Wine fermentation process control by PCA-based forecasting of infrared spectra

Early detection of fermentation deviations is essential, since delayed recognition of process disturbances can lead to quality losses, sluggish or stuck fermentations, and reduced production efficiency. This work extends Evolving Window Dissimilarity Index (EWDI)-based monitoring towards anticipatory process control by proposing a practical strategy to forecast EWDI from at-line ATR-FTIR spectra during wine fermentations. The workflow combines a moving-window PCA representation of the fermentation state with short-horizon extrapolation of principal component scores, reconstruction of future spectra, and EWDI calculation on the forecasted trajectories. Forecast uncertainty is estimated from the score prediction model and propagated to EWDI by Monte Carlo resampling. A probability of being in control (P C ) is also calculated from the forecasted EWDI distribution to express how likely the process is to remain within the control region. Validation on under-control and intentionally deviated fermentations showed that EWDI can be forecast reliably in the short term, particularly in the early fermentation stage, enabling earlier recognition of emerging deviations. Longer horizons may diverge from the measured trajectory during phases of rapid kinetic change, where abrupt transitions cannot be anticipated until new measurements enter the calibration window. P C provided a useful complement by distinguishing between predictions clearly outside the control region and predictions that, although centred within control limits, still showed enough uncertainty to indicate increased process risk. Together, EWDI forecasting and P C provide a short-horizon, risk-oriented tool for fermentation monitoring.