The evaluation of Fourier transform infrared (FT-IR) spectroscopy for quantitative and qualitative monitoring of alcoholic wine fermentation

[摘要] ENGLISH ABSTRACT: Fermentation is a complex process in which raw materials are transformed into high-valueproducts, in this case, grape juice into wine. In this modern and economically competitivesociety, it is increasingly important to consistently produce wine to definable specifications andstyles. Process management throughout the production stage is therefore crucial to achieveeffective control over the process and consistent wine quality. Problematic wine fermentationsdirectly impact on cellar productivity and the quality of wine. Anticipating stuck or sluggishfermentations, or simply being able to foresee the progress of a given fermentation, would beextremely useful for an enologist or winemaker, who could then take suitable corrective stepswhere necessary, and ensure that vinifications conclude successfully. Conventional methods offermentation monitoring are time consuming, sometimes unreliable, and the information limitedto a few parameters only. The current effectiveness of fermentation monitoring in industrial wineproduction can be much improved. Winemakers currently lack the tools to identify early signs ofundesirable fermentation behaviour and to take preventive actions.This study investigated the application of Fourier transform mid infrared (FT-IR)spectroscopy in transmission mode, for the quantitative and qualitative monitoring of alcoholicfermentation during industrial wine production. The major research objectives were firstly toestablish a portfolio of quantitative calibration models suitable for quantification of the majorquality determining parameters in fermenting must. The second major research objectivefocused on a pilot study aimed at exploring the use of off-line batch multivariate statisticalprocess control (MSPC) charts for actively fermenting must. This approach used FT-IR spectraonly, for the purpose of qualitative monitoring of alcoholic fermentation in industrial wineproduction. Towards these objectives, a total of 284 industrial-scale, individual, activelyfermenting tanks of the seven major white cultivars and blends, and nine major red cultivars, ofNamaqua Wines, Vredendal, South Africa, were sampled and analysed with FT-IRspectroscopy and appropriate reference methods during vintages 2007 to 2009.For the quantitative strategy, partial least squares regression (PLS1) calibration models fordetermination of the classic wine parameters ethanol, pH, volatile acidity (VA), titratable acidity(TA) and the total content of glucose plus fructose, were redeveloped to provide a better fit tolocal South African samples. New PLS1 models were developed for the must componentsglucose, fructose and yeast assimilable nitrogen (YAN), all of which are frequently implicated inproblem fermentations. The regression statistics, that included the standard error of prediction(SEP), coefficient of determination (R2) and bias, were used to evaluate the performance of theredeveloped calibration models on local South African samples. Ethanol (SEP = 0.15 %v/v, R2 =0.999, bias = 0.04 %v/v) showed very good prediction and with a residual predictive deviation(RPD) of 30, rendered an excellent model for quantitative purposes in fermenting must. Themodels for pH (SEP = 0.04, R2 = 0.923, bias = -0.01) and VA (SEP = 0.07 g/L, R2 = 0.894, bias= -0.01 g/L) with RPD values of 4 and 3 respectively, showed that the models were suitable forscreening purposes. The calibration model for TA (SEP = 0.35 g/L, R2 = 0.797, bias = -0.004g/L) with a RPD of 2, proved unsatisfactory for quantification purposes, but reasonable forscreening purposes. The calibration model for the total content of glucose plus fructose (SEP =0.6.19 g/L, R2 = 0.993, bias = 0.02 g/L) with a RPD of 13, showed very good prediction and canbe used to quantify total glucose plus fructose content in fermenting must. The newly developedcalibration models for glucose (SEP = 4.88 g/L, R2 = 0.985, bias = -0.31 g/L) and fructose (SEP= 4.14 g/L, R2 = 0.989, bias = 0.64 g/L) with RPD values of 8 and 10 respectively, also proved fitfor quantification of these important parameters. The new calibration models of ethanol, total glucose plus fructose; and glucose and fructose individually, showed an excellent relation tolocal South African samples and can be easily implemented by the wider wine industry.Two calibration models were developed to determine YAN in fermenting must by usingdifferent reference methods, namely the enzyme-linked spectrophotometric assay and Formoltitration method, respectively. The results showed that enzyme-linked assays provided a goodquantitative model for white fermenting must (SEP = 14.10 mg/L, R2 = 0.909, bias = -2.55 mg/L,RPD = 6), but the regression statistics for predicting YAN in red fermenting must, were lesssatisfactory (data not shown). The Formol titration method could be used successfully in bothred- and white fermenting must (SEP = 16.37 mg/L, R2 = 0.912, bias = -1.01 mg/L, RPD = 4). Aminor, but very important finding was made with respect to the storage of must samples thatwere taken from tanks, but that could not immediately be analysed with FT-IR spectroscopy orreference values. Principal component analysis (PCA) of frozen samples showed that mustsamples could be stored frozen for up to 3 months and still be used to expand the calibrationsample sets when needed. Therefore, samples can be kept frozen to a later stage if immediateanalyses are not possible.For the purpose of the pilot study that focused on the use of FT-IR spectroscopy forqualitative off-line monitoring of alcoholic fermentation, a total of 21 industrial-scale fermentationtanks were monitored at 8- or 12-hourly intervals, from the onset of fermentation to completeconsumption of the grape sugars. This part of the work excluded quantitative data, and onlyused FT-IR spectra. MSPC charts were constructed on the PLS scores of all the FT-IR spectrataken at the various time intervals of the different batches, using time as the y-variable. Theprimary aim of this research objective was to evaluate if the PLS batch models could be used todiscriminate between normal and problem alcoholic fermentations. The models that wereconstructed clearly showed the variations in patterns over time, between red- and white winealcoholic fermentations. One Colombar tank that was fermented at very low temperature inorder to achieve a specific wine style, was characterised by a fermentation pattern that clearlydiffered form the rest of the Colombar fermentations. This atypical fermentation was identifiedby the batch models constructed in this study. PLS batch models over all the Colombarfermentations clearly identified the normal and problem fermentations.The results obtained in this study showed that FT-IR spectroscopy showed great potentialfor effective quantitative and qualitative monitoring of alcoholic fermentation during industrialwine production. The work done in this project resulted in the development of a portfolio ofcalibration models for the most important quality determining parameters in fermenting must.The quantitative models were subjected to extensive independent test set validation, and havesubsequently been implemented for industrial use at Namaqua Wines. Multivariate batchmonitoring models were established that show good discriminatory power to detect problemfermentations. This is a very useful diagnostic tool that can be further developed by monitoringmore normal and problem fermentations. Future work in this regard, will focus on furtheroptimisation and expansion of the quantitative and qualitative calibration models andimplementation of these in the respective wineries of Namaqua Wines.

[发布机构] Stellenbosch University