Towards Industry 4.0 in the Bioprocessing Industries: ’Real-time’ monitoring and control of lignocellulosic ethanol fermentations

Abstract

Finding renewable alternatives to fossil fuels for the production of fuels and bulk chemicals is fundamental to sustain the life of future generations on Earth. For many years, bio-based ethanol has been produced as a renewable alternative to alleviate the dependence on fossil fuels. However, the low gasoline prices and the lack of firm governmental support in most countries have hindered ethanol from becoming a consolidated liquid fuel. Nowadays, the increasing demands for fossil fuels caused by a growing world population and the environmental effects associated with the use of fossil fuels are slowly encouraging many countries to promote the production of fuels using renewable feedstocks such as lignocellulosic biomass. However, nearly a hundred years of bio-based ethanol production shows that transitioning from a fossil-based to a bio-based economy is challenging and needs to be further developed in order to meet the demand for fuels. The objective of this Ph.D. thesis is to address the main challenges limiting the development of lignocellulosic based fermentations from a process engineering perspective to optimize the production of renewable fuels and hemicals. The development of the lignocellulose-to-ethanol industry reveals that the inhibitors generated during the pretreatment of the biomass, the diauxic growth on the different carbon sources, the inherent variability of the feedstocks and the risk of contamination are the major challenges hindering the consolidation of lignocellulosic ethanol at industrial scale. While some of these challenges have been partially approached using strain engineering, the present thesis addresses them developing ‘real-time’ monitoring and control frameworks that increase the operational flexibility of the fermentation to account for the substrate variability, to limit the inhibitory effects, and to promote the co-consumption of the different substrates. After evaluating the most relevant monitoring targets and benchmarking the available measurement technologies in fermentation processes, the cell biomass, and the composition of the liquid phase were chosen as variables to be monitored in ‘real-time’. The cell biomass was monitored using flow cytometry and multiwavelength fluorescence (MWF) spectroscopy, whereas the compounds in the liquid phase were monitored using attenuated total reflectance mid-infrared (ATR-MIR) spectroscopy. Flow cytometry was used to evaluate the changes in the membrane potential, permeability, and the cytosolic concentration of reactive oxygen species of Saccharomyces cerevisiae upon contact with the inhibitors. Both, the stress generated by the inhibitors and the adaptative response to them were reflected in the membrane potential and permeability, allowing for the ‘real-time’ in-depth assessment of the physiological conditions of S. cerevisiae during lignocellulose-to-ethanol processes. At-line MWF spectroscopy was applied to monitor the biomass concentrations indirectly by measuring the tryptophan concentration. On-line spectroscopy and data-driven models were applied to monitor the concentration of glucose, xylose, and ethanol inside the fermentor. In addition to that, mechanistic models describing the dynamic change of the different state variables were developed to forecast the progress of the fermentation. Since the mechanistic models cannot account for process deviations, the data-driven and mechanistic models were combined into hybrid models. Two different configurations of the hybrid model were implemented: first in series and then in parallel. The serial configuration used the ‘real-time’ measurements from the data-driven model to update the kinetic parameters of the mechanistic model recursively, allowing to make new long-horizon predictions, including the collected measurements during the actual fermentation. The parallel configuration used a continuous-discrete extended Kalman filter to fuse the measurements from both models and to update the estimates of the fermentation states. Both approaches allowed for a digital representation of the fermentation running in parallel to the process that can be used to detect process deviations (e.g., caused by lactic acid bacteria contaminations) and to calculate the end-time of the fermentation. Optimizing the operation of the process is the ultimate goal of implementing monitoring schemes. For this reason, the data-driven models for glucose were combined with a closed-loop feedback controller that automatically adjusted the feed-rate in lignocellulosic ethanol fed-batch fermentations to operate at a constant glucose concentration. By keeping the glucose set-point low, it was possible to limit the inhibitory effects and to promote the co-consumption of glucose and xylose, resulting in volumetric productivities that were between 20-33 % higher than the reference batch process. The monitoring methods developed and implemented throughout this thesis provide an operational framework based on existing measuring technologies that improve the performance of fermentation processes using lignocellulosic feedstocks and can be extended to other fermentation processes.