A chemometric approach to minimise the diffusive effect of the biomass when using near infrared spectroscopic measurements for the monitoring of bioprocesses

International audienceThis article focuses the problem of the light absorption and diffusion of the bacterial concentration on the calibration of near infrared (NIR) spectroscopic methods for the monitoring of bioprocesses. The biomass or bacterial concentration produces multiplicative and additive effects on the NIR spectral measurements. The process monitored is a batch anaerobic production of bio-hydrogen from molasses in a small-scale reactor. This monitoring consisted on the prediction of the main compounds of the process i.e. the concentrations of acetate, butyrate, sucrose and biomass. The system used was a near-infrared spectrometer scanning the range 1000-2500 nm with a 3 nm resolution, connected to the bioreactor by an in-situ optical fibre device. The work shows that the measurements were heavily influenced by the presence of biomass in the liquid medium. Dedicated pre-treatments like second derivative or Standard Normal Deviate (SNV) and the selection of wavelengths were used in order to improve the prediction accuracy of the compound concentrations. The Partial Least Square (PLS) regression associated with SNV pre-treatment of the transmittance values gave the best results leading to a prediction error of 0.56, 0.12 and 2.5 g.L-1 for butyrate, acetate and sucrose concentrations, respectively for compound concentrations ranges comprised between 0 and 6.65, 1.7 and 15.7 g.L-1, respectively. The biomass concentration had a prediction error of 75 NTU (Nephelometric Turbidity Unit) without pre-treatment for a biomass range comprised between 0 and 680 NTU