Elsevier

Particuology

Volume 52, October 2020, Pages 57-66
Particuology

Scale-up effect analysis and modeling of liquid–solid circulating fluidized bed risers using multigene genetic programming

https://doi.org/10.1016/j.partic.2019.12.003Get rights and content

Highlights

  • Model validated by comparing model predicted and pilot-scale LSCFB data.

  • GP model prediction and experimental data are in agreement.

  • Statistical performance measures of the GP model are quite competitive.

  • Three different scales are compared with experiments for model validation.

  • Different parameters are considered for comparing scale-up effects.

Abstract

Understanding scale-up effects on the hydrodynamics of a liquid‒solid circulating fluidized bed (LSCFB) unit requires both experimental and theoretical analysis. We implement multigene genetic programming (MGGP) to investigate the solid holdup and distribution in three LSCFB systems with different heights. In addition to data obtained here, we also use a portion of data sets of LSCFB systems developed by Zheng (1999) and Liang et al. (1996). Model predictions are in good agreement with the experimental data in both radial and axial directions and at different normalized superficial liquid and solid velocities. The radial profiles of the solid holdup are approximately identical at a fixed average cross-sectional solid holdup for the three LSCFB systems studied. Statistical performance indicators including the mean absolute percentage error (6.19%) and correlation coefficient (0.985) are within an acceptable range. The results suggest that a MGGP modeling approach is suitable for predicting the solid holdup and distribution of a scaled-up LSCFB system.

Introduction

Liquid‒solid circulating fluidized beds (LSCFB) have attracted substantial attention in the fields of chemical and biochemical applications owing to their superior mass and heat transfer between phases and solids in regeneration facilities (Zhu, Karamanev, Bassi, & Zheng, 2000). A LSCFB comprises two interconnected circulating fluidized bed units known as a riser and a downer. Solid particles are continuously circulated between these two units using liquid as the fluidization medium. The application of LSCFB is therefore particularly appropriate for processes where contact is important between liquid and solid. The effective distribution of solids in a LSCFB system maximizes the interphase contact area. A LSCFB system accommodates high heat and mass transfer rates, different particulate materials with high liquid throughputs, prevents back mixing, and facilitates simultaneous production and regeneration.

These advantages make LSCFB a suitable choice for many industrial applications such as wastewater treatment (Cui, Nakhla, Zhu, & Patel, 2004; Muroyama & Fan, 1985; Patel, Zhu, & Nakhla, 2006), linear alkylbenzene production (Devassy, Lefebvre, & Halligudi, 2005), heavy oil upgradation (Atta, Razzak, Nigam, & Zhu, 2009), protein recovery (Lan, Bassi, Zhu, & Margaritis, 2002), phenol polymerization (Bassi, Geng, & Gijzen, 2004), and heavy metal removal (Feng, Jing, Wu, Chen, & Song, 2003). The study of solid holdup and its distribution is therefore crucial because it directly affects the design, performance, and hydrodynamics of LSCFB (Razzak, Rahman, Hossain, & Zhu, 2012). Only a few experimental hydrodynamics studies have been reported on LSCFB and other liquid‒solid systems. For suitable industrial applications of LSCFB systems, further study on the scale effects on hydrodynamic behavior is necessary. However, experimental studies at different scales are lacking owing to high experimental time and cost, and the scale effect on hydrodynamic behavior remains poorly understood.

One way to minimize the cost and experimental time of a large-scale LSCFB study is the development of mathematical correlations, numerical methods, modeling, and simulations. Model correlations can be used for scaling-up systems and predicting the hydrodynamic behavior for different scale systems. In this regard, Cheng and Zhu (2008), Cheng & Zhu (2005), Dadashi, Zhang, and Zhu (2015), and Han, Yang, Zhu, and Liu (2015) employed computational fluid dynamics, which uses different numerical methods and algorithms to analyze and describe LSCFB characteristics. We recently implemented an artificial neural network, adaptive neuro-fuzzy interface system, and support vector machine for this purpose (Razzak, 2012; Razzak et al., 2012; Razzak, Hossain, Rahman, & Hossain, 2014). Neural network (NN) models for hydrodynamic studies of LSCFB provide relatively improved predictions over conventional statistical models. However, a fundamental problem with NN models is their working characteristics, which cannot relate inputs with outputs by an analytical equation. This problem can be overcome by applying soft computing approaches known as genetic programming (GP). Lahiri and Ghanta (2008) used a NN model approach to study solid holdup in slurry pipelines. Otawara et al. (2002) used a NN model to predict bubble motion characteristics in a three-phase fluidized bed.

In the present study, we develop a multigene genetic programming (MGGP) model to predict the scale effects of LSCFB systems on hydrodynamics behavior. MGGP is a modified version of GP. Experimental data used in the model development and validation are collected from three different scales of LSCFB systems from heng (1999), Liang et al. (1996), and the present study. In these experiments, glass beads with 500-μm particles and water are used as solid and liquid phases, respectively. We consider the scale effect on the hydrodynamic behavior in axial and radial directions under various operating conditions such as superficial liquid velocity (Ul), auxiliary liquid velocity (Ua), superficial solids velocity (Us), and normalized superficial liquid velocity (Ul/Ut). The model suitability is evaluated using statistical indicators.

Section snippets

Genetic programming approach

Genetic programming (GP) is a type of evolutionary algorithm that can be used as an effective alternative to NN and regression modeling. GP is a machine learning method inspired by the Darwinian theory of natural evolution such that when GP is applied to model a physical process, fitting can be viewed as survival of the best values (Koza, 1992). GP generates a mathematical model of the process using symbolic regression of input and output data of that system (Babu & Karthik, 2007), information

Experimental set-up and methodology

All experiments were conducted in the LSCFB system installed at the King Fahd University of Petroleum and Minerals, and a schematic diagram is illustrated in Fig. 5 for system 1. Similar setups were used by Zheng (1999) (system 2) and Liang et al. (1996) (system 3). The LSCFB system has a riser, downer, liquid‒solid separator, measuring device (manometer, half butterfly valve, optical fiber probe, and pressure transducer), liquid tank, centrifugal pump, and inclined solids feed pipe and

Model development

Different model parameters and process variables were systematically examined to develop and implement the appropriate MGGP model for the LSCFB systems of this study. Details of the developed MGGP model parameter settings are given in Table 2. Fig. 6 shows a schematic flowchart that describes the developed MGGP algorithm and its implementation for the LSCFB systems.

Model evaluation

To evaluate the developed model, experimental and MGGP model-predicted data are compared in Fig. 7. The experimental data are represented by symbols and a continuous dotted line is used for the model data in Fig. 7(a). The results show good agreement between experimental and predicted data. For further analysis, solid holdup data obtained from experiments are also plotted against the MGGP predictions in Fig. 7(b). Model predictions for both the training and testing data have a close

Conclusions

Multigene genetic programming (MGGP) is used to study and model the distribution of solid holdup in three LSCFB systems. Our results show that the MGGP model is adequate to predict axial and radial solid holdup and distribution in all of the studied LSCFB systems. Furthermore, the developed model predicts the effect of superficial liquid velocity, normalized superficial liquid velocity, and superficial solid velocity on LSCFB hydrodynamics. Despite riser height differences between the three

Declaration of interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgement

The authors would like to acknowledge support provided by King Abdulaziz City for Science and Technology (KACST) through the Science & Technology Unit at King Fahd University of Petroleum & Minerals (KFUPM) for funding of this work, project No. NSTIP # 13-WAT96-04, as part of the National Science, Technology and Innovation Plan.

References (25)

Cited by (6)

  • Prediction of minimum fluidization velocity in pulsed gas–solid fluidized bed

    2021, Chemical Engineering Journal
    Citation Excerpt :

    Gas-solid fluidized beds mix solids and gas, giving the mixture fluid-like properties. Due to their strong mixing and high-heat and mass-transfer efficiency, they have been widely used in petrochemistry, energy storage, environmental protection, and pharmaceutical industries [1–5]. The vibration energy is introduced into the traditional gas–solid fluidized bed to form a pulsed gas–solid fluidized bed (PGFD).

View full text