Periodic Reporting for period 1 - DComEX (Data Driven Computational Mechanics at EXascale)
Okres sprawozdawczy: 2021-04-01 do 2022-09-30
DCoMEX is a large-scale European project that aims to provide unprecedented advances to the field of computational mechanics by developing novel numerical methods enhanced by artificial intelligence algorithms. The key innovation of this project is the development of a novel scalable library of AI-enhanced algorithms for the solution of large scale sparse linear systems of equations that lay at the core of computational mechanics. The potential of the DCoMEX computational framework will be demonstrated by detailed simulations in two case studies: (i) patient-specific optimization of cancer immunotherapy treatment, and (ii) design of advanced composite materials and structures at multiple scales. Both these case studies are representative of a family of computational mechanics problems that necessitate peta- and exascale computations.
The DCoMEX is comprised of the following Strategic Objectives (SO):
SO1: Construction of AI-Solve an AI-enhanced linear algebra library
SO2: Exascale deployment of MSolve and Korali software engines.
SO3: Pre-processing of experimental image data
SO4: Integration of the DCoMEX framework, application and performance evaluation.
SO5: Scientific contributions and dissemination.
The DCoMEX is comprised of the following Strategic Objectives (SO):
SO1: Construction of AI-Solve an AI-enhanced linear algebra library
SO2: Exascale deployment of MSolve and Korali software engines.
SO3: Pre-processing of experimental image data
SO4: Integration of the DCoMEX framework, application and performance evaluation.
SO5: Scientific contributions and dissemination.
Sumarry of work performed in the projcet and results achieved with respoect to its objectives:
- We have developed two sets of ML methodologies dimensionality reduction and surrogate modelling, including, namely DMAP and Convolutional Autoencoders (CAEs) (Deliverable D 2.1 and ref [1-3].)
- We are continuously developing and improving AI-Solve library fusing data-driven methods and surrogate models with efficient block-iterative sparse linear system solvers. The proof of concept for this library has been published (ref [4]).
- We are continuously optimising MSolve to fully utilise the combined CPU and GPU potential of modern supercomputers. We demonstrated software capabilities by providing performance and correctness test on physics-based computational models.
- We are extending and upgrading the Korali engine to include state of the art sampling algorithms that harness extreme computation architectures (Deliverable D4.1).
- We have developed 3D image and data processing routines that extract geometries together with estimates of their uncertainties that can be propagated to predictive simulators (Deliverables D5.1 and 5.2).
- The combined Korali and MSolve/AI-Solve machinery has been successfully integrated and tested in Piz Daint HPC platform (Deliverables D6.1)
- We are in the process of producing the baseline frameworks DCOMEX-BIO and DCOMEX-MAT for the applications of WP7
- Continuous monitoring-evaluation of the software in terms of scalability, parallel efficiency, energy efficiency, and data locality is in progress. Energy efficiency considerations are foreseen at a later stage of the project progress.
- We have applied DCoMEX framework multiscale material design applications and presented the results in corresponding Journal Papers and Conferences
- We have presented in DCoMEX platform as novel approach applicable to engineering problems to corresponding exhibitions
- We have developed two sets of ML methodologies dimensionality reduction and surrogate modelling, including, namely DMAP and Convolutional Autoencoders (CAEs) (Deliverable D 2.1 and ref [1-3].)
- We are continuously developing and improving AI-Solve library fusing data-driven methods and surrogate models with efficient block-iterative sparse linear system solvers. The proof of concept for this library has been published (ref [4]).
- We are continuously optimising MSolve to fully utilise the combined CPU and GPU potential of modern supercomputers. We demonstrated software capabilities by providing performance and correctness test on physics-based computational models.
- We are extending and upgrading the Korali engine to include state of the art sampling algorithms that harness extreme computation architectures (Deliverable D4.1).
- We have developed 3D image and data processing routines that extract geometries together with estimates of their uncertainties that can be propagated to predictive simulators (Deliverables D5.1 and 5.2).
- The combined Korali and MSolve/AI-Solve machinery has been successfully integrated and tested in Piz Daint HPC platform (Deliverables D6.1)
- We are in the process of producing the baseline frameworks DCOMEX-BIO and DCOMEX-MAT for the applications of WP7
- Continuous monitoring-evaluation of the software in terms of scalability, parallel efficiency, energy efficiency, and data locality is in progress. Energy efficiency considerations are foreseen at a later stage of the project progress.
- We have applied DCoMEX framework multiscale material design applications and presented the results in corresponding Journal Papers and Conferences
- We have presented in DCoMEX platform as novel approach applicable to engineering problems to corresponding exhibitions
DCoMEX provides advances beyond state of the art to the field of computational mechanics by developing novel numerical methods enhanced by artificial intelligence (AI) algorithms that are deployed with scalable and open source software to exascale computing architectures. A key innovation of our project is the development of a novel scalable library of AI-enhanced algorithms for the solution of large scale sparse linear system of equations that lay at the core of computational mechanics. Our methods fuse physics-constrained machine learning with efficient blockiterative methods and incorporate experimental data at multiple levels of fidelity to quantify model uncertainties.
We continouusly deploy these methods in exascale computing architectures and provide scientists and engineers with efficient HPC-ready tools for predictive simulations of mechanical systems in applications ranging from bioengineering to manufacturing. We are developing DCOMEX platform as a modular and customisable software that can be used by the broader scientific community as well as by SMEs.
DCoMEX will provide a robust and user friendly framework that can be adopted in various applications. We will demonstrate its potential by detailed simulations in two case studies: (i) patient-specific optimisation of cancer immunotherapy treatment, and (ii) design of advanced composite materials and structures at multiple scales. These case studies are representative of a family of computational mechanics problems that necessitate peta- and exascale computations. We will demonstrate the integration of experimental data for quantifying the model predictive capabilities, demonstrate the analysis of extensive datasets as well as the integration of multiphysics systems at multiple spatiotemporal scale.
The foreseen results of the DCOMEX project and the percentage of their completion in the first 18M period are as follows:
Result 1: UQ-aware image pre-processing engine (Compltetion: 100%)
Result 2: CPU+GPU-enabled MSolve multiscale/multiphysics solver (Completion 60%)
Result 3: Adaptive UQ and Bayesian analysis Korali engines (Completion 100%)
Result 4: AI-Solve library (Compltetion 60%)
Result 5: The DCoMEX HPC framework (Compltetion 60-70%). Missing part is the DCOMEX-BIO and DCOMAX–MAT integration
Result 6: DCoMEX-BIO software for cancer immunotherapy optimisation (Compltetion 60%)
Result 7: DCoMEX-MAT software for material design (Compltetion 60%)
We continouusly deploy these methods in exascale computing architectures and provide scientists and engineers with efficient HPC-ready tools for predictive simulations of mechanical systems in applications ranging from bioengineering to manufacturing. We are developing DCOMEX platform as a modular and customisable software that can be used by the broader scientific community as well as by SMEs.
DCoMEX will provide a robust and user friendly framework that can be adopted in various applications. We will demonstrate its potential by detailed simulations in two case studies: (i) patient-specific optimisation of cancer immunotherapy treatment, and (ii) design of advanced composite materials and structures at multiple scales. These case studies are representative of a family of computational mechanics problems that necessitate peta- and exascale computations. We will demonstrate the integration of experimental data for quantifying the model predictive capabilities, demonstrate the analysis of extensive datasets as well as the integration of multiphysics systems at multiple spatiotemporal scale.
The foreseen results of the DCOMEX project and the percentage of their completion in the first 18M period are as follows:
Result 1: UQ-aware image pre-processing engine (Compltetion: 100%)
Result 2: CPU+GPU-enabled MSolve multiscale/multiphysics solver (Completion 60%)
Result 3: Adaptive UQ and Bayesian analysis Korali engines (Completion 100%)
Result 4: AI-Solve library (Compltetion 60%)
Result 5: The DCoMEX HPC framework (Compltetion 60-70%). Missing part is the DCOMEX-BIO and DCOMAX–MAT integration
Result 6: DCoMEX-BIO software for cancer immunotherapy optimisation (Compltetion 60%)
Result 7: DCoMEX-MAT software for material design (Compltetion 60%)