CFD-Based Surrogate Modeling for Cavitation Intensity Prediction in Obstructed Venturi Flows Using CatBoost and Gaussian Process Regression

Mahmoudi, Fallah; Esmaeili, Mostafa; Saadatbakhsh, Mohammad

doi:10.22075/jhmtr.2026.41363.1961

CFD-Based Surrogate Modeling for Cavitation Intensity Prediction in Obstructed Venturi Flows Using CatBoost and Gaussian Process Regression

Articles in Press

Document Type : Full Length Research Article

Authors

Fallah Mahmoudi

Mostafa Esmaeili

Mohammad Saadatbakhsh

Department of Mechanical Engineering, Kharazmi University, Tehran, 15719-14911, Iran

10.22075/jhmtr.2026.41363.1961

Abstract

This study investigated the effect of placing a rectangular obstacle in the divergent section of a venturi on hydrodynamic cavitation intensity. The geometry was simulated and analyzed over a range of pressure ratios from 3 to 6 and divergence angles from 5 to 7 degrees. The research aimed to develop a surrogate modeling framework based on Computational Fluid Dynamics (CFD) data for accurate cavitation intensity prediction. The results of CFD simulations show that in both venturi with and without obstacle, the average vapor volume fraction with increase of divergent angle decreases and with increase of pressure ratio increases. The presence of obstacle while maintaining these trends, the cavitation intensity significantly intensifies. Following data preprocessing, two machine learning models CatBoost and Gaussian Process Regression (GP) were trained and evaluated. Quantitative results demonstrated that both models performed with high accuracy, with CatBoost showing slightly superior performance (test R²: 0.903, RMSE: 0.00896) compared to GP (test R²: 0.886, RMSE: 0.00967). Comprehensive error analyses confirmed the models' robustness and generalizability. SHAP sensitivity analysis identified the divergence angle as the most influential parameter.

Keywords

Hydrodynamic cavitation

Numerical simulation

Rectangular obstacle

CatBoost

Gaussian Process Regression

Subjects