Predictive modeling can de-risk bio-based production

Technologies devised for bio-based production are typically based on single feedstock types. This approach is applicable in the Midwest where there is a sufficient single feedstock, corn stover, but not for several parts of United States that have abundant but diverse feedstocks. In order to reduce the risks associated with biomass availability, it is essential to develop technologies that utilize variable and multiple biomass feedstock streams, simultaneously. In this project, using statistical design of experiments defined by SAS JMP®, we developed a predictive model to optimize biomass mixtures, pretreatment types, and pretreatment reaction conditions to maximize sugar yield and minimize inhibitor production. Three biomass feedstocks: corn stover, switchgrass, and energy cane were pretreated with dilute alkali (DL), dilute acid (DA), and ionic liquid (IL) at severities predetermined by the model and subsequently, enzymatically hydrolyzed. The resulting sugar yields, which varied from 5 to 90% theoretical, were then fed into the model that generated a continuous envelope of optimal feedstock mixtures and reaction conditions. Model predicted that a biomass mixture consisting of 84, 6, and 10% (w/w) stover, switchgrass, and energy cane can deliver 70% (theoretical) sugar yield when DL was conducted on it at 103^°C for 16 hours. Techno-economic analysis was performed for IL in ASPEN Plus® and compared with the established DA model to identify economic bottlenecks for the IL system. In the future, such predictions coupled with experimental and economic evaluations can be applied to convert mixed feedstocks in several geographical areas and thereby help de-risk biobased production.