SIMOICS : Simulation and Instrumentation for Multi-Objective Optimization by Artificial Intelligence – Cold Spray.
Coordinator: Nicolas JOZEFOWIEZ
LORIA (Université de Lorraine)
Keywords: Artificial Intelligence, bayesian optimization, quality diversity,
multi-objective optimization, instrumentation, multi-physics simulation, cold spray, laser remelting, thermal treatment, characterization.
The synthesis processes for materials are multi-parametric, multi-physical, and multi-scale, and
the corresponding analytical and numerical models are generally incomplete. Optimizing them is
all the more complex as objectives can be multiple and not easily defined by mathematical laws.
For example, how can one simultaneously optimize the material properties and the energy efficiency of the process? The SIMOICS project proposes using artificial intelligence models to address this issue and applying them to ColdSpray processes and post-treatments associated with the CorrIS platforms at MATEIS (A-DREAM project) and the CEA (ARTEMIS project). The goal is to combine tools to converge quickly on a solution by performing the minimum number of simulations and/or experiments on the real system.
To achieve this, INRIA and LORIA will focus on building surrogate models that incorporate data from experiments and simulations, generating multiple high-performance solutions using “Quality-Diversity” evolutionary algorithms, and utilizing multi-objective “black-box” optimization tools.
To provide data for the AI models, the ColdSpray processes and post-treatments will be instrumented and simulated. For better understanding and to optimize the performance of AI models, the material synthesis processes will be considered as subsystems corresponding to the following phases: (1) projection (transport of particles in the carrier gas [distributor and enthalpic source]), (2) deposition (impact and adhesion of particles on the substrate), and (3) post-treatment (laser remelting, densification annealing via non-conventional thermal treatment). Characterizing samples at the end of the line will also be necessary to assess the material properties and performance, which are essential optimization objectives. Experimental data will serve to calibrate the AI models and optimization algorithms.
The SIMOICS project, therefore, rests on three interdependent pillars: instrumentation, numerical simulation, and data collection and processing through artificial intelligence to solve problems with multiple solutions and objectives. SIMOICS aims to develop both online and offline tools that can be deployed on other process platforms and accelerated materials research projects and to expand the high- and low-pressure ColdSpray synthesis platforms to applications beyond corrosion and shape memory alloys (PEPR A-DREAM and ARTEMIS), thanks to AI-assisted optimal parameter search.
The project relies on teams from LORIA, recognized for their expertise in designing tools and methods for optimizing complex systems through Artificial Intelligence; INRIA, for developing optimization tools for robotic systems based on machine learning; and the teams from MATEIS and CEA, which host the high- and low-pressure Cold Spray platforms, respectively. These teams will contribute their resources and expertise in material synthesis and characterization, instrumentation, and numerical simulation of processes.