Common Horizon of Open Research in Uncertainty for Simulation – CHORUS

Our goal is to get the robustness, speed and scalability of some reduction modelling techniques when using the reduced models in a UQ study. The specificity of this research is to associate the reduction techniques with the simulation objectives: central tendency analysis, reliability analysis, sensitivity analysis, ...

The reduction techniques which are studied and tested during the first phase of CHORUS are the followings:
- Low rank tensor reduction techniques
- PGD techniques
- Reduced basis techniques
- Multi fidelity kriging
- Bayesien calibration

An other axis is to incorporate HPC capabilities inside the algorithms used classically in reduced modelling techniques. A H-matrix solver has been studied at this stage of progress.

Scientific axis
- 3 workshops organised with extended community (reduction techniques, kriging & gaussian processes techniques, statistical learning)
- Publication of articles and oral communications in national and international conferences

Software axis
- Diffusion of 2 release of Open TURNS integrating advanced features developed during the ANR project OPUS (better Bayesian functionnalities, advanced polynomial chaos expansion)
- Link with a H-matrix solver to solve kriging matrixes
- Deployment of the test-cases on the platforms(Open TURNS et Feel++)
- Release Feel ++

Axe cas-test
- Formalisation of the use-cases in a common language
- Mutualization of R&T needs

- Mise à disposition des cas-tests au consortium

Valorisaton
- Organisation of a training session «HPC & Uncertainty Quantification« in the framework of PRACE

Axe logiciel
Prototyping the low-rank methods with Open TURNS
Follow-up of the tests PGD/Reduced basis
Use of weakly intrusive techniques in the solvers of some partners

Scientfic axis
Goal-oriented sensitivity analysis
Reduction model techniques

Methodological axis
Formalization of UQ problems at system level

6 Articles in international journals
9 oral communications in conferences
5 courses/lectures
1 training organised by PRACE

Target: The CHORUS platform tackles the growing need of quantitative uncertainty management during the risk analysis of complex systems in early design, certification or operating phases. By uncertainty management, we mean the ability to quantify, to propagate and to analyse the effects of uncertainties in a given modelling and simulation workflow. This is an opportunity to ramp-up many existing initiatives on the subject in a structured way and give access to up-to-date methods to the R&T community. Indeed, even if the subject is well identified in the scientific community and already pre organised in some professional circles, there remain crucial challenges to face the current scalability problems: link with HPC capabilities, development of reduced models and treatment of multi disciplinary problems. CHORUS aims at solidifying the scientific and technological bricks in an adequate eco system (SMEs, industrial groups, academic).
Innovations are expected in three directions: 1. Methodological treatment of uncertainty management problems for multi disciplinary purposes, 2. Development of new mathematical models and algorithms to face scalability problems (goal-oriented uncertainty analysis, specific reduction techniques using sparsity or multi fidelity), 3. Accessibility of interoperable advanced algorithms linked to HPC capabilities for a larger community than the CHORUS consortium in a recognised open source environment (OpenTURNS).
Status: The French R&T community has developed a specific approach to tackle this topic first at a methodological level and then at a tool level. This community is recognized at an international level on this topic. CHORUS will benefit from different running projects. In parallel, large companies are under the pressure of justifying and/or diminishing their technical margins but maintaining a zero risk objective. The recent years have demonstrated the existence of quick-wins in many disciplines. CHORUS aims at reinforcing this stream.
Positioning: CHORUS aims at echoing the various initiatives in the scope of uncertainty management and filling remaining gaps for specific topics. As the community already exists with shared scientific and methodological references (around 200 people active in GDR MASCOT, 100 regular users of OpenTURNS), it is very likely to irrigate the different applicative initiatives… This platform will be the cornerstone of scientific, applicative and software activities based on the generic requirements of realistic industrial use-cases: workshops, software publications, new service offer, initial and professional training are planned during the next four years.
Use-case centric: The maturity will be assessed on realistic industrial use-cases through “maturity gates”. Seven use-cases originated in various industrial sectors (aerospace, energy, and telecommunications) and disciplines (bio-electromagnetism, thermics, thermo-mechanics, acoustics, automatic, seismic) will be run following a scheme that allows a complementary approach as well as a large coverage of generic functions and associated tools.
“Open” valorization and economic sustainability: The consortium gathers academic and applicative teams that already have a strong experience in the field of uncertainty management. Most of OPUS contributors are part of CHORUS and guarantee an efficient continuity in methodology. Newcomers reinforce excellence in research aspects, and enlarge the scope from an industry point of view. To ensure the transfer capabilities, two thriving SMEs already positioned on this market will take opportunity of CHORUS to enhance their leadership either by commercializing a software solution (PHIMECA SOFT is based on OpenTURNS) or by providing services around it (installation, training, study…). Thanks to the past activities and various initiatives, the consortium aims at evolving towards a community and an extended “eco-system” dealing with risk management (IRT, Labex will benefit from the results).