Modeling and control process of first and second generation ethanol production applying neural networks

Author(s)

William Eduardo Herrera Agudelo

• Document type: Doctoral Thesis
• Press: Campinas, SP.
• Institution: Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Química
• Defense date: 2017-09-27

Examining board members

Rubens Maciel Filho; Carlos Eduardo Vaz Rossell; Adriano Pinto Mariano; Solange Inês Mussatto Dragone; Daniel Ibraim Pires Atala

• Advisor: Elmer Alberto Ccopa Rivera; Rubens Maciel Filho

Abstract

The present research work aims at developing and applicating computational techniques and tools, such as mathematical models, software sensors, and a neural network model based predictive control (NNMPC) for ethanol production process from hydrolysate mixture with sugarcane molasses (1G+2G). For this purpose, experimental studies of pre-treatment, enzymatic hydrolysis and fermentation were carried out for five different ethanol production systems, mainly defined by the pretreatment type and two levels of solids loading in the enzymatic hydrolysis. Experimental data of substrate, ethanol and cell concentrations, as well as online measurements (temperature, CO2 flow, pH, capacitance), among other experimental observations, were used to adjust and validate the mathematical models and generated computational tools. The developed kinetic models can predict cell growth, substrate consumption, and ethanol production for the different ethanol fermentation systems. An advanced temperature-dependent parameter estimation procedure was used to ensure the proposed models accuracy. In this approach, the temperature influence on the fermentation kinetic behavior was explicitly demonstrated. Subsequently, the kinetic models were applied in the simulation of a continuous fermentation process for ethanol production. The fermentation system is a typical large-scale industrial process composed of four fermenters attached in series and operated with cell recycling. The models were implemented in Matlab® programming language. As a result, a virtual 1G+2G ethanol production plant was obtained for each studied system. The fermentation performance of different systems was evaluated. In this work, a great prominence was given to the use of Artificial Neural Networks (ANN) in soft-sensors development for online monitoring of fermentation process of 1G+2G ethanol, as well as to obtain accurate measurements of necessary state variables for the proposed predictive control schemes. The results of soft-sensors application showed the possibility of accurate inference of substrate, ethanol, and cells concentrations from easily measurable online information. An additional advantage of this approach is the use of a relatively low-cost sensor system, such as thermocouples, pH meter, and capacitance probe. Based on the dynamic simulation of 1G+2G ethanol production plant, a NNMPC was applied to deal with fluctuations in raw material sugar concentration. The control purpose is to maintain the sugar concentration of the fourth reactor output in its set-point, by manipulating the feed flow. Through closed-loop simulations, it was verified that the proposed nonlinear control algorithms proved to be efficient and robust, since they provided satisfactory results in control problems of servo and regulatory types (AU)

• FAPESP's process: 12/24326-0 - Modeling and control processes for first and second generation ethanol production applying neural networks
• Grantee: William Eduardo Herrera Agudelo
• Support Opportunities: Scholarships in Brazil - Doctorate