Air Entrainment in Drop Shafts: A Novel Approach Based on Machine Learning Algorithms and Hybrid Models

Air entrainment phenomena have a strong influence on the hydraulic operation of a plunging drop shaft. An insufficient air intake from the outside can lead to poor operating conditions, with the onset of negative pressures inside the drop shaft, and the choking or backwater effects of the downstream and upstream flows, respectively. Air entrainment phenomena are very complex; moreover, it is impossible to define simple functional relationships between the airflow and the hydrodynamic and geometric variables on which it depends. However, this problem can be cor-rectly addressed using prediction models based on machine learning (ML) algorithms, which can provide reliable tools to tackle highly nonlinear problems concerning experimental hydrodynamics. Furthermore, hybrid models can be developed by combining different machine learning algorithms. Hybridization may lead to an improvement in prediction accuracy. Two different models were built to predict the overall entrained airflow using data obtained during an extensive experimental campaign. The models were based on different combinations of predictors. For each model, four different hybrid variants were developed, starting from the three individual algorithms: KStar, random forest, and support vector regression. The best predictions were obtained with the model based on the largest number of predictors. Moreover, across all variants, the one based on all three algorithms proved to be the most accurate.