Characterization of ventilated supercavitation regimes using Bayesian optimized Random Forest models

Ventilated supercavitation, a complex two-phase flow, has mostly been explored through experiments and simulations, with machine learning yet to emerge as a complementary research approach. This study combines experiments, numerical simulations, and machine learning models to explore the characteristics of ventilated supercavitation and the different cavitating regimes behind a disk-shaped cavitator. The experimental data were used to validate the simulation models. Subsequently, an optimized Random Forest model, enhanced using the Bayesian Optimization Algorithm (BOA-RF), was trained on the simulation results to predict cavitation length, cavitation number, and classify the cavitation regime type. The BOA-RF model exhibited high accuracy when compared with both experimental and numerical results. The results indicate that the supercavity length increases with the ventilation coefficient (CQ). However, at the formation ventilation coefficient (CQf), the cavity length undergoes a significant and rapid increase. Moreover, as the ventilation coefficient (CQ) increases, the cavitation regime evolves from foamy cavitation to a continuous transparent and asymmetric cavity, and when CQ ≥ CQf, it transitions to clean supercavity. Additionally, the results indicate that as the Froude number increases, the CQf initially increases and then decreases.