (1) Develop a sole, enlarged database on MH by integrating available databases, performing CALPHAD calculations of their thermodynamic properties at low and high pressure (2) Unravel interdependencies on structure-property relationships by identifying feature combinations that are the primary contributors to the ML (3) Use advanced synthesis and characterization methods to produce and integrate new experimental data in the ML analysis, advancing modelling and estimation power (4) Boost Experienced Researcher (ER) carrier, with a 2-way transfer of knowledge and by expanding USA-EU collaborative research in the field (5) Maximize exploitation of research outputs by wide dissemination, communication, and open science.