Prediction and analysis of bed-to-tube heat transfer in fluidized bed heat exchangers based on ANN-PSO hybrid approach and CFD simulation

Fluidized bed heat exchangers (FBHXs) are essential for enhancing energy efficiency in various industrial applications. However, optimizing their heat transfer performance requires a comprehensive understanding of the complex interactions among operational variables. A challenge lies in accurately capturing these nonlinear relationships and assessing the relative importance of the operational variables affecting the bed-to-tube heat transfer coefficient (HTC). To address this issue, an ANN-PSO hybrid model was developed to predict the bed-to-tube HTC, with the artificial neural network (ANN) capturing the nonlinear relationships between variables and the particle swarm optimization (PSO) algorithm for optimizing the model parameters. Computational fluid dynamics (CFD) simulation data were used to generate a comprehensive dataset for studying the effects of particle size, fluidization velocity, bed temperature, and wall temperature on the bed-to-tube HTC. The model's architecture was optimized by screening various configurations, leading to the selection of the best three-layer feedforward backpropagation structure (4:10:1), which significantly improved the model's performance. The model was evaluated using R² and RMSE metrics, achieving good predictive accuracy (R² = 0.9168, RMSE =31.2758). Furthermore, the relative importance analysis further revealed that particle size was the most influential parameter, contributing 42.4% to bed-to-tube HTC variation. These findings present a practical and efficient modeling approach for FBHX design and optimization, offering a viable alternative to computationally expensive CFD simulations and facilitating future integration into engineering tools for industrial-scale design and optimization.