Modelling and uncertainty assessment for Simultaneous Saccharification and Co-Fermentation (SSCF) processes
Conference poster, 2017
Bioethanol production from lignocellulosic raw materials has attracted interest as a renewable and sustainable alternative to fossil fuels in the transportation sector. Compared to fossil fuels it reduces greenhouse gas emissions and contributes to mitigating climate change. Thus, increased effort has been expended to develop sustainable and at the same time economically feasible process configurations for bioethanol production. A widely used process configuration is the simultaneous saccharification and co-fermentation (SSCF). In SSCF, pre-treated lignocellulosic raw material is enzymatically hydrolyzed to pentose and hexose monomers which are simultaneously fermented to ethanol by recombinant Saccharomyces cerevisiae yeast, as the enzymatic hydrolysis proceeds.
The SSCF process configuration has several hurdles that remain to be overcome. The optimal hydrolysis temperature is 50-55 °C whereas the optimal cultivation temperature for S. cerevisiae is 30-35 °C. High solid loadings (> 20 %) lead to severe mass and heat transfer limitations, dividing the bioreactor in different zones which may have large effects on enzymes and microorganisms. Additionally, with ongoing hydrolysis, viscosity changes which makes it difficult to provide optimal pH control throughout the fermentation. To find the optimal process configuration, mathematical modelling techniques can be applied.
Mathematical modelling enables a rational design of processes based on existing process data. The validity of models relies on the measurement data which are used for parameter estimation. Thus, model validity is constrained to the quality of data generation in terms of the precision and accuracy of measurements and the quality of the process (e.g. batch-to-batch variation). In SSCF processes such variations are introduced by several sources. The natural turbidity and particle contents of the medium impede photometric measurements. Cell measurements of colony forming units, a commonly applied method, suffers from high variability. The high viscosity impedes representative pH measurements and sampling during SSCF processes, both of which lead to a high variance in measurement data. Furthermore, difficulties to accurately measure the raw material properties result in problems to link these properties to the bioprocess performance. Therefore, developed SSCF models are usually valid only in narrow ranges, e.g. for one type of raw material. This calls for tools to cope with process variation for industrial and economical SSCF process development.
A powerful tool reducing technical variation is experimental design. In a first step, we will use factorial Design of Experiments (DoE) to investigate the impacts of steam explosion pre-treatment on the composition, hydrolyzability and fermentability of the pre-treated raw material. Thereby, we will link the raw material properties to bioprocess performance. In a second step, we will apply an optimal experimental design strategy to the validation of a SSCF bioprocess model which is further used for model-based process optimization. With the application of a model-based optimal experimental design we aim to design experiments such that the model parameters are estimated with the lowest possible inaccuracy. They can be used to identify and remove parameters causing ill-posed problems, e.g. caused by linear correlations, and to measure the uncertainty of the remaining parameter estimates. This would give valuable information about the reliability limits of the developed model. Moreover, the optimal experimental design for parameter estimations will take into account the technical variance and, since there will be replicate runs of the same designs, also the batch-to-batch of the bioprocess.
The utilization of a process model including the fermentation parameters, which are responses of the factorial DoE, allows for building a direct relationship between the raw material properties and the SSCF model. Thus, pre-treatment parameters can be related to fermentation outcomes, leading to a general process optimization scheme. Since the DoE includes replicates of sample points, it is even possible to include the batch-to-batch variation of the pre-treatment process in the optimization scheme. Using model-based optimization instead of response surface methodology take the highly non-linear nature of the SSCF process into account and improves the reliability of the optimization results.