Modeling (based on a precise understanding of physical mechanisms) and numerical simulation are bound to play a strategic role in the design and evaluation of systems. Increasingly often, simulation indeed replaces actual testing on real systems or plays a key role in the optimisation of qualification plans, based on "smart" experiments
Modeling and simulation are relevant in almost all fields of technological fields and this thematic will be part of different Excellence laboratories, inducing opportunities of interactions. Digiteo will emphasize on the software aspects with the objective of addressing, in an integrated approach modeling, algorithmic, programming, interfacing and architectures issues.
More and more, simulation is part of complex processes that require iterative and intense campaigns of computations (inverse problems solutions, simulation coupled to statistics).
Thus, applications in industrial context require software always faster, robust, including automatic pre and post processing. Due to the facts that modeling addresses multi-scale (propagation distances/wavelengths) and multiphysics phenomena (e.g. electro-mechanical behaviours of sensors, wave-matter interaction) and that simulation has to cover a great variety of materials, geometries, techniques, this is a real challenge which imply the development of innovative computation strategies.
Hybrid codes involving different numerical methods, in particular finite elements and semianalytical formulations, are identified as very promising solutions. The objective is to get the best advantages of the different methods by an optimized assignation of computation tasks.
Such coupling (formulation and implementation) generates specific developments to the various addressed applications. Particularly challenging is the search for generic aspects in methods initially developed for specific applications.
This approach does not overcome all the limitations encountered for 3D realistic
configurations in terms of computation time or data storage. The exploitation of parallel computations on multicore and SIMD architectures and the increasing computing power they offer appears as a tremendous opportunity of significant breakthroughs. Such advances require the development of numerical schemes and algorithms dedicated to the targeted architectures.
Numerical simulation applies to virtually all industries where economic competitiveness is closely linked to the duration of the design/development cycles, to risk management and to the reduction of environmental impact. It can drastically reduce design and maintenance costs, time to market and the impact of regulatory changes.
Numerical simulation programs become more complex as they must handle multiple physical phenomena (to take into account the various components of the systems) and multiscale aspects (to manage the components hierarchically). They benefit from ever-increasing computing power, call upon intensive high performance or grid computing and often have to handle very large volumes of data. The schedules of development of software components and hardware components must be consistent. The codes must also meet internationally recognised standards of verification, validation and qualification.
A non-exhaustive list of topics of interest to Digiteo is :
Multiphysics and multiscale numerical modeling in time and space (neutronics,
thermodynamics, hydraulics, mechanics, chemistry, elastodynamics, waves, electromagnetism, non destructive evaluation, metallurgy...) : one of the challenges is to interface software components dealing with different physics while maintaining their best performance with current architecture (cache memory, multicores, SIMD) and without creating defects in the convergence of methods or in the continuity of results.
- Modeling of non-destructive evaluation techniques : Electromagnetic, X-ray techniques, ultrasound, guided waves etc...
- Semi-analytical and hybrid approaches for wave propagation in materials and their interaction (scattering) with heterogeneities and defects.
- Model-based reconstruction and imaging algorithms.
- Solution of inverse problems (characterisation of materials properties, localisation and sizing of effects, etc...).
- Probabilistic approaches (noise modeling, probability of detection...)
Adaptation of software to hardware :
- Current CPU, GPU and memory architecture
- Use of clusters : load balancing, scheduling, repartition, heterogeneous hardware, automatic code generation and adaptation
- Dedicated algorithms for computation on grids/High performance computing (access to TER@TEC, CNRS / Blue gene)
- A priori, a posteriori data compression
- Large data set visualisation
- Real time model-based reconstruction and data processing
Code supervision :
- Pre- and post-treatment adapted to models with very large meshes (millions of degrees of freedom) (see Visualisation)
- Programming paradigm (automatic parallelisation, deep memory hierarchies...) for strategies to couple codes and to achieve multi-scale computations
- Learning : construction of simplified models, metamodels, response surface method for non-linear models, genetic algorithms,
- Treatment of uncertainties and errors (modeling and propagation, cf. Uncertainty),
- Probabilistic methods (cf. 2.1).