Sistemas Inteligentes Aplicados à Otimização de Parâmetros Cinéticos e Modelagem de Bioprocessos

Abstract

Technological advancement is taking the industry to the fourth revolution – Industry 4.0 –, characterized by the integration between production system and information and communications technology. From an operational point of view, the use of these technologies aims to reduce costs with raw materials, labor and production time, resulting in a higher productivity. Among the techniques used are Artificial Intelligence and metaheuristic optimization methods, for the analysis and simulation of virtual models capable of representing real processes. In Chemical Engineering, it is common to use computational methods to solve and improve complex problems through classical techniques of modeling, simulation and optimization of processes. Although, there are difficulties with this approach, which depends on knowledge of chemical and physical phenomena, complex mathematical representation; and the distinction between local and global optima in optimization. Regarding bioprocesses, biological phenomena must be considered, since the organisms involved are sensitive to several influences, often difficult to determine. In this scenario, this academic work made a construction of strategies for determining kinetic parameters, empirical modeling and simulation of alcoholic fermentation processes, from real experimental data, through metaheuristic optimization methods and machine learning. The kinetic parameters were obtained through Particle Swarm Optimization (PSO) and Genetic Algorithm (GA), while Artificial Neural Networks (ANNs) and Support Vector Machines (SVMs) were used in the empirical modeling and simulation of the process, implemented in the Python programming language, with the strategies being built, evaluated and validated. The results showed the efficiency in the use of these tools in Chemical Engineering, especially in the Bioprocesses, where the use of the Genetic Algorithm in the estimation of kinetic parameters showed results consistent with the academic literature, while Artificial Neural Networks and Support Vector Machines, applied to the empirical modeling of the fermentation process, demonstrated high predictive capacity of the main process variables.