PhD Position Surrogate Models and Multi-Objective Optimization for the Cold Spray Process

Objectives The candidate will develop an integrated approach built on two main methodological pillars. 1. Building intelligent surrogate models • Development of machine-learning-based surrogate models integrating experimental data and, where applicable, data from numerical simulations. • Exploration of various modeling approaches: Gaussian Processes (including heteroscedastic variants to capture uncertainty from different data sources), neural networks, or hybrid methods. • Multi-source data processing and analysis: – Exploratory analysis and preprocessing of heterogeneous data (in-situ instrumentation, post-deposition characterization, and possibly simulations). – Extraction and selection of physically guided, relevant features to improve model performance and interpretability. – Management of dimensionality and sparsity of experimental data. – Definition of relevant and measurable quality criteria for coating evaluation (composition, microstructure, porosity, deposition efficiency). 2. Constrained multi-objective optimization • Development and adaptation of multi-objective optimization algorithms suited to expensive black-box functions. • Exploration of multi-objective Bayesian optimization approaches leveraging the developed surrogate models. • Investigation of metaheuristics suited to this context (evolutionary algorithms, swarm optimization, etc.). • Management of trade-offs between conflicting objectives (e.g., mechanical properties vs. process efficiency vs. energy cost). • Sequential sampling strategies for adaptive updating of surrogate models. Collaboration This work is part of a larger project in which numerical simulation aspects of the process will be developed by other collaborators. The candidate will interact with these teams, in particular to: • integrate simulation results as complementary data sources for the surrogate models; • provide recommendations of optimal configurations to be tested experimentally or through simulation; • validate the predictions of the surrogate models. Expected impact The tools developed will: • accelerate the discovery of optimal process parameters for various application objectives; • significantly reduce the number of required physical experiments; • improve the understanding of relationships between input parameters and coating properties; • provide a generic methodology transferable to other additive manufacturing processes. Candidate profile Master's degree (or equivalent) in Computer Science or Applied Mathematics, with skills in optimization and/or machine learning. An interest in industrial applications and interdisciplinary work will be appreciated. Application Please send by email or through the ADUM platform (https://www.adum.fr/): • a detailed CV; • academic transcripts; • the master's thesis report (if available). Contact: bernardetta.addis@loria.fr, nicolas.jozefowiez@univ-lorraine.fr, tanguy.lacondemine@cea.fr