Estratégia de otimização em tempo real do processo de fermentação alcóolica usando metamodelagem.

Abstract

In a world in constant growth, the production processes in which mankind can obtain their goods and consumer products are going through advances every day, in this the real-time optimization of processes comes to add and innovate this advance, in order to make processes more versatile and able to meet the needs of manufacturing units to meet an increasingly demanding market and with a behavior so variant, in the most attractive way from the economic point of view. Real-time optimization is a performance optimization technique that relies on mathematical models to adhere to the most assertive operating condition of a process and expand the profitability of an industrial plant, allowing operating facilities to respond efficiently and effectively to changing conditions of feed rates and composition, equipment, and dynamic processing economics. Software is responsible for cross-referencing the variables, performing the calculations, and running the performance simulations to induce new setpoints for the automation systems, but it is up to the plant operators to deliberate whether the software's suggestions will be used or not. Through this, this project aims to develop optimization strategies for the alcoholic fermentation process exploring metamodeling, including Multiple Linear Regression and Kriging, with Real Time Optimization for industrial application, providing the optimal operation of the process in real time. The process was simulated in SIMULINK® and metamodeled in MATLAB®, always at the optimal point. The proposal of this work initially aims to obtain an algorithm (through metamodeling), which allows an increase in the speed of resolution and efficiency of problems proposed through a case study, by applying a real-time optimizer with a high enough execution frequency to allow its application. The simulation was performed according to data presented in the thesis of Andrietta (1994), which dealt with an industrial process of alcoholic fermentation, with the system equipped with the perfect-mix reactor. And having as base the program developed in the thesis of Fernandes (2022), this program is capable of building substitute models or metamodels in an automatic way, where it is not necessary to inform the sample or metamodel data, because the algorithm provides the user with the number of samples necessary for the construction of the model, the transformations of the response variables and the best model that represents the data. The modeling inserted into the MATLAB® software made it possible to follow the stationary behavior of the variables of product, cell and substrate concentration, besides a fast way, it can be obtained predictable behavior for the fermentative process when changing adopted parameters, where the values of the modifications served only as a comparative study in relation to the initial considerations. And the application of the methodology to the system studied allows us to conclude among the seven solvers studied that the "Nomad", became the best optimizer of this process, because it showed good yield results with a Q²<0.97, and with a fast processing time compared to the others.