Artificial intelligence–driven hybrid metaheuristic optimization of mixed-flow: Balancing efficiency, power consumption, and cavitation performance

Document Type

Article

Publication Date

3-1-2026

Abstract

Multiphase operation in mixed-flow pumps often leads to efficiency loss, increased power demand, and early cavitation, limiting reliability and usable operating ranges. This study develops a hybrid artificial-intelligence framework that combines CFD modelling, a Bayesian-tuned artificial neural network (ANN) surrogate, and multi-objective optimization to improve pump performance. CFD simulations using the SST–kω turbulence model method were carried out to capture gas–liquid interaction and vapour formation within the impeller passages. The ANN surrogate trained on the CFD-generated design dataset reproduced pump responses with high accuracy (R² > 0.99; MAPE < 3 %), enabling rapid performance evaluation during optimization. Four multi-objective algorithms, NSGA-II, GWO, MOALO, and MAOA, were benchmarked to identify designs that balance efficiency, shaft power, and cavitation behaviour. NSGA-II and MAOA provided the most favourable trade-offs, achieving up to a 1.2 % improvement in efficiency and a 0.22 reduction in the cavitation index relative to the baseline configuration. Flow-field analysis showed that these optimized designs weakened vapour-core development and improved pressure recovery. The proposed CFD–ANN–optimization framework offers an efficient and physics-consistent path for mixed-flow pump design enhancement and is adaptable to other turbomachinery operating under multiphase conditions.

Keywords

cavitation, CFD, gas-volume fraction, Mixed flow pump, multi-objective optimization, surrogate modeling, Thoma number

Publication Title

Progress in Engineering Science

ISSN

2950-4252

DOI

10.1016/j.pes.2026.100211

Funders

Petroleum Technology Development Fund, University Grant Research Program (CORG004-2025)

Volume

3

Issue

1

Publisher

Elsevier B.V.

Share

COinS