Weight Variation of Biscuits
Weight Variation of Biscuits
Weight Variation of Biscuits
To cite this article: Susanta Kumar Gauri (2003) Statistical process control procedures for controlling the weight of
packets of biscuits, Total Quality Management & Business Excellence, 14:5, 529-539, DOI: 10.1080/1478336032000053564
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of
the Content. Any opinions and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied
upon and should be independently verified with primary sources of information. Taylor and Francis shall
not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other
liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or
arising out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
TOTAL QUALITY MANAGEMENT, VOL. 14, NO. 5, 2003, 525–535
The thickness and weight of a biscuit are its two most important quality characteristics.
Although packets are not sold by weight, the declared weight is considered as the lower specification.
Any deviation of these characteristics from their respective nominal values leads to a variation in the
weight of packets of biscuits. High pack weight variation is a problem with considerable implications,
because underweight packets lead to customer dissatisfaction and excess weight results in loss of profit.
Conventionally, the weight and thickness of biscuits are monitored to control the weight of packets.
This study describes how the variation in weight of packets was reduced by setting appropriate target
values for dough weight and rise per gram (RPG ) and controlling these values using simple statistical
tools. This led to a substantial reduction of underweight packets from 12.2% to 2.0% and excess
weight packets from 2.1% to 0.9% of the declared weight. The statistical techniques used for the
purpose of controlling these characteristics are the test for equality of two means, and X-charts for
controlling dough weight and RPG on a regular basis.
Introduction
Biscuits are sold to consumers in packets of some declared weight. Each of these packets is
also expected to contain a specified number of biscuits. To satisfy a customer, the weight of
a packet should not fall below its declared weight. In order to eliminate the chance of packets
with less weight reaching the customer, the manufacturing target is set above the declared
weight. However, if the variation in pack weight during manufacturing is higher, the required
targeted weight will be higher. This will result in loss of profit.
This study demonstrates that profitability as well as customer satisfaction can be
increased substantially through
For the above purposes, simple statistical tools and Shewhart type control charts are found
to be useful.
Correspondence: S. K. Gauri, SQC and OR Unit, Indian Statistical Institute, 27B Camac Street, Calcutta
700016, WB, India. E-mail: susantagauri@hotmail.com
ISSN 1478-3363 print/ISSN 1478-3371 online/03/050525-11 © 2003 Taylor & Francis Ltd
DOI: 10.1080/1478336032000053564
526 S. K. GAURI
This study was carried out in an Indian biscuit manufacturing unit, where packets
containing biscuits in three stacks are sold in the market. The declared weight of a packet is
400 g with 69 biscuits.
The process
Different ingredients are mixed as per the recipe and dough is produced. Then a continuous
dough sheet of the required thickness and width is formed through a series of operations.
This dough sheet moves over a conveyor for further processing. The side of the conveyor
facing the operator is known as the working side (WS) and the opposite side is known as the
non-working side (NWS). The thickness of the dough sheet depends on the final gauge roller
setting, where sidewise adjustment is also possible. Dough pieces of desired shapes are then
Downloaded by [York University Libraries] at 06:53 30 December 2014
cut from the dough sheet by identical dies mounted on a rotating roller. Another conveyor
carries the dough pieces forward where a milk solution is sprayed on the top surface of the
dough pieces. Subsequently, the dough pieces are baked in ovens and the biscuits of required
thickness (also called height) and weight, are formed at the oven end. The biscuits are carried
to the packing section through another conveyor allowing sufficient time for cooling prior to
packing. In packing, a long row of biscuits is maintained on the inclined chute through which
biscuits are fed to the packing machine. The pusher fitted into the packing machine pushes
a stack of biscuits, which can be accommodated within the pusher length, into the channel.
The width of this channel is 124 mm. The stack is dragged towards the packet assembly unit
by the moving separators inside the channel.
The thickness of biscuits should be such that a stack of a specified number of biscuits
can be accommodated within the channel width. In this case, the specified number of biscuits,
i.e. count, is 23. If the pusher pushes fewer biscuits, the packing operator makes the necessary
correction by inserting biscuits during the movement of the stacks through the channel.
Three consecutive stacks are moved together to the packet assembly unit, where such sets of
stacks are wrapped to form a packet. The pack weight is mainly dependent on the count per
stack and the weight of individual biscuits. As the channel width is set according to the
specified packet length the count per stack is dependent only on the thickness of individual
biscuits. Thus, a biscuit’s thickness and weight are the only two characteristics that affect the
pack weight. Although the specified count per packet is 69 (3î23) at present, due to
variations in count per stack this value is observed to be between 67 and 73.
In order to assess whether the specified count and weight could be achieved in packets,
the following procedure is in use in the factory. The number of biscuits that can be stacked
in a wooden gauge box of 228.7 mm length is counted. This number is called ‘Gauge’. Then
the total weight of these biscuits is measured and, if this weight differs from 250 g, adjustment
is made to discover how many biscuits are required to get a weight of 250 g. This number is
used as a measure of weight and is simply called ‘Weight’. The target values for Gauge and
Weight for this particular variety of biscuit are 43.5 and 43 respectively. The sample biscuits
are collected randomly from the whole width of the conveyor for these Gauge and Weight
measurements.
roller setting is approved, one set of measurements on Gauge and Weight is made after baking,
following the existing procedure. If the supervisor judges that the target values for Gauge and
Weight have been achieved, the process is approved for regular production. Otherwise,
necessary adjustments are made. It is important to note that although the thickness of biscuit
is a critical factor with respect to variation in count per stack the existing practice does not
consider the possibility of differences in thickness of biscuits during the setting approval.
and dough weights are measured for monitoring purposes. The Gauge and Weight of biscuits
after baking are also measured on a routine basis and the observed values are compared with
their respective target values. If the observed Gauge and Weight are considered by the
supervisor/operator to be away from the target values action is initiated. The sample dough
weights are also taken into consideration while deciding upon the corrective action. This is
because both the Gauge and the Weight are dependent on the dough weight.
The problem
(1) The present target values for Gauge and Weight of biscuits (43.5 and 43 respectively)
have been fixed on an ad hoc basis based on experience. The gauge of 43.5 cannot
be counted and it is estimated subjectively by visual examination of the remaining
gap in the gauge box after stacking 43 biscuits. Obviously, measurement error is
high in the present measurement system leading to a high chance of inappropriate
process setting. The problem is further compounded because of:
Ω not knowing what should be the target dough weight to achieve a particular weight
of biscuits, taking into consideration the loss in weight due to baking;
Ω not utilizing statistical methodology for testing the equality of dough weight
between the two sides. This may result in improper sidewise adjustment in the
final gauge roller, leading to an increase in variability of a biscuit’s weight;
Ω non-consideration of the requirement of achieving equality of thicknss of biscuits.
A detailed investigation has revealed that there was a significant difference in
thickness of biscuits. This would lead to an increase in variability of thickness.
(2) No control charts are used to identify whether process adjustments are required.
Clearly, this results in frequent unnecessary process adjustments. It is to be noted
that as both the thickness and weight of biscuits are affected by dough weight, the
two characteristics are related. Monitoring by T 2 statistic, where the covariance
between the characteristics is taken into consideration, could be ideal to detect the
control state of the process. However, shop floor computation of this statistic is a
serious problem.
(3) The thickness of a biscuit depends on its weight (which in turn depends on dough
weight) as well as on some other factors, such as amount of milk solution spray,
oven heat and recipe. Because of this dependence of thickness on multiple factors,
identification of action to be taken often becomes difficult. Consequently, the time
needed for corrective action is usually high.
528 S. K. GAURI
(4) To identify the appropriate characteristics for controlling the height and weight of
biscuits, such that identification of corrective action becomes easier.
(5) To suggest a methodology for approving process setting and online control of the
identified characteristics based on statistical principles.
Approach
(1) Instead of expressing the thickness and weight of biscuits by count, if the total
thickness and weight of a particular number of biscuits are measured, the measure-
ment error will be less. Therefore, the best proposition is to measure the overall
height and weight of 23 biscuits, i.e. the target number of biscuits per stack, and to
control these parameters.
(2) The target weight of packets containing 69 biscuits is to be determined statistically
based on the inherent variation of the total weight of 69 biscuits. Obviously the
target total weight for 23 biscuits should be a third of the target weight of the packet.
The corresponding target dough weight should be decided based on the loss of
weight during baking. A one-to-one correspondence between samples of dough
pieces and samples of biscuits is to be ensured using the observed time lag based on
the conveyor speed. Regression analysis is thought to be useful here.
(3) The number of packets containing 69 count will be maximized if the number of
stacks with 23 count within the channel width in the packing machine is maximized.
23 biscuits will be accommodated in a stack as long as the overall height of 23
biscuits lies between the channel width and the channel width minus the thickness
of a biscuit. As the overall height of 23 biscuits follows a normal (i.e. symmetric)
distribution, the targeted height of 23 biscuits should be equidistant from the
channel width and the channel width minus the thickness of a biscuit (See Fig. 1).
This yields the following equation:
H
WñHóHñ Wñ (1)
23
where W is the channel width, H is the mean height of 23 biscuits and (H/23) is the
mean thickness of individual biscuits.
(4) In order to make the identification of action easier, a quantity called rise per gram
(RPG) is introduced. The RPG is defined as the quantity obtained by dividing the
overall height of 23 biscuits by the total weight of the same 23 biscuits. Thus, RPG
is a ratio of two correlated normal variables. The RPG approaches normality if
k2 p2ñ1 ê, where k2 and p2 are the mean and standard deviation (SD) of the total
CONTROLLING BISCUIT WEIGHT 529
Downloaded by [York University Libraries] at 06:53 30 December 2014
Mean SD
Calculated
SI. No. Characteristics WS (X̄1) NWS(X̄2) WS (s1 ) NWS (s2 ) Pooled (sp ) ‘t’ value
1 Weight of 10 dough
pieces (g) 75.47 75.34 0.79 0.62 0.71 0.45
2 RPG (mm/g) 0.900 0.888 0.011 0.008 0.0096 3.06
independent normal distributions was carried out using a t-test for each of the characteristics.
The results are summarized in Table 1.
Since t0.025,22 ó2.074, it was concluded that no significant difference was present in
Downloaded by [York University Libraries] at 06:53 30 December 2014
dough weight between WS and NWS, but a significant difference was present in RPG
between WS and NWS. The root cause of the sidewise difference in RPG was investigated
and found to be the non-uniformity of oven heat between the two sides. It might be worth
noting here that this finding was contrary to the popular belief in the factory that a sidewise
difference in a biscuit’s characteristics is created by a sidewise difference in dough weight.
The sidewise equality of RPG was tested again after initiating corrective action and no
significant difference was observed.
MR method. The moving ranges were based on two successive observations. The estimates
of the inherent variations for these characteristics are given in Table 2. The X-MR charts for
these characteristics showed that the process was stable during the data collection period.
The standard deviation of weight of a set of 69 biscuits was estimated from the estimated
SD of weight of 23 biscuits, and was found to be 1.75 g. Under the present system, weights
Downloaded by [York University Libraries] at 06:53 30 December 2014
of a set of 13 dough pieces are used for controlling the overall dough weight. So the SD of
the weight of 13 dough pieces was obtained from the estimated SD of the weight of 23 dough
pieces and it was found to be 0.87 g.
The correlation coefficient between the weight of a biscuit and the weight of a dough
piece was found to be 0.85 for the 21 pairs of observations. The regression analysis yielded
the following equation:
Weight of a biscuitó0.761îWeight of a dough piece (2)
It is interesting to note that the above finding matches the existing standard of 24% loss due
to baking. Based on equation (2), the required overall weight of 23 dough pieces to achieve
an overall weight of 134.5 g for 23 biscuits was obtained as 176.74 g, implying that the target
weight for a set of 13 dough pieces should be 99.9 g.
Target value of RPG. In the present case, the value of W is 124 mm. Putting this value of W
in equation (1), the value of H is obtained as 121.35 mm. Therefore the target value of RPG
Downloaded by [York University Libraries] at 06:53 30 December 2014
Control limits for the weight of a set of 13 dough pieces and the RPG
The target values of the characteristics are chosen as the central lines (CL) of the respective
control charts. A distance of three times the inherent variation from the central line yielded
the control limits for the characteristics as shown in Table 3.
In order to ensure adequate confidence that the present level of process variation is
maintained, the equality of sidewise dough weight and RPG should be tested twice during a
shift as well as at the time of process setting.
Procedure for the approval of uniformity of oven heat and/or amount of milk
solution spray over the two sides
The procedure for approval of uniformity of oven heat and/or amount of milk solution spray
over the two sides is recommended, taking into consideration the pooled estimate of sidewise
SD of RPG, which was found to be 0.0096 mm/g (see Table 1). The procedural steps are as
follows.
Step 1. Collect samples of five sets of 23 biscuits from WS as well as NWS. Measure heights
and weights of these samples and obtain the values of RPG. Compute the average
RPG ( ȳ1) and ( ȳ2) at WS and NWS respectively. Find the difference of averages,
( ȳ1 ñȳ2).
Step 2. If the value of ( ȳ1 ñȳ2) lies beyond the range [ñ0.012, 0.012] (95% confidence
Downloaded by [York University Libraries] at 06:53 30 December 2014
level) take appropriate corrective action to ensure uniformity of oven heat and/or
quantity of milk solution spray over both sides. Otherwise no corrective action should
be taken.
Step 3. Whenever any corrective action is taken, verify whether the action is taken correctly.
For this purpose, after every action, collect samples, take measurements and make a
computation following Step 1. If the value of ( ȳ1 ñȳ2) obtained after corrective action
lies within the range [ñ0.006, 0.006] (68.27% confidence level) consider that the
adjustment is made correctly. Otherwise continue readjustment and re-verification
until the value of ( ȳ1 ñȳ2) is obtained within the range [ñ0.006, 0.006].
Approval of both the sidewise final gauge roller setting and oven heat and milk solution spray
setting are to be made by the supervisor.
Step 1. If the first observation, or two out of three consecutive observations, on dough weight
fall within the green zone marked in the control chart for dough weight, approve the
settings of the gauge roller.
Step 2. If the first observation or two out of three consecutive observations on RPG fall
within the green zone marked in the control chart for RPG, approve the settings of
oven heat, milk solution spray and the recipe adjustment.
Before After
SI. No. Characteristics implementation implementation
*The difference of average pack weight from the declared weight is considered
as excess weight and is expressed as a percentage of the declared weight.
Downloaded by [York University Libraries] at 06:53 30 December 2014
coded region the cross mark (î) falls. During implementation, this form of control chart
was found to be highly acceptable to the operators.
No adjustment in the process should be made as long as the cross mark (î) falls in the
green zone or the white zone. The purpose of the green zone is to approve the setting or
adjustment. The adjustment will be made only when any cross mark (î) falls in the red zone.
In case the observation on dough weight falls in the red zone, the adjustment is to be
made in the final gauge roller setting. If the observation on RPG falls in the red zone, the
adjustment is to be made either in the oven heat setting or the air pressure in the milk
solution spray system or in the recipe as decided by the operator/supervisor based on
experience/expertise.
Since all adjustments are made on a trial and error basis, approval of the adjustments is
very important. For this purpose the following guidelines are provided.
After every adjustment, check the value of the particular characteristic. If the value falls
in the green zone the adjustment made for the characteristics will be considered appropriate.
Otherwise continue readjustment and re-verification until the observation after adjustment
falls in the green zone. Plot the value obtained after the final adjustment in the same column
where the out of control point was indicated and then go on plotting observations at regular
intervals as usual.
Training programmes were organized for the management, supervisors and operators of
this factory to explain the concept of natural variability, its relevance to the control limits
and the interpretation of different zones. In the training programmes, the effects of taking
actions with and without the help of control limits were demonstrated through real-life data
collected within the factory. In addition, the operators were given hands-on training on
rounding off of data, plotting of cross marks (î) and interpretation of control charts.
References
H, D.V. (1969) On the ratio of two correlated normal random variables, Biometrica, 56, pp. 635–639, 683.
R, K.C.B., D, R.J.M.M. & S, Y. (1993) Shewhart-type control charts for individual observa-
tions, Journal of Quality Technology, 25, pp. 188–198.
CONTROLLING BISCUIT WEIGHT 535