Ofertes treballs fi d'estudis

Títol: Machine Learning Surrogate model for a water-air ejector for vacuum generation

Director/a: Castilla Lopez, Roberto

Email del professor/a: robert.castilla@upc.edu

Departament del professor/a: 729-MF

• 2024 - MÀSTER UNIVERSITARI EN RECERCA EN ENGINYERIA MECÀNICA

Identificador de l'oferta: 205-06981

Modalitat: Universitat

Possibilitat de beca/finançament: No

Idioma d'elaboració del treball: Anglès

Paraules clau:

Descripció:

A water-air ejector is sometimes used in industry for vacuum generation replacing some other devices, such as liquid ring vacuum pumps. Its advantages are robustness, simple design and low maintenance cost. In this ejector a high velocity jet of water (primary flow) creates a low pressure region that is used to entrain and exhaust a secondary flow of air. Computational simulations to predict the performance of an ejector's geometry for a given secondary pressure are complex and high computational resources demanding. This project will generate a dataset through 2D axisymmetric CFD simulations in OpenFOAM, systematically varying key geometric parameters and the secondary pressure while maintaining fixed primary conditions. The resulting model will map the design-operational space to the entrainment ratio (secondary and primary flow rate ratio), enabling both design optimization and off-design performance analysis. This tool will significantly accelerate engineering workflows by replacing numerous CFD simulations with instant predictions.

Objectius:

Primary Objective: To develop a machine learning model that predicts the entrainment ratio as a function of both geometric parameters and secondary pressure. Secondary Objectives: - To establish an automated parametric CFD workflow for 2D axisymmetric ejector simulation in OpenFOAM. - To generate a dataset covering a practical range of geometric parameters and secondary pressures using a Space-Filling Design of Experiments. - To analyze the interaction effects between geometry and secondary pressure on performance. - To implement the trained model in a practical Python-based performance prediction tool.

Tasques a realitzar / Característiques:

1.- Problem definition and setup. Review of available works in literature (75 h) 2.- Dataset generation (200 h) 3.- Machine Learning Modelling Analysis (75 h) 4.- Tool development (15 h) 5.- Thesis writing (10 h)

Tema: ENGINYERIA INDUSTRIAL

Tipus: Estudi

Càrrega de treball:

18 ECTS, equivalents a 450 hores.

Requisits:

Knowledge of: - python - Computational Fluid Dynamics (specifically OpenFOAM) - Fluid dynamics - Thermodynamics (heat and mass transfer)

Temàtica ambiental: No

Confidencial (informatiu): No

Ambit de cooperació: No

Publicació: 13/02/2026

Caducitat: 11/02/2027