Multi-objective optimization of hydrocyclone by combining mechanistic and data-driven models

This paper presents a cost-effective method to optimize hydrocyclones used for particle separation. It integrates a mechanistic model for data generation with data-driven models for prediction and optimization. The mechanistic model is based on a validated two-fluid model (TFM), and the data-driven models are the artificial neural network (ANN) and non-dominated sorting genetic algorithm II (NSGA-II). In this integration, the response surface methodology (RSM), coupled with the steepest ascent, is used to design the numerical experiments based on the TFM, aiming to achieve reliable prediction through limited numerical experiments or training data. The applicability of the proposed method is demonstrated by multi-variable and multi-objective optimization of hydrocyclone geometry to achieve low pressure drop and accurate separation, especially for fine particles. The optimization result is elucidated using the multiphase flows predicted by the TFM.