Industrial Multi-Fidelity Surrogate-Based Optimization Framework for Compressor Blade Design: Balancing Aeromechanics, Contact Robustness, and NI-POD Modeling
Year: 2026
Type: article
Abstract: This article presents the improvements of a modular, multi-disciplinary, and multi-fidelity surrogate-based optimization (SBO) framework for rotor blade design, aiming at reducing computational cost and accelerating the design process in aero-mechanical applications. The concept of “useful accuracy” is central to our approach, allowing the selection of appropriate fidelity levels for each stage of the optimization chain. The modeling strategy relies on a 3D blade parametrization, which is first introduced. The optimization chain is then presented and each chain block is detailed. In particular, (1) both 2.5D and 3D aerodynamic simulations are considered at two different flight conditions, (2) maximum static stresses and frequency margins with respect to specific engine orders are evaluated, (3) the blade robustness to potential blade-tip/casing contact events is assessed using the clearance consumption simplified criterion as well as nonlinear contact simulations. The proposed numerical strategy is applied to the design of an industrial low-pressure compressor rotor blade. The objective function is to maximize the polytropic efficiency at cruise conditions while satisfying aerodynamic and mechanical constraints. An online SBO strategy is employed to efficiently handle the associated high-dimensional and highly constrained optimization problem. Robustness to blade-tip/casing contact interactions is first investigated using two fidelity levels: simplified criterion and nonlinear simulations. The ability of the simplified criterion to guide optimization process is assessed. Results indicate that, although the locations of the maxima differ, the minimization procedure consistently converges toward similar regions of the parameter space. For aerodynamic modeling, a multi-fidelity surrogate strategy combining 2.5D and 3D simulations is adopted, leveraging low-fidelity 2.5D simulations to enhance the quality of the surrogate models associated with the 3D outputs while reducing the need for expensive 3D evaluations. Advanced surrogate modeling based on non-intrusive proper orthogonal decomposition (NI-POD) are compared with tuned radial basis functions approaches for the prediction of criticity fields. Results show that NI-POD improves prediction accuracy across all derived quantities, despite a higher computational cost. Finally, the multi-fidelity aerodynamic optimization is performed using the simplified contact robustness criterion as constraints. Optimized geometries are compared with initial designs, demonstrating the potential of our framework to efficiently guide aero-mechanical design while balancing fidelity, computational cost, and predictive accuracy.
Source: Journal of Turbomachinery
DOI: 10.1115/1.4072649
Institution: Cenaero (Belgium)
Language: English
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W7204573233