Abstract

The total gas holdup and the holdup of large gas bubbles were predicted in bubble column reactors (BCRs) and slurry bubble column rectors (SBCRs) using two Back-Propagation Neural Networks (BPNNs). Over 3880 and 1425 data points for gas holdup and Large gas bubble holdup respectively, covering wide ranges of gas-liquid-solid physical properties, operating variables, reactor geometry, and gas sparger type/size, were employed to develop, train and validate the two neural networks. The developed BPNN for gas holdup has a topology of [14,9-7,1] and was able to predict the trained and untrained data with an average absolute relative error (AARE), standard deviation, and regression coefficient (R²) of 16, 19 and 90%, respectively. The developed BPNN for large gas bubble holdup has a topology of [14,8,1] and was capable of predicting the trained and untrained data with AARE, standard deviation, and R² of 10, 14 and 93%, respectively. The BPNNs were then used to predict the effects of pressure, superficial gas velocity, temperature and catalyst loading on the total syngas holdup for Low-Temperature Fischer-Tropsch (LTFT) synthesis carried out in a 5 m ID SBCR. The predicted total syngas holdup appeared to increase with increasing reactor pressure, superficial gas velocity and the number of orifices in the gas sparger. The predicted syngas holdup, however, was found to decrease with increasing catalyst loading and reactor temperature. Also, under similar LTFT operating conditions (P = 3 MPa, T = 513 K, C_W = 30 and 50 wt%), the total syngas holdup values predicted for H₂/CO ratio of 2:1 and cobalt-based catalyst are consistently lower than those obtained for H₂/CO ratio of 1:1 and iron oxide catalyst in the superficial gas velocity range from 0.005 to 0.4 m/s. These predictions are in perfect agreement with reported literature trends, which underscore the reliability and validity of the developed BPNNs in predicting the total syngas holdup and the holdup of large gas bubbles in large-scale bubble columns and SBCRs operating under industrial conditions.

Recommended Citation

Behkish, Arsam; Lemoine, Romain; Sehabiague, Laurent; Oukaci, Rachid; and Morsi, Badie I. (2005) "Prediction of the Gas Holdup in Industrial-Scale Bubble Columns and Slurry Bubble Column Reactors Using Back-Propagation Neural Networks," International Journal of Chemical Reactor Engineering: Vol. 3: A53.
Available at: http://www.bepress.com/ijcre/vol3/A53