Improving the precision of solids velocity measurement in gas-solid fluidized beds with a hybrid machine learning model

Abstract

Accurate measurement of solids velocity is crucial for fluidized bed reactor design, optimization, and scaling in chemical engineering. However, the cross-correlation calculation of dual-channel particle concentration signals has limitations in dense-phase fluidized systems. To address this, we propose a data-driven approach using machine learning algorithms in numerical simulations. Four popular algorithms (linear regression, random forest, support vector machine, and neural network) are evaluated. Individual models show poor generalization with a determinable coefficient below 0.2. To overcome this, a hybrid model combining a neural network and random forest is developed, improving data prediction and generalization. The determinable coefficient increases to 0.90, and the relative error in average particle velocity prediction reduces to less than 0.01. This hybrid machine learning model has the potential to overcome limitations of cross-correlation-based particle velocimetry, expanding its applications in multiphase flow systems. It offers a promising solution for accurate solids velocity measurement in fluidized beds.