Слайд 1 - Хемометрика в России

Download Report

Transcript Слайд 1 - Хемометрика в России

Sixth Winter Symposium on Chemometrics
Kazan 18 – 22 February 2008
_______________________________________________________
Instrumental measurements of
beer taste attributes using an
electronic tongue
Alisa Rudnitskaya Evgeny Polshin Dmitry Kirsanov
Katrien Beullens Jeroen Lammertyn Bart Nicolai
Freddy Delvaux Andrey Legin
Purpose of the study
•
•
•
•
Brewing and aging of beer are complex processes
Several parameters have to be controlled to ensure reproducible quality of
the finished product
One of the most important – taste and flavour are evaluated by the sensory
panel
Similarly to all sensory analyses of any foodstuffs assessment of beer taste
and flavour is slow, expensive and suffering from irreproducibility of the
assessors.
• The aim of the present study is evaluation of the electronic
tongue sensor system as a screening tool for the beer taste
and flavour attributes
2
Experimental
Samples
• Samples
– 50 Belgian and Dutch beers of different types
• Sensory evaluation
–
–
–
–
–
–
trained sensory panel has estimated 72 attributes in total
beer aroma
taste
mouthfeel
appearance
global quality
3
Experimental
ET measurements
• Sensor array
– 29 potentiometric chemical sensors of different types
– multichannel custom-made voltmeter
• Sample preparation
–
–
–
–
filtering
dilution
thermostatting at 270C;
7-9 replicated measurements
4
Data processing
• Data exploration:
– Principal Component Analysis (PCA)
– Canonical Correlation Analysis (CCA)
• Prediction of sensory attributes:
– Partial Least Square Regression (PLS)
5
Data set
Flavour and taste attributes
• 72 attributes were evaluated:
– 27 aroma attributes
intensity, sour, fruity, alcoholic, …
– 29 taste and flavour attributes
– 9 aftertaste attributes
– 4 mouth feel attributes
– 2 foam attributes
sour, sweet, bitter, hoppy, …
intensity, duration, body, …
astringency, CO2, warming, …
colour, texture
– global quality
• Sets of attributes are overlapping
– Aroma and taste attributes sets include the same parameters
– Taste and aftertaste attributes include the same parameters
6
Data set
Flavour and taste parameters
•
Correlation between beer taste and aroma attributes
– The same attributes related to the taste and aroma or aftertaste were highly
correlated (correlation coef. 0.8-0.9) with exception of attributes fusty and metallic
Considering high correlation of the same aroma and taste attributes and
the fact that ET is measuring liquid only attributes pertaining to the beer
taste were chosen for the further data processing, i.e. 45 in total
•
Correlation was also observed between following groups of attributes
(correlation coef. 0.9-0.7):
• aroma, taste and after taste intensity and mouthfeel, duration and body;
• mouthfeel, CO2, warming and aftertaste intensity, duration and body;
• taste, aftertaste and aroma sour, artificial and fruity;
• taste and aroma alcoholic and mouthfeel and warming;
• taste and aroma ester and solvent;
• taste and aroma sulphury, DMS and rubber
7
Data set
Sensory panel
•
Problem of the sensory panel data set
– Sensory panel consisted of 18 people
– Not all of them were present at each tasting session
– Each sample was tasted by a “sub-panel” of 7-11 assessors
– None of them tasted all samples
•
Data sets used
– 7 tasters that tasted more than half of the samples
–
Average values of each attribute were calculated using scores of those of the 7
tasters that assessed this particular sample, which resulted in 50x45 data matrix.
– Sensory attributes data set was centered and standardized
8
Beer samples discrimination
Sensory panel data
PCA score plot
4
2
1119
0
775
1233
362 489
1138
1139
881
619
622
802
932
533
1171
1170
1168
882
1167
125
486
817
890
363
744
1689
368
251
1688
620
364
365
891
9411140
774 616 602742 1134
1234
892 893
487 615
933
815
-4
PCA loading plot
1183
514
0.3
-6
bitter
0.2
-8
684
0.1
-10
PC2 (17%)
PC2 (17%)
-2
-12
-14
845
-16
-8
-6
-4
-2
0
PC1 (20%)
2
4
0.0
-0.1
6
-0.2
-0.3
-0.4
-0.3
bitter_at
hoppy
solvent
warming_m
ester
spicy
sulphury DMS
alcoholic
vinylguaiacol
astringent
CO2_m
rubber acetaldehyde
yeast
oxidation
global quality
mouthfeel
H2Sautolysis
body_at
metallic
chlorophenol
burned liquorice_at
worty
intensity_at
caramel
liquorice
duration_at
floral
intensity
texture diacetyl fustyfruity_at
colour
fruity
sweet_at
sweet
artificial
artificial_at
sour
sour_at
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
PC1 (20%)
9
Beer samples discrimination
Sensory panel data
1138
1119
PCA loading plot
815
891
775
1140
1167
893
742 1688
622
1168
684
882
1233
892
616 1183
881
744
1170
1134
774 615933620
817
363
487
362364
1139
602
890
486
1171 1689
365
932368
802
533251
619
125
941
1234
845
514
0,3
PC4 (7%)
PC4 (7%)
PCA score plot
5
4
3
2
1
0
-1
-2
-3
-4
-5
-6
-7
-8
-9
-10
-6
489
-5
-4
-3
-2
-1
0
1
2
PC3 (10%)
3
4
5
6
7
sulphury floral
0,2 global quality
DMS
hoppy
sweet_at liquorice
fruity_at ester
alcoholic worty
sweet
caramel
rubber
0,1
acetaldehyde
mouthfeel liquorice_at
colour
fruity bitter_at
body_at
metallic
warming_m
yeast
chlorophenol
0,0
astringent_m
artificial solvent
sour_at
bitter
burned
sour fusty
texture
duration_at
artificial_at
oxidation intensity_at
intensity
-0,1
CO2_m
-0,2
diacetyl
spicy
8
-0,3
vinylguaiacol
-0,4
H2 S
-0,5
-0,3
-0,2
-0,1
autolysis
0,0
0,1
0,2
0,3
0,4
0,5
PC3 (10%)
10
Beer samples discrimination
ET data – comparison with sensory panel
ET data -PCA score plot
5
845
845
4
774
774
PC2 (24%)
2
1
0
-1
684
684
Sensory panel - PCA score plot
4
2
-2
-3
-4
-6
-4
-2
0
2
4
PC1 (34%)
775
1233
362 489
1138
1139
881
619
622
802
932
533
1170
1171
1168
882
1167
486125
817
890
744
363
1689
368
251
1688
620
364
365
891
9411140
774 616 602742 1134
1234
892 893
487 615
933
-2
-5
-8
1119
0
251
881
1233
891
1233 1134
881893
9321688 890
11191119
533
619
932
891
619
533 890
1688
489363
1170
1170
363 802
486
1138
1138
486802
PC2 (17%)
3
620
882 622
622
620
368
365
365
1689 364
882
892
616
892
368
364
1140
817
514 1689
742
615
514
362
602
817
1168 1140
941616
933
615
941
775
775
933 362 742
125125
1167
1167
1234
487
1171
487
1234 1171
744
1139
744
815
1183
251
1134
8151139
1183
6
8
10
12
14
1183
514
815
-4
-6
-8
684
-10
-12
-14
845
-16
-8
-6
-4
-2
0
2
6
4
PC1 (20%)
11
Beer samples discrimination
ET data – comparison with sensory panel
ET data -PCA score plot
6
5
881
881
882
882
4
Sensory panel - PCA score plot
2
1134
815
815
1138 1134
616
620
615
363
486
622
622
620 1138
932
487
802
615
251 363
932
1234
487
365
802
619362
368
251 817
6161689
486 933
1233
362
365 602
489 817
1233
933
744
1139
602
845
845
775
619 1140
368
1689
684
775 684
890
1119
1234
364
364744
489
1140
774
533
742 890
514
1688
1139
1119774
533 514 742
1688
1167125
1167
893
1170
941
1168
125
1183
1170
941
892
1183
893
891
892
1171
1171
891
1
0
-1
-2
-3
-4
-8
-7
-6
-5
-4
-3
-2
-1
0
PC3 (16%)
1
2
3
4
5
PC4 (7%)
PC4 (10%)
3
6
5
4
3
2
1
0
-1
-2
-3
-4
-5
-6
-7
-8
-9
-10
-6
1138
1119
815
891
775
1167
8931140 742
622
1168
1688 616 1183
684
882
1233
892
881
744
1170
1134
774 615933620
817
363
487
362364
1139
602
890
486
1171 1689
365
932368
251
533802
619
125
941
1234
845
514
489
-5
-4
-3
-2
-1
0
1
2
3
4
5
6
7
PC3 (10%)
12
8
Beer samples discrimination
ET data
ET data -PCA score plot
5
845
845
3
620
882 622
622
620
368
365
365
1689 364
882
892
892
368 616
364
1140
817
514 1689
742
615
514
362
602
817
1168 1140
616
941
933
615
941
775
775
933 362 742
125125
1167
1167
1234
487
1171
487
1234 1171
744
1139
744
815
1183
251
1134
8151139
1183
774
774
PC2 (24%)
2
1
0
-1
684
684
ET data - loading plot
0.5
0.4
-3
-4
-6
-4
-2
0
2
4
PC1 (34%)
C12C3
0.2
-5
-8
E2 C14
E1
0.3
251
881
1233
891
1233 1134
881893
9321688 890
11191119
533 890
619
932
891
619
533
1688
489363
1170
1170
363 802
486
1138
1138
486802
-2
6
8
10
12
PC2 (24%)
4
A9
C8
C4
A10
0.1
C11
G10
0.0
14
C13
A5 A6
pH
G4 A4
G5
A8
A3
-0.1
C2C6
-0.2
C10
-0.3
A12
-0.4
-0.05
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
PC1 (34%)
13
Beer samples discrimination
ET data
ET data -PCA score plot
6
5
881
881
882
882
4
2
1134
815
815
1
0
-1
-2
-3
-4
-8
-7
ET data - PCA loading plot
1138 1134
616
620
615
363
486
622
622
620 1138
932
487
802
615
251 363
932
1234
487
365
802
619362
368
251 817
6161689
486 933
1233
362
365 602
489 817
1233
933
744
1139
602
845
845
775
619 1140
368
1689
684
775
890
684
1119
364
364
489 1234
744
1140 533
774
742 890
514
1688
1139
1119774
533 514 742
1688
1167125
1167
893
1170
941
1168
125
1183
1170
941
892
1183
893
891
892
1171
1171
891
-6
-5
-4
-3
-2
-1
0
PC3 (16%)
1
2
3
4
5
0.5
0.4
0.3
C8
C12
C13
C2
G4
G5
0.2
PC2 (10%)
PC4 (10%)
3
C6
0.1
0.0
G10
C10
C11
E1 A12
C4
6
-0.1
A6
-0.2
C3
E2
pH
A5
C14
A9
A3
A4
A10
A8
-0.3
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
PC1 (16%)
14
Comparison of ET and sensory panel data sets
using CCA
•
•
Four significant canonical roots were extracted
Correlations between first four pairs of canonical variables were: 0,96, 0,91, 0,79 and
0,77
Similarity maps
Sensory panel
ET
3
4
684
CV2
0
-1
1
0
684
845
-1
-2
-2
-3
893
514
-3
1119
775
1138
1233
362
489
622
941
251
1139
619 365
932
487
1140
1689
368
774
802
817
364
881
363
615
882
1167
1171
125
1168
933
486
1170 533
891 744 1688 1234
892
1134
742
890
602
616
893
620
815
2
CV2
1
845
3
1119
882
489
1139 619251
1233 774
775
622
365
802 363
1134
881
1140
487
941
362
533
364
932 368
1689817
1138
1171
815 1688 933
744 615
11671170
620
616
1168
602
125 892 890
742
1234
486
891
2
-4
1183
1183
514
-5
-4
-12
-10
-8
-6
-4
CV 1
-2
0
2
4
6
-9
-8
-7
-6
-5
-4
-3
-2
-1
0
1
2
CV1
15
3
Prediction of the attributes using ET
Correlation
Slope
Offset
RMSEP
Sd (assessors)
1.7
-0.05
0.08
1.21
0.26
0.27
0.16
0.25
0.14
0.28
0.24
0.29
0.42
0.53
0.36
0.35
0.53
0.33
0.60
0.14
0.68
0.85
1.17
0.50
0.19
0.38
0.19
2.1
2.1
0.5
0.39
0.41
0.64
0.65
0.75
0.99
1.4
2.1
0.41
0.12
0.27
0.16
0.05
1.65
0.23
0.39
0.47
0.22
0.35
0.24
0.09
0.28
0.82
0.82
1.13
0.06
0.40
0.46
0.05
0.75
1.2
0.56
0.74
0.46
0.63
0.74
Taste and aroma
intensity
sour
sweet
bitter
fruity
caramel
artificial
burned
sulphury (SO2)
0.85
0.74
0.96
0.84
0.97
0.94
0.93
0.79
0.8
0.7
0.7
0.8
0.6
0.7
0.6
0.7
0.55
0.41
mouthfeel
mouthfeel
CO2
warming
0.77
0.72
0.82
0.53
0.48
0.66
aftertaste
intensity
duration
afterbitter
sour
sweet
fruity
artificial
body
0.9
0.79
0.79
0.88
0.64
0.86
0.87
0.89
0.71
0.5
0.58
1.05
0.51
0.49
1.1
0.6
foam
colour
texture
0.81
0.76
0.48
0.76
16
Conclusions
• Electronic Tongue multisensor system seems to be very promising
tool for instrumental beer taste screening
• Determination of taste attributes with ET in some cases allows to
achieve lower error values than sensory panel
• Sensory panel data set handling and methods of ET data processing
are still to be improved
17
Acknowledgments
• RFBR project 05-03-34824-МФ_а
• Centre for Malting and Brewing Sciences, Leuven, Belgium
• InBev Brewery Company, Belgium
• Sensor Systems LLC, St Petersburg, Russia
18
THANK YOU FOR
YOUR ATTENTION !
19
Data set
Flavour and taste attributes
•
72 attributes pertaining to the beer aroma and taste were evaluated:
– 27 aroma attributes (intensity, sour, fruity, alcoholic, hoppy, floral, spicy, caramel,
liquorice, worty, artificial, ester, solvent, burned, yeast, autolysis, sulphury, H2S,
DMS, diacetyle, fusty, oxidation, metallic, chlorophenol, vinylguaiacol, rubber,
acetaldehyde)
– 29 taste and flavour attributes (intensity, sour, sweet, bitter, fruity, alcoholic, hoppy,
floral, spicy, caramel, liquorice, worty, artificial, ester, solvent, burned, yeast,
autolysis, sulphury, H2S, DMS, diacetyle, fusty, oxidation, metallic, chlorophenol,
vinylguaiacol, rubber, acetaldehyde)
– 9 aftertaste attributes (intensity, duration, bitter, sour, sweet, fruity, liquorice,
artificial, body)
– 4 mouth feel attributes (mouthfeel, astringency, CO2, warming)
– 2 foam attributes (colour, texture)
– global quality
•
Sets of attributes are overlapping
– Aroma and taste attributes sets include the same parameters with exception of
sweet and bitter that are only taste characteristics
– Taste and aftertaste attributes include the same parameters (e.g. intensity, bitter,
sour, sweet, fruity, liquorice, artificial)
20