Artificial neural network model for predicting minimum fluidization velocity and maximum pressure drop of gas fluidized bed with different particle size distributions

Abstract

The minimum fluidization velocity (Umf) and maximum pressure drop (ΔPmax) of a gas-solid fluidized bed are important hydrodynamic characteristics. The accurate information of these characteristics is required for obtaining the optimum design and operating conditions. In this study, a multi-layer perceptron (MLP) based on an artificial neural network was developed to accurately predict these hydrodynamic characteristics dealing with the influence of the particle size distribution (PSD). The MLP model parameters were adjusted by the backpropagation learning algorithm using wide ranges of experimental data from conducted experiments and collected literature. The five influential dimensionless groups of parameters were used for simultaneous estimation of the Umf and ΔPmax. Statistical accuracy analysis confirmed that a two-layer feedforward with thirteen hidden neurons was the best architecture for the MLP model in terms of absolute average relative deviation (AARD), mean square error (MSE) and regression coefficient (R2). The accuracy of Umf and ΔPmax was 10.36% and 8.35% with AARD, 1.7 × 10−4 and 0.0188 with MSE, and 0.9935 and 0.9152 by R2, respectively. Besides, the predictive performance of the developed model was compared with other literature models. The comparison shows the performance of the developed MLP model was acceptable.

Keywords

• Artificial neural networks
• Bed inventory
• Bed temperature
• Maximum pressure drop
• Minimum fluidization velocity
• Particle size distribution (PSD)