Development of a novel glucose online analysis approach in fermentation: Combining PLS and SVR for Raman spectroscopy...

Fermentation processes are critical in industries such as pharmaceuticals or food production, where glucose plays a key role as a primary carbon source for microorganisms. Accurate and real-time glucose monitoring is essential for optimizing fermentation conditions, but traditional methods often provide delayed results and lack the robustness required for industrial applications. To overcome these challenges, the present work applies a hybrid modelling framework not previously used for glucose monitoring in fermentation combining Partial Least Squares (PLS) and Support Vector Machine (SVM) regressions (PLS-SVR) for quasi-real-time glucose monitoring, on a 5 to 50 g L^-1 concentration range, using Raman spectroscopy. The Raman acquisition parameters were optimized using a response surface method. Missing high concentration references were supplemented by glucose estimates derived from Oxygen Uptake Rate (OUR) data resulting in an original bootstrapping strategy for this type of Raman based fermentation modelling. Finally, the model was subjected to twelve independent fermentation runs performed over six months with deliberately varied medium compositions and operating conditions. Together, these factors constitute an extensive validation, covering multiple full-scale fermentations, long term temporal drift, and broad process-to-process variability. The results demonstrate that the PLS-SVR model significantly outperforms traditional PLS models, with a Root Mean Square Error of Prediction (RMSEP) of 1.7 g L^-1 over the 6-month validation, highlighting its accuracy and reliability. This work provides a practical and reliable solution for glucose monitoring in fermentation, paving the way for a new benchmark in bioprocesses monitoring.