Comparison of Individual and Integrated Inline Raman, Near-Infrared, and Mid-Infrared Spectroscopic Models to Predict the Viscosity of Micellar Liquids

Abstract

In many industries, viscosity is an important quality parameter which significantly affects consumer satisfaction and process efficiency. In the personal care industry, this applies to products such as shampoo and shower gels whose complex structures are built up of micellar liquids. Measuring viscosity offline is well established using benchtop rheometers and viscometers. The difficulty lies in measuring this property directly in the process via on or inline technologies. Therefore, the aim of this work is to investigate whether proxy measurements using inline vibrational spectroscopy, e.g., near-infrared (NIR), mid-infrared (MIR), and Raman, can be used to predict the viscosity of micellar liquids. As optical techniques, they are nondestructive and easily implementable process analytical tools where each type of spectroscopy detects different molecular functionalities. Inline fiber optic coupled probes were employed; a transmission probe for NIR measurements, an attenuated total reflectance probe for MIR and a backscattering probe for Raman. Models were developed using forward interval partial least squares variable selection and log viscosity was used. For each technique, combinations of pre-processing techniques were trialed including detrending, Whittaker filters, standard normal variate, and multiple scatter correction. The results indicate that all three techniques could be applied individually to predict the viscosity of micellar liquids all showing comparable errors of prediction: NIR: 1.75 Pa s; MIR: 1.73 Pa s; and Raman: 1.57 Pa s. The Raman model showed the highest relative prediction deviation (RPD) value of 5.07, with the NIR and MIR models showing slightly lower values of 4.57 and 4.61, respectively. Data fusion was also explored to determine whether employing information from more than one data set improved the model quality. Trials involved weighting data sets based on their signal-to-noise ratio and weighting based on transmission curves (infrared data sets only). The signal-to-noise weighted NIR–MIR–Raman model showed the best performance compared with both combined and individual models with a root mean square error of cross-validation of 0.75 Pa s and an RPD of 10.62. This comparative study provides a good initial assessment of the three prospective process analytical technologies for the measurement of micellar liquid viscosity but also provides a good basis for general measurements of inline viscosity using commercially available process analytical technology. With these techniques typically being employed for compositional analysis, this work presents their capability in the measurement of viscosity—an important physical parameter, extending the applicability of these spectroscopic techniques.