Dynamic Modeling and Parameter Estimation for Unit Operations in Lignocellulosic Bioethanol Production

FEDERICO ANDERSEN; SUSANA MORENO; MARIA SOLEDAD DIAZ

Pittsburgh

Congreso; AIChE Annual Meeting 2012; 2012

AIChE (American Institute of Chemical Engineers)

Lignocellulosic biomass in nature is by far the most abundant and lowcost feedstock for the production of the socalled second generation biofuels. These biofuels are likely to replace first generation ones in the future to avoid the controversies resulting from the use of food crops fortransport fuel production. Lignocellulosic ethanol is one of the major secondgeneration biofuels and a considerable amount of research is being made in order to develop appropriate technologies for achieving a large conversion through biological processes (Martín & Grossmann, 2012; Andersen et al., 2011). Direct bioconversion of lignocellulosic materials to ethanol needs a pretreatment stage aimed to remove lignin and hydrolyze hemicelluloses. This operation makes cellulose more accessible allowing the liberation of fermentable sugars from their carbohydrate polymers in a further hydrolysis stage. In the following microbial fermentation, the sugars extracted from cellulose and hemicellulose are converted to ethanol. Although different pretreatment techniques have been proposed, the dilute acid hydrolysis is preferred for industrial applications due to its simplicity and high sugar yield from hemicelluloses. However, it has some disadvantages such as generation of toxic compounds, like furfural and hydroxymethyl furfural, which can be inhibitory to microorganisms in the downstream process. Therefore, an additional operation of detoxification is required after this pretreatment where the toxic materials are converted to less inhibitory components. A number of kinetic models for the hydrolysis, detoxification, and fermentation operations involved in the production of bioethanol have been reported in the literature. However, they have not included process variables such as operating temperatures, concentrations of inhibitory compounds, etc., that would allow evaluating their influence when they are used in process simulation and optimization studies. Consequently, the focus in this study is on the parameter estimation of kinetic models considering the main process variables for the unit operations of hydrolysis, detoxification, and fermentation in a bioethanol production plant. More specifically, we aim at identifying realistic dynamic models for the dilute acid hydrolysis, the detoxification by addition of Ca(OH)2(overliming) and the simultaneous fermentation of mixtures of glucose and xylose (cofermentation). Particularly, for the diluted acid hydrolysis, the model postulated by Lavarack et al. (2002) has been adopted. They proposed several schemes for the production of xylose, glucose, furfural, and soluble lignin, considering the influence of temperature, acid concentration, and mass ratio of solid to liquid. In this work, we have extended original model with the inclusion of acetic acid generation. Regarding the overliming process, the model published by Purwadi et al. (2004) was selected. In this model the sugars or furans generate transient complexes with calcium ions which are degraded to a certain product. Here, we have proposed that the reaction rate constants are represented by the Arrhenius equation to account for the influence of the temperature in the process. Furthermore, a new expression for determining the initial concentration of calcium cation related to pH level was proposed. Finally, for the cofermentation operation we have modified the unstructured model developed by Leksawasdi et al. (2001) by including a new factor that accounts for the furfural toxicity. Also, a first order kinetics with respect to furfural and biomass concentration is proposed for representing the conversion of furfural by the fermentative microorganism. The original high number of estimated parameters was significantly reduced in this study. In this work we have formulated the dynamic parameter estimation problem for the above mentioned models within a simultaneous approach. In this approach, each differential algebraic equation (DAE) model is discretized using orthogonal collocation on finite elements (Biegler & Zabala, 2008, Estrada et al., 2009). Thus, the DAE constrained optimization problems are transformed into largescale nonlinear programming (NLP) problems using a weighted leastsquares objective function. In order to accomplish parameter estimation, the experimental data of Cassales et al. (2011), Purwadi et al. (2004), and GutiérrezPadilla & Karim (2005) are used for hydrolysis, overliming, and fermentation processes, respectively. Main unknown kinetic parameters such as activation energies, preexponential factors in the hydrolysis and detoxification operations and the maximum overall specific growth rate and furfural inhibition coefficient for the fermentation process have been obtained. Results show reasonable agreement with the reported experimental data.