A Closed-Loop Optimized System with CFD Data for Liquid Maldistribution Model

Wei Zhang,Liyi Li,Jian Zhai,Junwen Wang,Baoping Zhang,Xin Xu

doi:10.3390/pr8111332

Abstract

For the simulation of a trickle-bed reactor (TBR) in coal and oil refining, modeling the liquid maldistribution of the gas-liquid distributor incurs enormous pre-processing work and bears a huge computational cost. A closed-loop optimized system with computational fluid dynamic (CFD) data is therefore proposed for the first time in this paper. A fast prediction model based on support vector regression (SVR) is developed to simplify the modeling of the liquid flow rate in TBRs. The model uses CFD simulation results to determine an optimized set of structural parameters for the gas-liquid distributor in TBRs. In order to obtain an accurate SVR model quickly, the particle swarm optimization (PSO) algorithm is employed to optimize the SVR parameters. Then, the structural parameters corresponding to the minimum liquid maldistribution factor are calculated using the response surface methodology (RSM) based on the hybrid PSO-SVR model. The CFD validation results show a good agreement with the values predicted by RSM, with liquid maldistribution factors of 0.159 and 0.162, respectively.

Highlights

Trickle-bed reactors (TBRs) are widely used in the chemical and oil industries
The validity of the particle swarm optimization (PSO)‐support vector regression (SVR) model as the proxy model was evaluated by comparing the accuracy rapidity of the algorithm
We investigate the utility of using the PSO algorithm for optimizing the parameters of SVR in order to obtain an effective hybrid model (PSO-SVR) for application in modeling liquid maldistribution of the gas-liquid distributor in TBRs

Summary

Introduction

Trickle-bed reactors (TBRs) are widely used in the chemical and oil industries. TBR efficiency relies on the good distribution of liquid feed over the catalyst bed cross section, so predicting liquid maldistribution is one of the critical issues in the efficient use of TBRs [1].Several approaches have been developed to estimate liquid maldistribution. There is a possibility of flow redistribution at the exit of the bed [2,3,4] To overcome this disadvantage, several groups have focuses on the use of tomographic measurements using photon attenuation (x-ray and γ-ray tomography), magnetic resonance imagining, and electric tomographic techniques, which provide more quantitative flow distribution information [5,6,7,8,9]. With the drawback of a high cost and unsafety, the performance of these techniques depends on the complexity of the reconstruction algorithms Compared to these approaches, computational fluid dynamics (CFD) modeling has a relatively low cost and can simulate both realistic and ideal conditions [10,11,12,13,14]. A CFD simulation of the full geometry of a gas-liquid distributor may be time-consuming, depending on the domain size and the mesh number

Objectives

Discussion

Conclusion