(353d) Developing a Hybrid Model of a Biochemical Fermentation Process

With ever-increasing energy demands, rising pollution, and over-reliance on the crude oil industry, there is a definite need for a cheap, environment-friendly energy source. Among various renewable sources, ethanol has been perceived as one of the most promising candidates, one which can meet the growing global energy demands. At present, primary methods of ethanol production are the chemical conversion of food sources like corn with a high starch content [1], and thermochemical conversion of lignocellulosic biomass to ethanol [2]. However, both have drawbacks like sustainability in both food and fuel market or high energy demands, respectively. Hence, the recent focus is shifting towards biochemical production of ethanol, which has the potential to utilize many different cellulosic biomass substrates and pretreatment technology options without the use of food resources, which solves the issues of the previous methods [3].

It is important to develop a mathematical model to understand and describe the fermentation process, predict optimal parameters, and help in future automation. However, representing such a process is quite complicated as it has coupled nonlinear differential equations that are difficult to solve and often have uncertain parameters or states. It is difficult to incorporate everything occurring inside a plant, in a model which may result in a plant-model mismatch between optimal and actual ethanol production rate. This mismatch is mainly due to assumptions of constant parameters, a lack of complete process information, and homogeneity of properties for making the process computationally less intensive. Hence, it is crucial to identify the uncertain parameters resulting in this plant-model mismatch. To this end, we perform a global sensitivity analysis by varying the nominal parameters in a stipulated range to obtain the set of the most important model parameters, which profoundly affects the model outputs.

Based on the global sensitivity analysis results, we built a neural network model with temperature as input and ethanol rate as the output maximizing the productivity and keeping the ethanol concentration constant. We then combined it with the first principle model to develop a hybrid model that gives us better prediction accuracy, extrapolation capabilities, and generalization enabling us to reduce the existing plant-model mismatch [4]. Once we had this hybrid model, we needed to estimate the state variables in the process to maintain them at optimal operating conditions adequately. For this, we designed a state observer that estimated the internal states of the biochemical system like the ethanol production and substrate concentrations. We then validated our hybrid model and the designed observer against an entirely new set of process data, and results showed that the model works well in describing the process. Then, we determined optimal conditions to achieve a set of target values of the ethanol production rate while ensuring the safety of the operation and separation of the product from biological residues. It is essential to develop a relationship between the estimated optimal production rate of ethanol and the variation of temperature needed to attain it. In this regard, we developed a Model Predictive Control (MPC) algorithm to maintain the temperature at suitable levels considering all practical plant constraints. Finally, we combined these algorithmic models and implement them on a real industry-scale ethanol fermentation plant and demonstrated the performance of the proposed MPC algorithm with real plant data and monitoring.

1. Thompson, P., The Agricultural Ethics of Biofuels: The Food vs. Fuel Debate. Agriculture, 2012. 2: p. 339-358.

2. Niziolek, A.M., et al., Coal and Biomass to Liquid Transportation Fuels: Process Synthesis and Global Optimization Strategies. Industrial & Engineering Chemistry Research, 2014. 53: p. 17002-17025.

3. Raftery, Jonathan Patrick(2017). Continuous Biochemical Processing: Investigating Novel Strategies to Produce Sustainable Fuels and Pharmaceuticals. Doctoral dissertation, Texas A&M University.

4. Bangi, Mohammed Saad Faizan, and Joseph Sang-Il Kwon. “Deep Hybrid Modeling of Chemical Process: Application to Hydraulic Fracturing.” Computers & Chemical Engineering, vol. 134, 2020, p. 5. Crossref, doi:10.1016/j.compchemeng.2019.106696.