Online capacitance-integrated development of a hybrid bioprocess model for industrial fungal fermentations

The growing demand for fermentation-based products has increased interest in bioprocess modelling approaches compatible with routinely available industrial process data. Capacitance probes are promising because they provide online data related to viable biomass, but their use in filamentous fungal fermentations remains challenging, as state-of-the-art cell-count-based calibration methods of capacitance data are not applicable to hyphal organisms under industrial settings. This study presents a capacitance-integrated hybrid modelling workflow for highly viscous, enzyme-producing filamentous fungal fermentations at pilot scale under data-scarce conditions. The oxygen mass transfer coefficient k L a was developed using an intra-batch design covering agitation, pressure, and aeration dynamics. The correlation achieved a model fit of R 2 = 0.95 and approximately 8% relative validation error. Capacitance data were used to define a pseudo-viable biomass state based on a single CDW-derived anchor point, reducing the need for repeated offline CDW measurements. A ±20% perturbation of this calibration point had only a minor effect on overall model accuracy within the dataset investigated. The hybrid model was assessed on two pilot-scale fermentations with distinct operating trajectories. One of the two batches was used for viscosity model selection and therefore does not represent a fully independent validation batch. Product concentration was predicted with NRMSE values of 5.8% for a closed-loop and 4.3% for an open-loop simulation, although several flow-, oxygen-transfer-, and biomass-related states showed larger deviations. Overall, the results provide a proof of concept for capacitance-integrated hybrid modelling, reducing the number of fermentation batches by 66.7% and decreasing offline sampling requirements from previously 10-15 samples for both, CDW and viscosity per batch to one CDW sample and no viscosity samples.