Abstract
Terminal Settling velocity (Vts) is a significant phenomenon for two-phase solid-liquid flows in various process units, such as slurry pipeline, annular drillpipe, fluidized bed reactor, drier, and gasifier. A two-phase system operated below Vts encounters several technical issues, such as solid accumulation, surface wear, erosion, flow irregularity, increased energy requirement, and backflow or, even, bursting. Measuring and predicting the Vts of spherical and non-spherical particles in rheologically different fluids are of interest to the concerned research community. Although many empirical models with limited boundary conditions were proposed earlier, the application of artificial intelligence (AI) for such modeling is still in infancy. In the current study, we attempt to address the shortcomings of the previous methods of predicting Vts by using a generalized machine learning (ML) approach. The training and testing of the ML models were conducted using an extensive dataset including various particle shapes and fluid rheologies. The current dataset is substantially larger compared to the similar datasets used for previous studies. We investigated several state-of-the-art ML algorithms, including ensemble learning and multi-stage regression models to find out the most optimum model for predicting Vts. The proposed model is applicable to spherical and non-spherical particles in both Newtonian and non-Newtonian fluids. This new application of AI-based modeling in the field of engineering can be used for industrial-scale design and operation with much needed reliability in a cost-effective manner.
[Display omitted]
•Artificial intelligence (AI) was applied to predict terminal settling velocity.•Ensemble learning and multi-stage regression models were tested.•A large dataset comprising 3328 samples was used for training and testing.•The validated model is independent of particle shape and fluid rheology.•The outcome of the current study is a generalized machine learning model.