A Data-Driven Approach for Predicting the Onset of Entrainment in Two-Fluid Stratified Systems During Selective Withdrawal Process Using Machine Learning Techniques

Abstract

Abstract Selective withdrawal is crucial in the US Strategic Petroleum Reserve and other applications, such as micro-particle coating, thin glass fiber fabrication, and local drug delivery. Physics-based mechanistic models are commonly used to predict the onset of entrainment but have challenges and uncertainties in its computational solutions. Machine learning (ML) models, like Gaussian Process Regression (GPR) and Artificial Neural Networks (ANN), can provide more efficient solutions to these problems. This study uses experimental data from various sources to investigate GPR and ANN for predicting the onset of entrainment in two-fluid stratified systems. Non-dimensional parameters were used as model inputs to ensure generalizability, and dimensionality reduction techniques, like the Genetic Algorithm, were employed to optimize the models. The machine learning models’ predictions were compared with physics-based models across different flow regimes. Preliminary findings show that GPR and ANN models have excellent predictive capabilities, surpassing physics-based models. GPR demonstrated slightly better performance than ANN. However, predictions are only accurate if the underlying physical regime remains unchanged. Machine learning models can unify physics-based models based on available training data at different flow regimes. This study emphasizes the potential of machine learning techniques in solving complex problems and offers a novel approach to predicting the onset of entrainment in select two-fluid stratified systems.