Energy engineering and materials science are critical in addressing issues related to climate change because they aim at improving process efficiency in the use of renewable sources. One of the limitations of those sources is their intrinsic intermittency nature, and in this sense, energy storage systems turn out to be key technologies for collecting and storing energy and making it available regardless of the actual source availability.
A study conducted by the research team at Politecnico di Torino composed of Giovanni Trezza, Luca Bergamasco, Matteo Fasano and coordinated by Eliodoro Chiavazzo from the SMaLL laboratory – at the Department of Energy-DENERG – has been recently published in the prestigious scientific journal Nature npj Computational Materials.
The work of the researchers shows how Artificial Intelligence algorithms can be exploited, through techniques known as high-throughput computational screening, to generate a huge amount of possible new innovative materials with unprecedented characteristics and opportunities for energy engineering. Thus, increasingly efficient and fast methodologies are being developed for a targeted “selection” of the most promising materials given a certain application, such as energy storage. Specifically, algorithms have been trained on a broad set of new hypothetical materials, potentially suitable for thermal energy storage. Such materials, called metal-organic frameworks (MOFs), have a crystalline structure – similar to a cage – that can “trap” other molecules, and thus their energy. The research therefore focused on developing a technique to choose the best materials that maximize heat when they trap water.
The study proves to be very important precisely because materials science is increasingly moving toward the automatic analysis of a large body of new and “hypothetical” materials with new or improved properties over those already available. This is made possible by the continuous evolution of computer and technological systems. Indeed, the continuous improvement in the performance of computational systems makes it possible to solve increasingly sophisticated models. Particularly, this can be done for a large number of different materials, up to hundreds of thousands of new “hypothetical” compounds, which would require several years and substantial economic resources to be tested in practice.
“Choosing the optimal material for a certain application is a common need in many engineering fields. In this work, we have considered more than 5,000 hypothetical MOFs calculating, by means of partial literature data and developed ad-hoc developed models, the energy performance they would have in operation within a plant” explains Giovanni Trezza, first author of the research. “Being able to predict in advance the performance of these compounds allows us to make a first and necessary screening on which ones would be interesting to reproduce in the laboratory and then in industrial or civil practice. In addition, through artificial intelligence techniques, we have identified the most energy-relevant chemical characteristics. These can be used for sequential optimization of materials. For example, given a set of MOFs to test in the laboratory, such techniques allow to have an indication of which material to evaluate next without having to test all the possible compounds. This can save substantial time and economic resources; in particular, the algorithms developed have shown a potential 90% acceleration in the discovery of innovative materials for thermal storage.”
“We believe that the better and more efficient energy management processes stem from a multi-disciplinary effort, which combines engineering with more fundamental science in order to develop better technologies and materials than those we have access today,” says Luca Bergamasco, co-author of the research. “Specifically, in our study we combined our knowledge in energy engineering, notions of thermodynamics and materials science, and used the most innovative tools that computer technology makes available today, namely artificial intelligence”.
“The research is going to continue,” the four researchers say, “Indeed, the methodology and tools developed can easily be used to evaluate the performance of materials for other applications of interest in the energy and sustainability sector; for example, capturing carbon dioxide from the atmosphere or recovering water from air moisture in arid areas”.
Figure:Over 5000 hypothetical MOFs from ref. 46 are first featurized by CFID, with the corresponding full set of descriptors provided to a ML regression pipeline for a preliminary descriptor reduction and ML model training of sorption properties of interest. The Tree SHAP interpretation algorithm is thus used to finalize the identification and ranking of a reduced subset of ruling descriptors of the chosen property). Several sequential learning schemes are tested using both the full set of descriptors and the reduced one for a comprehensive comparison.
Source: www.nature.com
