(99c) Bridging the Gaps in Fermentation Modeling with Generative AI

Traditionally, chemical process modeling has relied on first-principles equations with constant parameters. While effective, these models often fall short in capturing the nonlinear and time-varying nature of real-world processes. To address this, we developed a hybrid modeling framework that combines physics-based dynamics with data-driven, time-varying parameters using attention-based time-series transformers (TSTs). Inspired by the transformer architecture behind ChatGPT, our model captures both immediate and historical process behaviors, offering contextual understanding of system dynamics. This TST-based hybrid model is adaptable across modeling approaches—from DFT to CFD—and applicable at scales ranging from lab experiments to industrial operations. We will demonstrate its capabilities through real-world applications enabled by collaborations with leading chemical companies.