NIR Technology Australia

Application Note 15: THE USE OF A NEAR INFRARED TRANSMISSION ANALYSER FOR GRAPE AND WINE QUALITY TESTING

Phase 1

Preliminary Investigations

Brad Swarbrick

Applications Chemist

April 2002

TABLE OF CONTENTS

1. Introduction 1

1.1 Background 1

1.2 Aims 1

1.3 Instrumentation 2

1.4 Sampling Procedures 3

1.5 Spectral Results 4

2. Calibration 5

2.1 Red Juice 5

2.1.1 pH 5

2.1.2 Total Acidity (TA) 5

2.1.3 Baume 6

2.2 White Juice 6

2.2.1 pH 6

2.2.2 TA 6

2.2.3 Baume 7

2.3 White Wine 7

2.3.1 Alcohol 7

2.3.2 pH 7

2.3.3 TA 7

2.3.4 Sugar 8

2.4 Red Wine 8

3. New Hardware Considerations 9

3.1 Instrumental 9

3.2 Sampling 10

4. Conclusion 11

LIST OF TABLES

Table 1: Calibration statistics for pH determination in red juice 5

Table 2: Calibration statistics for TA determination in red juice 5

Table 3: Calibration statistics for Baume determination in red juice 6

Table 4: Complete summary for the statistics of white juice parameters 6

Table 5: Summary of the calibration statistics for white wine parameters 7

Table 6: Summary of calibration statistics for red wine analysis 9

Table 7: Properties of detection systems currently used and proposed 10

LIST OF FIGURES

Figure 1: Performance of the NIT-38 Alcohol Analyser 2

Figure 2: Spectral data for wine and juice samples 4

Figure 3: Spectral data of red wine samples 8

ABBREVIATIONS

NIRTech NIR Technology Australia

AWRI Australian Wine Research Institute

TSS Total Soluble Solids

TA Total Acidity

Abs Absorbance units

EM Electromagnetic

PLS Partial Least Squares

SEC Standard Error of Calibration

SEP Standard Error of Prediction

FOP Fibre Optic Probe

NIR Near Infrared

1. Introduction

1.1 Background

The use of spectroscopy in the wine industry is well established, with many routine measurements for properties such as phenolics and colour being common applications for UV-Visible instruments. NIR on the other hand has recently gained much attention based on the versatility and ease of calibration of the method.

Based on previous work in the fruit industry, properties such as total soluble solids (TSS), pH, organic acids, moisture etc have been measured using NIR techniques and in a more recent development, NIR Technology Australia (NIRTech) in conjunction with the Australian Wine Research Institute (AWRI) have developed a portable fibre optic probe analyser, primarily for the purpose of measuring red grape colour.

The instrument operates in reflectance mode and requires a homogenisation step, which in turn requires an expensive high-speed blender. The precision and accuracy of results obtained depend on the uniformity of the sample preparation, the contact of the probe with the sample and the complete removal of air bubbles. Comments from industry suggest that this method is highly user dependent and hence has prompted NIRTech to develop a transmission based system for the measurement of grape quality.

This report outlines the preliminary studies performed on a 512-element silicon diode array NIR spectrometer, operating in transmission mode for a series of raw and finished wine products. This research was carried out in conjunction with Cabonne Vintners with the following aims:

1.2 Aims

1) To determine if an NIR spectrometer, operating in the wavelength region 500-1050nm can be used to determine a range of grape quality parameters based on raw juice sample presentation.

2) To determine if the instrument can determine the alcohol content, pH, TA and sugar of both red and white wine samples.

3) To determine whether colour characteristics of red wine can be measured.

4) To make recommendations on hardware design and sampling techniques in order to develop a universal grape and wine analyser.

1.3 Instrumentation

In a previous study, the feasibility of a transmission analyser for the measurement of grape quality parameters was investigated, and from the results obtained, it was suggested that although there were good correlations between the NIR spectra of the grape samples and the quality parameters, the wavelength range of the instrument was limited and that a fibre optic based instrument could provide the most convenient method of sampling.

The instrument used for this study was a 38-element silicon diode array spectrometer (NIT-38, NIRTech), scanning the wavelength region 720-1100nm at a resolution of 10nm. Such wavelength coverage was not optimal for measuring properties such as colour and total phenolics, as these are highly correlated to the visible portion of the EM spectrum. The instrument developed for the measurement of grape quality parameters, (Fibre Optic Probe (FOP) NIRTech), hence incorporated a spectrograph covering the wavelength region 350-1050nm. A 512-element silicon diode array detector was used for this purpose and the absorption spectra obtained were the result of averaging over 4-elements of the detector giving effectively a 128-element detector scanning at a resolution of 5.47nm.

The instrument provided to Cabonne Vintners for this study was a Near Infrared Transmission, (NIT-512) Analyser (NIRTech), covering the same wavelength region as the FOP analyser. The NIT-38 instrument, mentioned previously, is the current instrument used for the alcohol analyser. The predictive ability of this instrument is shown in figure 1.

Figure 1: Performance of the NIT-38 Alcohol Analyser.

1.4 Sampling Procedures

The sample housing for the NIT-512 analyser is based around a generic design for the range of whole grain analysers produced by NIRTech and therefore is not the optimal set-up for the analysis of grape products.

Starna Cells, Inc provide a range of cuvettes suitable for the analysis of liquid samples at pathlengths of 5, 20 and 30mm. The current alcohol analyser uses a 30mm pathlength cell for analysis, and in the region 720-1100nm, provides excellent sensitivity for –OH absprption. For the analysis of highly coloured samples, such as red wine, the molar absorptivity increases by several orders of magnitude and according to the Beer-Lambert law:

Where: A = Absorbance

e = Molar Absorptivity

b = Pathlength

c = Concentration

From the above equation, Absorbance is directly proportional to Molar Absorptivity, Pathlength and Concentration. In the NIR portion of the EM spectrum, molar absorptivity is low enough to implement pathlengths as large as 30mm for liquid samples and still be in the absorbance range 0-2abs units. When measuring samples in the visible portion of the spectrum all three variables of the Beer-Lambert expression contribute to absorbance.

The sensitivity of measurement depends on the dynamic range of the detector. For the NIT-512 analyser, a study was performed to determine the limits of detection of this system using neutral density filters and it was found that the maximum absorbance reading for this instrument is 2.5abs units. This corresponds to about 0.5% light transmission.

For white grape and wine samples, pathlength considerations are not so stringent as there is no strong colour absorptions in the visible region of the spectrum measured by this instrument. The major sampling strategies to be considered in this study for red wine samples are:

a) Pathlength Considerations.

b) Dilution.

The sampling procedures implemented in this study are in many ways primitive and are prone to high levels of variability, but for the purposes of a preliminary study, should form the basis of any continued study. Processes such as filtration and centrifugation of samples with low clarity were compared for spectral reproducibility and will form the basis for final method development.

1.5 Spectral Results

Figure 2 shows the spectral data obtained from selected samples, chosen from the calibration set, representing both juice and wine.

Figure 2: Spectral data for wine and juice samples.

The data obtained in figure 2, were all collected in a 20mm pathlength cuvette. It is obvious that red wine will require either reduced pathlength or dilution considerations if the instrumental configuration used here is to be effective, especially for colour.

2. Calibration

2.1 Red Juice

The parameters to be measured for red juice are as follows, pH, Total Acidity (TA) and Baume. The calibration set consisted of approximately 75 individual samples scanned three (3) times on the NIT-512 analyser. The sample presentation usually was a cloudy liquid, which was either centrifuged or filtered before spectral analysis. Partial Least Squares (PLS) regression was performed on the spectral data and reference laboratory data for all parameters measured. The calibration statistics are presented as follows.

2.1.1 pH

The calibration for pH was developed using absorbance spectra with no data pre-processing. Table 1 shows the calibration statistics for the pH calibration.

					Calibration		Prediction
Red Juice	n	Outliers	Range	PC’s	R	SEC	R	SEP
	231	35	2.8-3.3	8	0.8	0.04	0.85	0.05

Table 1: Calibration Statistics for pH determination in Red Juice.

Inspection of the spectral data obtained for these samples showed that most of the outliers occurred because of incorrect collection, i.e. the data were just noise. There are two points to note about the above data. Firstly, although the correlation for both calibration and prediction is low (R=0.8-0.85) the standard errors are extremely good. Secondly, the R-values may be highly effected by two variables, the laboratory data is limited and more importantly, the sampling procedure was not optimised.

2.1.2 Total Acidity (TA)

Table 2 shows the calibration statistics obtained for TA in red juice.

					Calibration		Prediction
Red Juice	n	Outliers	Range	PC’s	R	SEC	R	SEP
	231	31	6.0-18.0	9	0.82	0.93	0.92	0.98

Table 2: Calibration statistics for the determination of TA in Red Juice.

The range of laboratory values for TA is much broader, and as is the case for pH, there is a significant degree of linearity, but the low R-values, once again, may be indicating the effect of non-reproducible sampling.

2.1.3 Baume

Table 3 provides the calibration statistics for the determination of Baume in red juice.

					Calibration		Prediction
Red Juice	n	Outliers	Range	PC’s	R	SEC	R	SEP
	231	39	7.0-12.0	8	0.90	0.23	0.96	0.24

Table 3: Calibration statistics for the determination of Baume in Red Juice.

The calibration for Baume seems to show the best predictive ability of all the parameters

measured for red juice. This is most likely due to the high correlation of sugar to the –OH absorption around 960nm. Better sampling should increase the predictive ability for this parameter significantly.

2.2 White Juice

The sample parameters measured for red juice were also measured for white juice. Table 4 provides a summary for the calibration statistics for all parameters measured. As for red juice, calibrations were developed on absorbance spectra, with no data pre-processing, using PLS regression with the same filtration or clarification step used. Approximately 80 different samples were used for the purpose of calibration.

					Calibration		Prediction
Red Juice	n	Outliers	Range	PC’s	R	SEC	R	SEP
	231	39	7.0-12.0	8	0.90	0.23	0.96	0.24

Table 4: Complete summary for the calibration statistics of White Juice parameters.

White Juice					Calibration		Prediction
	n	Outliers	Range	PC’s	R	SEC	R	SEP
pH	240	20	2.7-3.8	8	0.86	0.07	0.94	0.07
TA	240	35	5.0-18.0	9	0.78	1.41	0.89	1.43
Baume	240	34	8.0-13.0	11	0.97	0.15	0.98	0.19

2.2.1pH

The predictive ability for pH is of similar quality to the red juice predictions. An SEP of 0.07 is comparable to that expected from standard laboratory measurements.

2.2.2 TA

The results obtained for TA have a much higher error associated with them compared to red juice, which is somewhat surprising as the detector operates at a higher sensitivity for less coloured samples. The calibration line showed some apparent curvature, but at this stage, it is impossible to differentiate this from actual scatter. Overall, it seems possible to determine TA based on the data obtained and more work is required to optimise this.

2.2.3 Baume

As for the case with red juice, the baume for white juice is predicted to a high level of accuracy, possibly for the same reason presented for red juice.

2.3 White Wine

The parameters of interest to be measured in white wine are alcohol, pH, TA and sugar. White wine samples are optimally measured in a 30mm pathlength cell to obtain the best sensitivity from the alcohol absorbance, but for this study, a 20mm pathlength cuvette was used. PLS regression was applied to the data as usual and table 5 presents the calibration statistics for all parameters.

White Wine					Calibration		Prediction
	n	Outliers	Range	PC’s	R	SEC	R	SEP
Alcohol	138	18	12.0-16.0	11	0.96	0.13	0.99	0.13
pH	138	19	3.0-3.5	10	0.95	0.02	0.98	0.02
TA	138	21	6.0-9.0	8	0.85	0.18	0.93	0.18
Sugar	138	48	0.0-25.0	7	0.94	0.93	0.98	0.95

Table 5: Summary of calibration statistics for white wine parameters.

2.3.1 Alcohol

The results presented in table 5 for white wine for alcohol content are consistent with the accuracy expected for the NIT-38 alcohol analyser.

2.3.2 pH

The pH results obtained are also to a high level of accuracy. The sample set is limited in range of laboratory data but both calibration and prediction have a high correlation to this reference data.

2.3.3 TA

&nb