Dynamic orthogonal projection. A new method to maintain the on-line robustness of multivariate calibrations. Application to NIR-based monitoring of wine fermentations

A new method is developed to maintain the on-line robustness of multivariate calibrations against unknown influence factors, especially in the case of spectroscopy. This method is based on the use of the calibration database and the periodical on-line reference measurements with their corresponding spectra to compute an orthogonal correction. The spectra that would have been obtained in the absence of influence factor occurrence are estimated from the calibration database using the reference measurements and a kernel function. The difference between the estimated and the on-line spectra is used to compute an orthogonal projection correction of the calibration database that is used for recalibration. The originality of this method is its capability to handle different kinds of perturbations (chemical, physical, and environmental) using the original calibration database. This method was applied for on-line monitoring of alcoholic wine fermentation in the presence of temperature variations as the influence factor, where it showed a good performance for temperature variation and batch effects. Also, this method provides an on-line diagnosis tool for the spectral causes of robustness problems thus providing a basis to better understand the process in order to perform adjustments, if necessary, at an earlier stage.