In the first part of this talk, we will present a multiscale modeling and simulation framework for biomass fractionation processes. This framework elucidates the relationship between molecular-level processes, such as lignin depolymerization, condensation and demethoxylation, and macroscopically observable process behavior. It also enables the computation of optimal design and operating conditions. Using biomass fractionation as a model system, the framework integrates several computational approaches: (a) ab initio molecular dynamics simulations to determine stable lignin structures at various temperatures; (b) density functional theory calculations to calculate energy barriers for key molecular processes based on specific lignin structures; (c) kinetic Monte-Carlo modeling to simulate molecular-level reactions and predict the evolution of lignin's molecular weight distribution and monomeric composition; and (d) an integrated multiscale computational approach that links molecular-level models with macroscopic continuous-phase equations, applicable to various biomass fractionation systems. The parameters and predictions from this multiscale model are validated through close collaboration with experimental groups.
In the second part of the talk, we will present model predictive controller designs that utilize insights from the multiscale modeling work and experimental measurements to manipulate operating conditions, enabling the production of lignin chains with desired molecular weight distributions and monomeric compositions. To address feedstock variability, a significant challenge in industrial biomass fractionation owing to agricultural cycles, seasonal changes, and geographic diversity, we will introduce a machine learning-based parameterization scheme. This approach employs the expectation-maximization algorithm and a Gaussian mixture model to efficiently update the controller's model parameters when processing biomass from various sources. By doing so, the framework ensures consistent production of lignin chains with desired molecular properties, regardless of feedstock variability.
