In a previous work (Sanchez et al., 2006), a preliminary approximation to the pro­cess synthesis of ethanol production from lignocellulosic materials employing the optimization-based approach was presented. The analyzed system comprised the step of biological transformation of the pretreated feedstock through different tech­nological options (separate hydrolysis and fermentation, simultaneous saccharifi­cation and fermentation, simultaneous saccharification and co-fermentation) and the step of ethanol dehydration using distillation. The initial stream entering the system contains the main components that are formed after the pretreatment using dilute acid, i. e., cellulose, pentoses (mainly xylose), glucose, lignin, and water. The system should process this stream in such a way that the final product stream has an ethanol content greater than 99.5% wt. For this preliminary study, the wastewater treatment step was not considered.

For tackling such a complex process as the bioethanol production, the Jacaranda synthesis package was employed, which has been described elsewhere (Fraga, 1998; Fraga et al., 2000). It has been successfully applied to the preliminary design of a hydrofluoric acid plant (Laing and Fraga, 1997), the generation of optimal downstream processing flowsheets of bioprocesses (Steffens et al., 1999b), and the process synthesis for the microbial production of penicillin (Steffens et al., 1999a). Any optimization-based strategy for process synthesis requires the implementa­tion of models of process units. In this case, the biological transformations were described by kinetic models considering the cellulose hydrolysis, glucose forma­tion and consumption, cell growth, and ethanol biosynthesis (South et al., 1995; see Chapter 7, Case Study 7.2). When co-fermentation using recombinant bacteria was

taken into account, the model reported by Leksawasdi et al. (2001) was used (see Section 7.2.2). For the calculation of distillation units, the Fenske-Underwood — Gilliland (FUG) method was utilized considering the presence of binary azeo­tropes in the system ethanol-water. The separation section of the process, used to generate 99.5% pure ethanol, consisted of distillation units alone. As there was an azeotrope formed by water and ethanol, an extractive distillation step was used with ethylene glycol as the solvent.

The objective function used, in this case, is net revenue defined as the value of the ethanol produced minus the annualized cost of the process, which is a function of both capital and operating costs. For the continuous bioreactors, operating and capital costs are directly related to the residence time. The capital costs of distilla­tion units are related to the vapor velocity inside the columns and to the number of stages, and the operating costs are linked to the energy consumption (mainly heat duty). The problem posed to Jacaranda consists of a superstructure, which is shown in Figure 11.14. The reaction section has a choice of three paths, the SSF reactor, the SSCF reactor, and the combination of cellulose hydrolysis followed by hexose fermentation (SHF). The separation section consists of three distillation steps: con­centration column, extractive column, and recovery column with a recycle of the solvent to the extractive column. The superstructure is the basis for an MINLP model. This model has some characteristics that make it difficult to solve, such as the physical properties models used (NRTL [nonrandom two-liquid model] in this case) and the equations for the concentrations in the reactors as a function of resi­dence time. These are present in the optimization problem as equality constraints and are difficult to satisfy. Furthermore, the use of different hot and cold utilities was allowed to meet the heating and cooling demands of any process alternative. Using discrete utilities means that the objective function is discontinuous even as a function of only the real valued variables. Furthermore, the capital cost function for the distillation units uses integer values for the number of stages determined by the FUG procedure, also leading to discontinuities in the objective function. The result is that the overall optimization problem is not solvable using standard mathematical programming approaches.

Jacaranda provides access to a number of optimization procedures including direct search methods (Kelley, 1999) and stochastic methods, such as genetic algo­rithms (Goldberg, 1989) and simulated annealing (van Laarhoven and Aarts, 1987).

In this case study, it was decided to use the genetic algorithm (GA) approach. The GA uses a replacement policy for the population at each generation, with an elite size of 1, a mutation rate of 10%, a crossover rate of 70%, and a roulette wheel selection procedure. The fitness function is based on the objective function value directly with infeasible solutions discarded if they arise (which they do with a fre­quency of approximately 5 to 6%).

For this first attempt at automated design for the production of ethanol from biomass, the number of degrees of freedom was limited. Specifically, the residence times of each reactor and the top and bottom key component recoveries in each distillation column were selected as the manipulating variables. Therefore, four residence time variables and six recovery variables were manipulated. The super­structure makes use of two binary variables for identifying the path taken through the reaction section of the process.

The results obtained identify the SSCF configuration as the best performing for the given process. This is reasonable given the high degree of integration achieved with this configuration, which makes possible the immediate consumption of the glucose formed during the cellulose hydrolysis. In this way, the inhibition of cellulose-degrading enzymes (cellulases) is avoided. In addition, the utilization of xylose allows an increase in the content of fermentation sugars and, therefore, in the overall amount of produced ethanol. This enhanced utilization of the feedstock is not characteristic for the SSF process. The SHF option implies the utilization of an additional bioreactor (the enzymatic hydrolysis and the fermentation are car­ried out in different units), which involves the increase in the capital costs for this configuration. Jacaranda allowed the determination of the values of the operating parameters corresponding to the separation section. In particular, the make-up of ethylene glycol and the recycle stream flow rate are determined automatically.

Early results demonstrate that the genetic algorithm used by Jacaranda handles the complexity of the problem design robustly with respect to the numerical diffi­culties that may arise. The solutions obtained show variability in the technological option. From 10 different runs, three of the solutions corresponded to SSCF con­figurations (two of them with the best values of the objective function), six solutions to the SSF process, and one solution to the SHF configuration.

Undoubtedly, the development of this approach will make possible the synthesis of technological flowsheets considering the structure of the system on a mathemati­cal programming basis. The complementation with tools of a knowledge-based approach will allow gaining a deeper insight of the overall process needed for the synthesis of technological configurations with increased performance.