Coupled Stochastic / AI / HPC approach to coating formation by particle impact/stacking in the plasma suspension spraying process – ACSIA

The Suspension Plasma Spraying (SPS) process uses a plasma jet to accelerate and melt submicron particles dispersed in a liquid to create coatings with adjustable microstructures and superior properties compared to coatings made by conventional plasma spraying. The SPS process is of great interest to industry due to the variety of its applications: next-generation thermal barriers for gas turbines in the aerospace and industrial sectors, on both new and used parts, high-performance wear-resistant coatings, antimicrobial coatings, as well as fuel cell components or solar power plant components, etc. However, its complexity hinders its industrial use due to the number of operational parameters and the difficulty of coating parts with complex shapes. Mastering this process requires understanding the relationships between the many operational parameters, the suspension treatment mechanisms in the plasma jet, and the final coating microstructure. However, in-situ diagnostic tools are limited due to the small size of the particles. The development of a multi-physical digital twin of the SPS process is therefore essential and will ultimately help reduce the development time for aerospace coatings by optimizing material and energy costs. Previous work conducted by the joint laboratory Prothéïs (IRCER (Limoges) / Safran / Oerlikon) and the I2M Institute (Bordeaux) has proposed four numerical models covering the formation of plasma in the torch, the flow of the plasma jet at the torch outlet with the development of turbulence, the suspension treatment, and the formation of the deposit. The formation of the deposit involves simulating, at the liquid/gas interface scale, the impact of a very large number of particles on a substrate, where they solidify upon contact. Despite the computational power now available in large computing centers, we remain in a domain where computational approaches through solving equations of continuum mechanics and heat transfer will not be sufficient to overcome all the obstacles to creating a digital twin of the SPS process. Therefore, it is necessary to couple direct numerical simulation with less computationally intensive approaches. Specifically, we propose, as part of this project, to develop a stochastic approach to the flow of particles impacting the substrate, coupled with an Artificial Intelligence (AI) algorithm to predict the particle contours after impact without significant loss of accuracy. The tool thus created aims to replace the CFD (Computational Fluid Dynamics) engine in the Notus fluid mechanics code with an AI that predicts the particle contours after impact and solidification. This ACSIA project, led by I2M in partnership with IRCER and Safran, opens up an exciting prospect for using AI as an acceleration tool for solving complex multi-physical problems by encapsulating results from CFD-type calculations and will allow a change in scale in the number of particles processed. Coupling with other software components for simulating the SPS process will ultimately lead to the creation of its digital twin, enabling predictive simulations of the coating structure. This will make it possible to link the operational parameters of the process with the resulting microstructure, thus optimizing existing coatings or developing new ones with cost and development time savings.