Use of a rugged mid-infrared spectrometer for in situ process analysis of liquids

The widespread application of mid-infrared (MIR) spectroscopy for process monitoring is currently limited by the poor transmission of MIR light through fibre optics. In this work, the performance of a novel and robust MIR spectrometer has been evaluated for practical deployment in a pilot plant or production environment. The spectrometer utilises a Sagnac interferometer design containing no moving parts and is directly attached to an attenuated total reflectance probe, eliminating the need for fibre optics. The quantitative performance of the spectrometer for the in situ analysis of ternary solvent mixtures was assessed. The predictions obtained by partial least squares were accurate (root mean square error of prediction of < 1% w/w) and comparable to those of a benchmark Michelson-based spectrometer with a fibre-coupled probe, which is more amenable to process development in a laboratory or pilot plant. Calibration transfer between the two spectrometers was performed using the spectral space transformation method to mimic the scenario of the scale-up of a process from the laboratory to pilot scale or from a pilot plant to production scale, where the two different MIR instruments might be deployed. The ability to perform in situ reaction monitoring with the robust Sagnac-based spectrometer was then demonstrated. Spectra acquired during an esterification reaction were resolved using multivariate curve resolution, to produce concentration profiles of each component. These results demonstrate the suitability of this rugged spectrometer for quantitative in situ monitoring of liquid processes, opening up new opportunities for process monitoring in the MIR region. • A rugged mid-infrared spectrometer was evaluated for in situ analysis of liquids • Quantitative analysis of ternary solvent mixtures was demonstrated • Successfully transferred calibration models between mid-infrared instruments • Reaction concentration profiles were obtained using multivariate curve resolution