Session: 36-02 Adjoint-based approaches - Part 2

Submission Number: 179256

Comparative Assessment of Gradient Based and Evolutionary Optimizers for Preliminary Impeller Design 

Preliminary design of centrifugal compressors is increasingly driven by algorithmic optimization, since early geometric decisions largely determine efficiency and operability within a multidimensional, constrained, nonconvex design space.

Two optimizers embedded in ARGO®, a quasi-2D framework for optimized preliminary sizing of centrifugal impellers, are evaluated: a generalized reduced gradient (GRG) solver and an evolutionary algorithm. The architecture uses a nested-loop scheme, in which an inner optimizer minimizes internal and parasitic losses for the candidate geometry, subject to admissibility limits.

The GRG optimizer, with active constraint handling, converges rapidly on smooth regions but remains sensitive to initialization and can stall at local optima. By contrast, the evolutionary optimizer is derivative free, and population based, employing selection, recombination, and mutation, exploring multiple local solutions, although at higher computational cost. Within this workflow, ARGO® returns an impeller geometry parameterization suitable for preliminary CAD and map generation, and its predictions are verified against ANSYS® CFX simulations.

Using identical loss closures and property models, a benchmark campaign on representative cases assesses the ability of both optimizers to approach global rather than local optima, sensitivity to initialization, constraint satisfaction, computational time, and robustness across operating conditions and working fluids. Results provide a comparison of optimizer performance for preliminary centrifugal compressor design.

Presenting Author: Gianluca Cevolani Università degli Studi Roma Tre

Presenting Author Biography: Gianluca Cevolani holds a Master’s degree in Mechanical Engineering from Roma Tre University. He previously served as a research fellow on the EU CO2OLHEAT project, focusing on supercritical CO₂ (sCO₂) power cycles for waste-heat recovery in energy-intensive industries. He is currently a second-year PhD candidate researching sCO₂ power systems, with interests spanning turbomachinery, cycle modeling, and optimization.

Authors:

Gianluca Cevolani Università degli Studi Roma Tre
Ambra Giovannelli Università degli Studi Roma Tre
Giuseppe Messina ENEA-Casaccia
Jacopo Romani Università degli Studi Roma Tre
Coriolano Salvini Università degli Studi Roma Tre