Motivation and goal
DCoMEX (Data-driven Computational Mechanics at EXascale) will fuse physics-constrained machine learning (ML), statistical methods and large scale linear algebra solvers to address extremely demanding computational mechanics problems. It will develop an open source, user friendly and customisable computational mechanics framework that harnesses the capabilities of next generation European Exascale systems.
- Construct an AI-based solver and AI-enhanced linear algebra library
- Develop innovative ML methods for dimensionality reduction and surrogate modelling, including the diffusion maps (DMAP) manifold learning and deep neural networks (DNN).
- Develop an AI-Solve library fusing data-driven methods and surrogate models with efficient block-iterative sparse linear system solvers.
- Exascale deployment of MSolve and Korali software engines
- Optimise MSolve to fully utilise the potential of combined CPU and GPU HPC systems, and validate performance and correctness
- Extend the Korali UQ and Bayesian analysis frameworks with highly efficient load-balanced sampling algorithms
- Pre-process experimental image data
- Develop 3D image and data processing routines that extract geometries and estimates of their uncertainties to use in predictive simulators
- Integrate and evaluate the DCoMEX framework
- Korali + AI-Solve + MSolve with application to immunotherapy and multi-scale material design
- Scalability, energy efficiency, fault-tolerance and error resilience of DCoMEX algorithms on extreme-scale systems
- Provide modular and adaptable software for the broader scientific community and SMEs interested in computational mechanics problems
- Scientific contributions and dissemination
- Application of the DCoMEX framework to optimise patient specific cancer immunotherapy.
- Application of the DCoMEX framework to the multiscale material design
- Dissemination of the novel DCoMEX approach to address large-scale engineering problems