Modelagem matemática e simulação computacional aplicadas ao estudo de processos de co-digestão anaeróbia de glicerol e/ou melaço

• Autor(a) principal: Ferreira, Carolina Machado
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFSCAR
• Texto Completo: https://repositorio.ufscar.br/handle/ufscar/17201
• topic: Glicerol; Melaço de cana-de-açúcar; Biometano; Co-digestão anaeróbia; Rede neural; Lógica fuzzy; Simulação computacional; Glycerol; Sugarcane molasses; Biomethane; Anaerobic co-digestion; Neural network; Fuzzy logic; Computational simulation; ENGENHARIAS::ENGENHARIA QUIMICA::PROCESSOS INDUSTRIAIS DE ENGENHARIA QUIMICA

Policies worldwide are increasingly incentivizing and pressuring countries to use renewable sources of energy. Brazil was a pioneer in the production of bioethanol and is one of the main producers of biodiesel. Due to the expanding production of biodiesel in Brazil, large quantities of crude glycerol are generated as a byproduct, while the high cost of its purification means that it is discarded by small and medium sized producers. Given the environmental impacts of such disposal, as well as the energy potential of glycerol (since it is carbon-rich), an attractive application of crude glycerol is in the generation of biomethane by means of anaerobic co-digestion, involving the decomposition of the organic matter in two or more substrates by bacteria and archaea, in the absence of oxygen. Co-digestion enables balancing of the availabilities of nutrients, so that biogas production is optimized. Like glycerol, sugarcane molasses has widespread availability, as a byproduct of sugar crystallization, and contains nitrogen, potassium, calcium, and other elements at concentrations that enable it to be considered as a complementary substrate for use in co-digestion processes. Computational simulations using artificial neural networks and fuzzy logic offer a way to rapidly predict the production of biogas in different scenarios. Therefore, the objective of this work was to evaluate the potential of these artificial intelligence techniques in the production of biomethane from the anaerobic co-digestion of glycerol and sugarcane molasses. Firstly, experimental data reported in the literature were used, where mixtures had a composition of distillery water in the range from 95 to 100%, and a concentration of glycerol and sugarcane molasses from 0 to 5%. A reactor model was implemented using Scilab, with Monod kinetics involving two substrates and an intermediate (M2SI model), in order to generate a database for subsequent fitting and evaluation of neural and fuzzy models. The neural network package of Matlab was used, with evaluation of the effect of the number of neurons in the networks and the distribution of data used for the training, validation, and testing sets. The Matlab package includes the multilayer perceptron artificial neural network framework and the Levenberg-Marquardt backpropagation algorithm (for training). Fuzzy modeling was applied using the Takagi-Sugeno approach available in the ANFIS package of Matlab. A Gaussian membership function and a hybrid algorithm were used for the training. The biomethane production results simulated by M2SI showed very satisfactory predictions for 8 scenarios, which were used in neural network modeling, firstly employing a “generic” network applicable to all 8 scenarios. A very good fit was obtained (R² > 0.99). The minimum quantity of neurons in the hidden layer was 14, with a small error between the intended output value and the output variable simulated by the neural network. Excellent performance was also observed for specific artificial neural networks (one for each condition). The kinetic parameters of the M2SI model for the 8 different conditions were also mapped using an artificial neural network, as a function of the organic material composition. In this case, due to the relatively low volume of data, test data were not allocated, with 90% and 10% being used for training and validation, respectively. A fit with R² > 0.99 was obtained using 25 neurons. In the case of the fuzzy logic, RMSE of 18.88 mL of methane was obtained with 216 rules, which was a value lower than 0.5% of the order of magnitude of the accumulated methane. It could be concluded from the results that fuzzy logic and artificial neural networks offer excellent ability to predict methane production, as well as to parameterize the M2SI kinetic model (using neural networks).