(662a) Modeling of Multicomponent Adsorption Isotherms for Bio-Butanol Separation from ABE Model Solutions

Depletion of fossil fuel resources, political instability in oil-producing countries and environmental challenges has attractedsignificantattention to alternative biofuel production from renewable and sustainable resources. Butanol is considered as one of the best renewable alternatives for gasoline since it has advantages over other biofuels such as ethanol. However, there are some challenges in biobutanol production such as the very low concentration of butanol and the presence of other components in the acetone-butanol-ethanol (ABE) fermentation broths. The major challenge remains to develop a cost-effective technique to separate and recover butanol as the final product.

Distillation, as a traditional recovery technique,which is still used for butanol separation, is not economically viable due to its high energy requirement. Liquid-liquid extraction, gas stripping, perstraction, pervaporation and adsorption are other techniques being investigated for biobutanol separation from dilute fermentation broths. According to different investigations, the energy requirement for adsorption is the lowest compared to other techniques.

In our previous study activated carbon F-400 was selected as the best adsorbent amongst several adsorbents that were tested. Indeed, AC F-400 has high adsorption rate, capacity and selectivity toward butanol. The breakthrough experimental results for ABE model solution (butanol 12.7, ethanol 2.2, acetone 6.1, acetic acid 4.8, butyric acid 4.9, glucose 5.3 and xylose 4.2 g/L) showed that butanol is preferentially adsorbed and the othercomponents that were also adsorbed downstream of the butanol adsorption zone werelater displaced by butanol and butyric acid during the breakthrough experiments.

In this study, different isotherm models (multicomponent Langmuir,modified Langmuir and a model based on artificial neural network (ANN)) were tested to find the most suitable model to predict a wide array of standard adsorption isotherm behavior for butanol adsorption from multicomponent solutions. Results showed that the neural network most adequately described the multicomponent equilibrium data. A neural network with a hidden layer having four neurons, including the bias neuron, was able to represent very accurately the adsorption isotherm data. A parity plot comparing the neural network model with the experimental data was generated with a regression coefficient of 0.98. Isotherms of the other species present in the fermentation broth were also derived using neural networks.