Abstract
In different parts of the world researches are focusing in the development of new technologies for liquid fuel production based on renewable resources. In this context, as a major producer of sugarcane-based bioethanol, Brazil aspires to reduce its fossil fuel consumption and its associated impacts on environment. These types of products are becoming an important option for the country sustainable development. In this context the broad objective of this work is to propose and evaluate an efficient way of reducing bioethanol fuel production costs, through the development of tools that allow the process to run under control, even when possible fluctuation on feedstock are present.
In this work it was developed a software sensor to infer the concentration rates of substrate, biomass and product from secondary measurements of pH, turbidity, CO2 flow rate and temperature. The software sensor uses a hybrid neural model to combine a multi-layer Artificial Neural Network (ANN) and the mass balance which describes the fermentation process kinetics. The experimental data used on the hybrid model training were obtained from fermentations that took place between 30 °C and 38 °C, measured at every 2 °C interval. The raw material for the fermentation is a mixture of 75 % hydrolyzed sugarcane bagasse and 25 % sugarcane molasses. This kind of composition may be a possible feedstock for is typical of second generation ethanol production. The three Hybrid Neural models developed are robust models that describe adequately the fermentation process even in the presence of changes in operating conditions. This is therefore a powerful tool for the prediction of kinetic rates in fermentation processes, which eventually may be used for online applications.
