Modelagem e simulação do processo de produção de etanol a partir do suco do pedúnculo de caju, visando a otimização das condições operacionais
Dissertation (Ms) 08/08/2016
Andrea da Silva Pereira
Ethanol is the most relevant biofuel in the Brazilian market, its rising demand generates interest in using other raw materials as an alternative to sugar cane. Therefore, the main purpose of the present work is to study a mathematical model for the process of cashew apple juice alcoholic fermentation by Saccharomyces Cerevisiae, making possible to check cells, sugar and ethanol concentration profiles, to optimize reaction operation conditions interfering in the reaction for scaling up the fermenter. For this purpose, laboratory experiments data of Pinheiro (2015) were used with variations of the initial cell concentration from 3 to 10 g/L parameters were varied as follows: initial cell concentration from 3 to 10 g/L, temperature from 26 to 42 °C, initial total sugar concentration from 70 to 170 g/L and stirring speed from 80 to 800 RPM. Initially, regressions and interpolations were performed to treat data, seeking for smoothing experimental data noise and increasing the amount of data for training artificial neural networks (NNAs) and to estimate parameters. Five mathematical models were proposed: four mechanistic unstructured nonsegregated models and one hybrid model that combine NNAs and mass balances. Model 1 considers negligible the substrate consumption due to cells maintenance and that all cells are viable, using the Ghose & Tyagi’s kinetic model. Model 2 differs by the hypothesis of death and cell maintenance, Model 3 derives from Model 2 with inclusion of the temperature. Model 4 adds the influence of initial concentration of cell and agitation by the hypothesis of a stagnant film surrounding the cells of flocculant S. cerevisiae. The set of parameters was determined by damped least-squares (DLS) method proposed by Marquardt. For the hybrid model (Model 5), the specific rates are predicted by NNAs as function of instant cell and substrate concentrations, temperature and stirring speed. Simulation results were statistically assessed by residual standard deviations (RSD), modified F-test and confidence interval (CI), apart from being validated by another experiment run used neither to estimate the set of parameters nor to train NNAs. Among the models studied, it was observed that the Model 4 and 5 were the most general and suitable in the studied process. The models were optimized by the bio-inspired method of particle swarm, indicated for global searches, based on the parameters efficiency and productivity of the reaction. After that, the scale up of the process for a 14 liters reactor was performed using the P/V criteria. The productivity obtained was 6 gL-1h-1 with the following optimal conditions: initial substrate concentration of 143.2 gL-1, initial cell concentration of 6.5 gL-1, temperature of 36.5 ºC and stirring speed of 80 rpm.
