Hybrid machine learning for scale-up of biomass production using photobioreactors

Artificial intelligence (AI) and machine learning (ML) have found widespread acceptance in the field of chemical engineering for the scale-up of processes. However, the translation of these scale-up strategies to biological processes has been challenging due to the emergence of unexpected phenomenology, including new metabolic pathways, during scale-up. While a key issue is the lack of interpretability and explainability in developed algorithms, the primary challenge lies in the transferability of ML models across different scales. To make these models readily deployable, it is crucial to incorporate comprehensive process information while harnessing the potential of AI. In this work, we present insights into a framework which uses hybrid models combining first-principles knowledge about photobioreactors and AI, validated using experiments at different process scales.

We present this framework in the context of scale-up of biomass production in photobioreactors. Using small-scale experiments with volumes of 500 mL and 3 L; a medium-scale experiment with 30 L experiment is also conducted, which is used to validate our approach. This comprehensive approach not only addresses the challenges of biological process scale-up but also ensures the reliability and adaptability of the developed models for real-world applications.