Monitoring yeast fermentations by nonlinear infrared technology and chemometrics-understanding process correlations and indirect predictions.

Fermentation processes are still compromised by a lack of monitoring strategies providing integrated process data online, ensuring process understanding, control, and thus, optimal reactor efficiency. The crucial demand for online monitoring strategies, not only encouraged by the PAT initiative but also motivated by modern paradigms such as circular economy and sustainability, has driven research and industry to provide "next-generation process technology": in other words, technology tailored toward industrial needs. Mid-infrared (MIR) spectroscopy as such is superior to near-infrared (NIR) spectroscopy since it provides significantly enhanced selectivity. However, due to high costs and a lack of instrumental robustness, MIR spectroscopy is outcompeted by NIR when it comes to industrial application. The lack of chemometric expertise, model understanding, and practical guidance might add to the slow acceptance of industrial MIR application. This work demonstrates the use of novel MIR, so-called non-linear infrared (NLIR) technology and the importance of model understanding, exemplarily investigated on a lab-scale yeast fermentation process. The six analytes glucose, ethanol, glycerol, acetate, ammonium, and phosphate were modeled by partial least squares (PLS) based on spectral data, demonstrating the potential of the novel technology facilitating online data acquisition and the necessity of investigating indirect predictions. KEY POINTS: * NLIR spectra were acquired online during a yeast fermentation process * PLS models were constructed for six components based on uncorrelated samples * Glucose, ethanol, ammonium, and phosphates were modeled with errors of less than 15% * Acetate and glycerol were shown to rely on indirect predictions.