Project

Back to overview

Open-Source Platform for Efficient FFT-based Continuum Mesoscale Modelling

English title Open-Source Platform for Efficient FFT-based Continuum Mesoscale Modelling
Applicant Junge Till
Number 174105
Funding scheme Ambizione
Research institution Laboratoire de modélisation mécanique multi-échelle EPFL - STI - IGM - LAMMM
Institution of higher education EPF Lausanne - EPFL
Main discipline Other disciplines of Engineering Sciences
Start/End 01.10.2017 - 30.09.2021
Approved amount 783'958.00
Show all

All Disciplines (4)

Discipline
Other disciplines of Engineering Sciences
Information Technology
Mechanical Engineering
Civil Engineering

Keywords (12)

Open science; Scientific computing; Multiscale modelling; Mesoscale modelling; Fast fourier transform; Reproducible science; Open data; Computational homogenisation; Computational solid mechanics; Scientific software development; Mesoscale-aware modelling; Computational materials science

Lay Summary (French)

Lead
En mécanique, l'homogénéisation numérique est un outil pour modéliser des cellules périodiques de matériaux hétérogènes. De telles cellules homogénéisées peuvent ensuite être utilisé dans des analyses dites multi-échelles, pour prédire le comportement de structures en matériaux composites, tel que par exemple des structures en béton, céramique, bois, etc. Les coûts importants des calculs multi-échelles sont l'obstacle principal à leur application répandue.
Lay summary
Objectifs visés par le projet
Le but principal de ce projet est de développer µSpectre, une plateforme open-source pour la homogénéisation numérique utilisant une nouvelle méthode extrêmement efficace basée sur la transformée de Fourier numérique (FFT).

Nous avons choisi le modèle de développement open-source pour encourager des collaborations et pour promouvoir la science réellement reproductible. Initialement, l'application principale de µSpectre sera la prédiction de l'endommagement de structures en béton par la réaction alcali-granulat (ASR) qui affecte environ 20% des barrages hydroélectriques suisses.

Contexte scientifique et sociétal du projet de recherche
Le projet mettra à disposition de la communauté de mécanique numérique des milieux continus un outil polyvalent pour l'analyse multi-échelles de structures en matériaux composites et contribuera à la compréhension de l'évolution de l'endommagement de structures en béton dû à l'ASR.
Direct link to Lay Summary Last update: 18.09.2017

Responsible applicant and co-applicants

Employees

Collaboration

Group / person Country
Types of collaboration
Computational Solid Mechanics Laboratory (LSMS) Switzerland (Europe)
- in-depth/constructive exchanges on approaches, methods or results
Scientific IT and Application Support (SCITAS) Switzerland (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Research Infrastructure
- Exchange of personnel
Laboratory for Multiscale Mechanics Modeling (LAMMM) Switzerland (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
- Research Infrastructure
Experimental Centre, Civil Engineering, Czech Technical University in Prague Czech Republic (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
- Research Infrastructure
- Exchange of personnel
Group of Prof. Jan Zeman, Czech Technical University in Prague Czech Republic (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
- Research Infrastructure
- Exchange of personnel
Laboratory for Construction Materials (LMC), EPFL Switzerland (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Publication
Simulation Group Prof Lars Pastewka, Imtek, Universität Freiburg Germany (Europe)
- in-depth/constructive exchanges on approaches, methods or results
- Research Infrastructure
- Exchange of personnel

Scientific events

Active participation

Title Type of contribution Title of article or contribution Date Place Persons involved
WCCM (13th World Congress in Computational Mechanics) Talk given at a conference Fast Spectral Solvers Without Linear Reference Medium 22.07.2018 New York, NY, United States of America Junge Till;
ECCM (6th European Conference on Computational Mechanics) Talk given at a conference Fast spectral Solvers Without Linear Reference Medium 11.06.2018 Glasgow UK, Great Britain and Northern Ireland Junge Till;
MEGA.seminar EPFL (MEchanics GAthering Seminar Series) Individual talk FFT-based Homogenisation without Linear Reference Media 31.05.2018 EPFL, Switzerland Junge Till;
Tech meetig at Ansys Zürich Individual talk FFT-based Homogenisation without Linear Reference Media 03.05.2018 Ansys, Zürich, Switzerland Junge Till;
Invited seminar in the Mechanics Department of ETHZ (MAVT) and the group of Prof Dennis Kochmann Individual talk Unconditional Convergence for FFT-based Homogenisation 30.01.2018 ETHZ, Switzerland Junge Till;
Seminar in the Simulation group of Prof Lars Pastewka at the IMTEK in the University of Freiburg Individual talk FFT-Based Homogenisation without Linear Reference Media 19.01.2018 University of Freiburg, Germany Junge Till;


Self-organised

Title Date Place
1 week research visit from Prof Jan Zeman 04.09.2018 EPFL, Switzerland
Research visit at the University of Freiburg 19.02.2018 University of Freiburg, Germany

Use-inspired outputs

Software

Name Year
µSpectre 2017


Abstract

Computational continuum mesoscale modelling (or computational homogenisation) involves computing the overall response of a periodic unit cell of material, a so-called representative volume element (RVE), to a given average (i.e., macroscale) strain. Typically, this is done using the finite element method (FEM), even though it is neither able to leverage its main strength, the trivial handling of complex geometries, nor otherwise particularly well suited for periodic problems. An alternative method for modelling periodic RVEs is based on the fast Fourier transform (FFT), developed by Moulinec and Suquet. This method has evolved substantially over the last two decades, with particularly important and currently underused improvements in the last two years.This new method for the solution of the core problem of computational homogenisation is significantly superior to the FEM in terms of computational cost and memory footprint for most applications, but has not been exploited to its full potential. One major obstacle to the wide adoption of the method is the lack of a robust, validated, open-source code. Hence, researchers choose the well-known and tested FEM that has numerous commercial, open-source or legacy in-house FEM codes.The goal of this project is to develop µSpectre, an open-source platform for efficient FFT-based continuum mesoscale modelling, which will overcome the major obstacle. The project is designed to i) establish a de facto standard implementation for the spectral RVE method (SRM) that subsequent implementations can be compared to, in order to concentrate the development effort of all interested parties in the field, in a similar way as, for instance, LAMMPS does for molecular dynamics modelling, ii) demonstrate the performance of the SRM by applying it to the simulation of alkali-silica reaction (ASR) damage evolution in concrete in a fully coupled, macro/meso multiscale simulation complementing and accelerating previous prohibitively expensive FEM-based attempts, iii) make µSpectre widely accessible for users by providing language bindings for virtually all relevant popular computing platforms and comprehensive user’s manuals in order to help widespread adoption, and, finally iv) make µSpectre eminently modifiable for developers by developing it in the open, with a clean architecture and extensive developer’s documentation in order to maximise outside contributions.Furthermore, this project places great importance on truly reproducible and verifiable science with a credible open data strategy in the firm belief that these qualifiers help to reach and guarantee a high level of scientific quality, difficult to reach otherwise, and to attract outside collaborations and contributions that help boost the scientific output beyond what can be achieved by a two person team (principal investigator (PI) and PhD student). To reach these objectives, the PI i) outlines in detail an open Bazaar-style software development plan designed to maximise external visibility and usability in order to attract valuable collaborations that help to reach the ambitious goals of µSpectre, and ii) intends to pioneer the use of containerised HPC which allows guaranteed reproducibility of simulations, thus making obsolete the storage of simulation results, enabling the application of credible open data policies when using µSpectre without exorbitant storage costs. In addition, ASR is a leading deterioration process of concrete structures and is as such an academic and private sector research priority, particularly in Switzerland. Hence, the application of µSpectre to ASR is likely to be widely used and significantly advance the field.
-