L-C.F.B.
15 AUG IM
ACK -
n L E
Scientific Aids to blending: A collaborative study
With I.T.L. Canada
(Progress report describing the recent visit to
I.T.L.
between the 31st May and 14th June 1980)
G. A. Few
File I
Circulation
R. M. Gibb, Esq. I.T.L. Canada
R. S. Wade, Esq -
Dr L. C. F. Black.~~)'~- Millbank
A. L. Heard, Esq.
Dr C. I. Ayres
Dr G. W. Boswall - I.T.L. Canada
Dr R. G. Hook
Dr T. Hirji
CO
BATCo document for Province of British Columbia 10
November 1999


1. INTRODUCTION
In April 1978 Imperial Tobacco Limited (Canada)
contacted the
GR&DC, Southampton and Hillbank requesting
assistance in solving
the following specific problem. This was one of
maintaining smoking
quality whilst modifying their current usings mix
so as to minimise
the usage of grades that are in short supply and
maximise the usage
of grades in over supply. To facilitate this the
Imperial blender,
Mr A. F. Schaffer, visited the GR&DC in May 1978
for discussions
(1). Following these discussions, it became clear
that imperial
felt that grade/ blend matching (2,3) was more
suited to the solution
of their problem than TPD. With this in mind,
Imperial set up a
programme of O/WP GC analyses of single grade
flue-cured tobaccos
from their market for the 1976, 1977 and 1978 crop
years with an
eye to using this data as a means of maintaining
the smoke quality
of their blends.
A year following Mr Schaffer's visit to the GR&DC
Dr R. G.
Hook visited Imperial to put into operation a
computer formulation
designed to be applicable to the Imperial problem.
This consisted
of constraints regarding grades for which supply
exceeded demand
and vice versa so manipulating the usings,
appended to a frame*work
which endeavoured to maintain smoking quality.
This framework was
composed of the O/WP data, and taste and flavour
maintenance was
affected by minimising a distance measure between
that of the O/WP
scan from the target grade to-be matched and that
of the generated
blend.
In the year following Dr Hook's visit his original
work was
supplemented by Judy Fudge, a student on temporary
assignment from
the University of Waterloo, who was at the same
time studying the
application of discriminant analysis to the O/WP
data, and several
written communications were exchanged between
CR&DC and Imperial.
c0
--_j
_Z~:u
ON
BATCo document for Province of BritiSh Columbia 10
November 1999


It soon became clear from these communications
that the O/WP
data from 1977 grades was subject to much larger
variability than
that observed for the 1976 data, and another visit
to Imperial was
suggested with the broad objectives of assessing'
and improving the
O/WP data and projecting the study from a paper
exercise to a
practical experiment in a systematic fashion. With
these objectives
in mind a visit was arranged to Imperial Tobacco
Limited in Montreal
between 31st May and 14th June this year.
2. SPECIFIC OBJECTIVES
Shortly following arrival, from a discussion with
members of
Imperial Tobacco's Leaf blending and R&D staff, it
was apparent
that the wish of the GR&DC to now take the study
from that of being
purely one on paper and turn it into a practical
exercise completely
agreed with the direction in which Imperial were
themselves progressing.
They had, in fact, produced cigarette samples from
three computer
derived blends (which were subjectively assessed
during the stay)
which were matches for GCI from the 1977 crop
year.
It was pointed out early in the discussion that
progress must
follow in a systematic fashion and the following
scheme v.is proposed.
1. Select a target grade to be matched from a
defined number of
grades.
2. Produce the 'best' match for that grade using
the distance
measure.
3. Make cigarettes from the blend recipe and
compare subjectively
with the target grade.
4. If this is an acceptable match go onto
investigate how taste
and flavour varies with distance measure.
X=
0 N
41-
BATCo document for Province of British Columbia 10
November 1999


- 3 -
It was also pointed out that if there is not an
acceptable match
from 3. one might consider mismatching parts of
the scan whilst
improving the match in other parts in a systematic
fashion until a
satisfactory subjective match is obtained.
The meeting concluded with all parties satisfied
with the
above scheme.
3. REVIEW OF WORK ACCOMPLISHED
The first couple of days were spent scrutinising
the O/WP data
from 1976 and 1977 grades ensuring compatibility
between scans from
within crop years and across crop years. I
confirmed that a
number of inconsistencies existed between scans
for grades from the
1977 crop year. These were brought about by
baseline errors and
peak resolution problems and were duly pointed out
to the O/WF
analysts.
Consideration of the time required to correct the
peak areas
in the 1977 data where the inconsistencies lay,
indicated that it
would be out of the question to edit the data and
make significant
progress in the specific objectives, outlined
previously, during
the time alloted to the visit. Hence, it was
decided to identify
and use only those peaks in the distance measure
which were compatible
throughout the analyses of the 1977 grades and
compatible with
analyses from the 1976 crop year. The distance
measure was defined
as the summation of the absolute error of the
blend peak from the
target peak over 46 chosen peaks.
Having scrutinised the 1976 and 1977 data and
decided which
distance measure was to be minimised, Dr Boswall
and myself spent
co
the next few days modifying the computer files to
include the
distance measure and re-running jobs run
previously to verify that
the computer was producing the same results as
before. Having regenerated
BATCo document for Province of British Columbia 10
November 1999


- 4 -
earlier results the computer was asked to minimise
the distance
measure for 4 target grades given below. At the
same time the
distance measure over all 105 peaks was retained
as a guide.
H2 X in 1976 using 12* 1976 grades
H2 X in 1977 using 12* 1977 grades
GC1 in 1976 using 9A 1976 grades
GCl in 1977 using ak 1977 grades.
FSC3, GCl, D2, H2, FS3, GC2, R142X, RGCl, AF3,
AF2, BFL, CL
,I DC2, AFX, CLX, H2X, CL, BF3, H2, AF2
The recipes obtained from these experiments are
given in
Tables I and 2 for H2X and GC1 respectively. Two
features are
readily apparent in the Tables. Firstly, the
distance measures for
1977 matches are larger than those from 1976. This
is because the
O/WP data from the 1977 crop year is more variable
than that from
1976. Secondly, considerable difference is
observed in the recipes
obtained for the same target between the two crop
years!
By the beginning of the second week the hand
blending and hand
making of cigarettes had begun. This had to be
perforined quickly
since only a few smoking sessions by the expert
panel were available
and was a co-operative effort between myself and
the R&D department.
Initially, one of the targets was hand produced
(using tobacco from
machine made cigarettes) and smoked by the panel
and myself against
its machine equivalent, each cigarette having a
low efficiency
filter. Since no significant difference in taste
and flavour was
found, it was decided to use machine manufactured
cigarettes as the
target controls. This saved considerable time.
Next, the 4 blends
given in Tables 1 and 2 were hand blended (again
using
tobaccos from machine made cigarettes) and the
cigarettes made
from each blend by hand. Each blend was then
smoked by the
Cr\
BATCo document for Province of BritiSh Columbia 10
November 1999


- 5 -
expert panel and myself against the taget, again
each cigarette had
a low efficiency filter.
The comments of the panel were very favourable.
Although some
differences in smoke intensity were apparent, the
panel agreed that
for all targets except H2X 1976 the blends showed
similar top and
base notes i.e. similar taste and flavour
characteristics. The
blend for H2X 1976, although having similar base
notes, had an
extra top note which produced the taste
difference.
Having been encouraged by the comments of the
panel, the next
experiment which was planned investigated how
taste and flavour
varied when the distance measure was varied from
'best' to 'worst' for
a target. Since little time now remained it was
difficult to produce
a good experimental design, however choosing
recipes for blends which
had already been made into cigarettes allowed GGI
1977 to be compared
with a selection of blends/single grades ranging
from the largest
value of the distance measure (the 'worst' match)
to the smallest
(the 'best' match). The blend compositions and
distance measures
are given in Table 3. Notice that the 'best' match
obtained in the
previous experiment (from a grade pool which did
not contain
GO) was not used since this had already been shown
to be a close match.
. After producing hand made cigarettes for the
blend with the
lowest distance measure in Table 3, the expert
panel and myself
smoked the cigarettes as paired comparisons i.e.
each blend/single
grade against the target, and again each cigarette
had a low efficiency
filter. The panel agreed that both single grades
were good taste
and flavour matches for the target. However, CBM2
and the blend
with the lowest distance measure differed from the
target in taste
BATCo document for Province of BritiSh Columbia 10
November 1999


- 6 -
with the latter being quite distinct. Hence, the
overall conclusion
from the subjective experiments which had been
performed was that
there does not appear to be a relationship between
taste and flavour
and the distance measures.
Following a presentation of the results obtained
to date to
Messrs R. H. Gibb, S. M. Candlish, R. S. Wade, A.
P. Schaffer and
several other members of the Leaf Department and
R&D on the afternoon
of 12th June, the remaining time was primarily
spent scrutinising
the O/WP data from the 1978 grades for within crop
year and across
crop year compatibility and applying discriminant
analysis to the
O/WF 1977 data. The discriminant analysis grouped
the grades from
that year as to stalk position and ripeness in the
same way that
they would be grouped by the Imperial Leaf
Department. The reason
for this exercise will become clear in the final
section. In
conclusion a final blending experiment was
performed. From Table 1
the blend recipe for 1976 was applied to 1977
grades and vice
versa. Again blending and cigarette naking were
performed quickly
by hand. Unfortunately, time had run out for the
expert panel and
only one member from the panel, a volunteer and
myself assessed the
cigarpttes. Two pairs were smoked: 1 H2X 1976
versus blend of
1976 grades using 1977 recipe.
2 H2X 1977 versus blend of 1977 grades using 1976
recipe, and
the panel found that with the exception of H2X
1977 the other three
cigarettes were similar in taste and flavour
although the intensity
of smoke was quite low from each. H2X 1977
appeared to be a much
strongpr cigarette.
This concluded the experiments performed in the
two weeks.
ON
CC)
BATCo document for Province of BritiSh Columbia 10
November 1999


4. CoNmusioNs
The first conclusion to be reached was that more
subjective
experiments of the type demonstrated during the
visit should be
performed. It was particularly pointed out by the
author and Mr
Candlish that differences may have axisen because
of the use of
hand made cigarettes and it was certa inly advised
that further
experiments with hand made cigarettes be performed
at a much slower
pace. Although machine made samples would
certainly be ideal,
problems with the quantities of tobacco remaining
in stock and the
time factors involved in production would probably
exclude the use
of machine made samples.
In the event that the distance measures explored
so far are
shown to be an inadequate reflection of taste and
flavour, the
question was discussed as to 'Where do we go
next?' My feeling
certainly, is that there is a lot more to be
obtained using O/WP in
grade matching than has yet been explored.
Although other approaches
(involving tobacco chemistry) for controlling
taste and flavour of
blends were discussed, it is evident that Imperial
is enthusiastic
in continuing the grade matching project based on
O/WP data. Depending
on the adequacy of O/WP data to achieve blend
matching it may of
course be necessary to augment it by including
elements of leaf
chemistry and physical properties in the ultimate
blend matching
model.
One point to emerge from the work and from the
discussions is
that the blending grid used by Imperial blenders
whose axes are
stalk position and ripeness (see Figure 1) is
accepted as generally cz:)
C:XD
defining the way in which grades smoke. For
instance, GC1 and CLX
are both cutter/leaf, mature grades. Now, although
the computer
ON
110
BATCo document for Province of British Columbia 10
November 1999


- 8 -
indicated that the O/WP traces were dissimilar,
blenders would
expect the two to smoke similarly, and, in fact,
this was found by
the panel to be the case. It was pointed out that
if the grid can
be used as a loose taste and flavour framework,
then if discriminant
analysis is applied to separate the groups of
grades on the grid
using sound O/WP peaks then perhaps these peaks
may be used as a
distance measure in grade matching. Certainly,
there is no problem
in separating the groups on the grid on the peaks
available.
Since Imperial have already performed a
considerable amount of
discriminant analysis, it seems appealing to link
this into the
grade matching programme. Dr Boswall is quite
happy to move in
that direction on the confirmation that the
present distance measures
are inadequate and has a student with him at
present until August
who would be able to take this up as his project.
Since time is short, and enthusias m high, I feel
this would be
a ripe time to advance this joint project further
by offering as
much guidance and technical assistance as we can,
Acknowledgement
I would like to thank everybody at Imperial
Tobacco Limited for
their co-operation, especially Dr Graeme Boswall,
and for making my stay
so pleasant.
References
1. File note on file 1 from Dr T. Hirji 2.6.78.
2. BAT report RD 1503 Restricted 14.7.77.
3. BAT report RD 1603 Restricted 17.7.78.
CD
110
CO
___J
BATCo document for Province of British Columbia 10
November 1999


0
0 TABLE I
CL
0 TARGET H2X
3
DISTANCE MEASURE
DISTANCE MEASURE GRADES
YEAR OVER 46 PEAKS OVER ALL PEAKS CL JFSC3 GC1 D2
H2 IFS3 'GC2 RH2X RGC1 AF3 AF2 B
Z FL
0
1976 41.6 80.3 11.3 1.7 15.2 58.2 8.2 5.4
0
< 1977 96.6 186.3 23.5 49
7 3
5 13
1 0
7 9
.
.
.
.
.4
0
:r
0
0
F
0
z
0
cr
c0
c0
t 6 2 6 0


0
0
CL
0
T
0
0
(D
0
h
=r
0
0
0
z
0
(D
3
Cr
(D
DISTANCE MEASURE DISTANCE MEASURE
YEAR OVER 46 PEAKS OVER ALL PEAKS
1976 96.2 128.9
1977 108.4* 194.0
TABLE 2
TARGET GC1
GRADES
DC21 AFX JCLX H2X CL I BF3* I H2 AF2
54.5 2.5 9.4 33.6
12.2 35.0 152.8 -
unavailable for 1976 at the time of hand blending
and hence was deleted.from the recipe.
Z 6 6 0


0
0
CL
0
0
-h
0
0
(D
0
h
0
0
z
0
CD
a
Cr
(D
TABLE 3
TARGET GCI 1977
DISTANCE MEASURE DISTANCE MEASURE GRADES (Z)
CODE OVER 46 PEAKS OVER ALL PEAKS
H2 X H2 DC2 CLX CL GC3 JAFX AF2 BF3
481.7 752.6 100
386.8 623.0 - - 100 -
CBM2 202.2 63.7 31.7 4.6
10.11 12.7
105.7 192.9 49.4 27.7
Not recorded.
2 16 S 6 0 1


w
>
I
0
0
CL
0
0
0
03
=r
0
0
E-
Q
z
0
Cr
to
to
jrl6LAac I. Classification of Flue-Cured Tobacco
Grades
Degree of M4turity
2 3 4 5
USA Highest Mature Medium Mature Trashy to
Grades Maturity JJnblemlshed Unblemished Coarse
1. Lug 13 DX 14 D 15 X2
02 X DZ F52
F53
2. Cutter 22 RH2X 23 RX 24 H 25 DC2
t (") X H2 FSr.2
FSC"3
3. Cutter/Leaf 32 RGCl 33 CLX 34 CL 35 GC2
Gb GC3
_j 4. Leaf 41 MlW 42 RAF3 43 AFX 44 AF
AF3 AF2
5. Tip 52 ROF3 53 GF3 54 BF2 55 G20
G3 CF2 18X
Experimental Grades OF4
and Blends (oo) QCB
QF8
M81
M82
MIE
CF2A
Q1
Q2
Q3
Q4
Q5
Q6
6 7
Green Very Trashy
to Crude
17 F54
26 FOL
36 GC4
56 CF 57 G4G
CF2G G40