Data-Driven Design of Disordered Materials

Informations projet

D4M

N° de convention de subvention: 101024077

DOI

10.3030/101024077

Projet clôturé

Date de signature de la CE 1 Avril 2021

Date de début 1 Avril 2021

Date de fin 31 Mars 2023

Financé au titre de

• EXCELLENT SCIENCE - Marie Skłodowska-Curie Actions

Coût total

€ 203 149,44

Contribution de l’UE

€ 203 149,44

Coordonné par

EIDGENOESSISCHE TECHNISCHE HOCHSCHULE ZUERICH
Switzerland

Periodic Reporting for period 1 - D4M (Data-Driven Design of Disordered Materials)

Période du rapport: 2021-04-01 au 2023-03-31

Résumé du contexte et des objectifs généraux du projet

Architected materials, i.e. materials that derive their properties from their structure and geometry, have enjoyed increasing popularity partly due to continuous advances in manufacturing techniques at different scales. They have been used successfully in different contexts including the design of stiff and tough materials for structural applications or aerospace components, or the design of materials that can absorb energy for impact applications. Yet, the design of such materials has so far been been based on regular and periodic patterns, which implies a very limited design space. On the contrary, nature provides plenty of examples of irregular materials with superior mechanical properties (e.g. flaw tolerance) compared to regular systems. The objective of the project D4M ("Deform") is to develop a framework for the data-driven, and hence experience-free, design of architected cellular materials that exploits disorder. In particular, the project focuses on materials which can be described as networks, and leverages modern graph machine learning techniques, and efficient experimentally validated mechanical models to facilitate their design. Through a combination of theoretical, computational and experimental studies, it was shown that a) disorder may be used beneficially to design materials with enhanced mechanical properties such as increased energy absorption during fracture, and b) graph machine learning may be used as an efficient tool to achieve these designs.

Travail effectué depuis le début du projet jusqu’à la fin de la période considérée dans le rapport et principaux résultats atteints jusqu’à présent

Since the beginning of the project, the following technical objectives were accomplished. First, a network representation of random disordered cellular architectures was developed based on statistical principles. Next, a versatile mechanical model was formulated for predicting the mechanical response (focused on fracture) of these cellular materials, which are comprised of slender beams. The model was implemented in an open-source framework, and was then validated by 3d-printing and testing disordered cellular specimens. Following the successful validation, a large dataset of disordered cellular materials and their response to fracture (fracture pattern, fracture energy) was compiled. A graph machine learning model was formulated and trained based on this dataset, which is able to learn the mapping between the disordered network geometry and topology, and its eventual fracture pattern. The machine learning model was combined with a statistical physics-inspired optimization technique to design optimal disordered cellular materials that maximize their fracture energy.

Besides these technical objectives, the project involved several actions towards exploitation and dissemination of the results. One journal publication has been accepted, and another manuscript is currently at its final writing phase. Moreover, two of the three planned conference presentations have been accomplished, and an additional contribution has been accepted for oral presentation in June 2023. The work was further disseminated in one workshop participation, one invited seminar, the generation of a dedicated website, and through meetings and presentations at ETH Zurich.

Progrès au-delà de l’état des connaissances et impact potentiel prévu (y compris l’impact socio-économique et les conséquences sociétales plus larges du projet jusqu’à présent)

Overall, the action has generated a significant impact, as it advanced the field of machine learning-based prediction and design of architected materials, a field which is still in its infancy and has several potential applications e.g. in the design of tough structural and aerospace components, as well as biomedical implants and sports equipment. So far, these designs were carried out predominantly following a trial-and-error approach and the design space was confined to regular periodic patterns. The project showed that it is possible to achieve improved mechanical behavior by extending the design to aperiodic disordered architectures, and in fact, in an experience-free data-driven manner. Moreover, practical computational tools were developed and became available to the public as part of the action.

Beyond the general impact, the action has a positive impact to university students through the supervision of PhD, Master and Bachelor theses, the teaching of a course on Multiscale Modeling by the ER, and the participation in outreach events at ETH. Also it has a positive impact in engaging high school students to pursue science and engineering through virtual lectures in schools Europe-wide. Finally. the results were further disseminated in the general public through participation in the European Researchers Night event 'Intersections'.

Additionally, the action has had a significant impact on the future career prospects of the experienced researcher (ER), since it helped secure a faculty interview at a major European university, according to the Career Development Plan. The various training activities that took place (high performance computing course at ETH Zurich, participation in the MaP Data-Driven Materials and Processes Workshop at ETH Zurich, participation in the Leadership Essential course for Postdocs, as well as practical training in the additive fabrication and experimental lab of the supervisor), helped expand the expertise and competency of the ER in machine learning-based design and fabrication, as well as physical and computational modeling of architected materials. The ER also taught a graduate level course at ETH focused on Multiscale Modeling, gaining important experience as a lecturer, while he also supervised 1 PhD, 3 Master and 2 Bachelor theses, and 1 visiting Master student research project. Finally, the ER was trained in grant writing, by preparing and submitting a Swiss National Science Foundation proposal jointly with the supervisor and two other postdoctoral scholars in the supervisor’s research group. Academic collaborations were formed with two institutions, the University of Cambridge and the Massachusetts Institute of Technology (MIT), Finally, the ER carried out visits to the National Technical University of Athens, Greece (School of Applied Mathematics and Physics, School of Civil Engineering) where he engaged in fruitful discussions with faculty members, and promoted the research carried out as part of the action. All of the above have significantly enhanced the network of the ER in Europe.

Overview of data-driven approach towards enhanced fracture toughness