Use of Control Charts in The Production of Concrete: by Ian Gibb and Tom Harrison October 2010

Use of control charts in the production of concrete
by
Ian Gibb and Tom Harrison
October 2010
© 2010 The authors and MPA/BRMCA - ERMCO
The text is published on the understanding that the authors are solely responsible for the
statements made and opinions expressed in it and that its publication does not necessarily imply
that such statements and/or opinions are of reflect the views or opinions of the publishers. While
every effort has been made to ensure that the statements made and the opinions expressed in this
Publication provide a safe and accurate guide, non liability or responsibility can be accepted in this
respect by the authors or publishers.
Table of Contents
Symbols
1 Introduction
2 Statistics for Concrete
2.1 Normal distribution of strength
2.2 Characteristic strength and target strength
2.3 Standard deviation
2.4 Setting the target strength
3 Simple Data Charts
4 Shewhart Charts
4.1 Introduction
4.2 Shewhart action criteria
4.3 Control of standard deviation
4.4 Example Shewhart chart
4.5 Modified application of Shewhart control chart
5 CUSUM
5.1 Introduction
5.2 Controlling mean strength
5.3 Controlling standard deviation
5.4 Controlling correlation
5.5 Design of V-mask
5.6 Action following change
6 Multivariable and Multigrade Analysis
6.1 Multivariable
6.2 Multigrade
7 Speeding the Response of the System
7.1 Early age testing
7.2 Family of mixes concept
8 Guidance on Control Systems
8.1 Abnormal results
8.2 Handling mixes outside of the concrete family
8.3 Handling mixes not controlled by compressive strength requirements
8.4 Test rates
9 EN 206-1 Conformity Rules for Compressive Strength
9.1 Basic requirements for conformity of compressive strength
9.2 Assessment period
9.3 Conformity rules for compressive strength
9.4 Achieving an AOQL of 5% with CUSUM
9.5 Non-conformity
10 Implementing Control Systems in EN 206-1
11 CUSUM Example
11.1 Reference mix and concrete family
11.2 Main relationship
11.3 Applying adjustments
11.4 CUSUM calculation
11.5 CUSUM action following change
11.6 Further data and a change in standard deviation
12 References
Page 1 of 53
Symbols
AOQ Average outgoing quality
AOQL Average outgoing quality limit
fci Individual test result for compressive strength of concrete
fcm Mean compressive strength of concrete
σ Estimate for the standard deviation of a population
k Statistical constant
fck Specified characteristic compressive strength
Ll Lower limit
n Number of samples
qn Statistical constant that depends upon n and the selected AOQL
s Sample standard deviation
UCL Upper Control Limit
UWL Upper Warning Limit
LWL Lower Warning Limit
LCL Lower Control Limit

Constant giving the cement content increase required to
Cmra
produce a 1N/mm2 increase in strength
DI Decision Interval
G Gradient
dc Change in cement content

Test result
xi NOTE: According to EN 206-1, a test result may be the mean value of two or
more specimens taken from one sample and tested at one age.
Mean value of ’n’ test results
Page 2 of 53
1 Introduction
It is safe to assume that ever since manufacturing commenced, attempts have
been made to control the process in order to improve quality and drive down
costs. The application of statistical techniques to manufacturing was first
developed by physicist Walter A. Shewhart of the Bell Telephone Laboratories
in 1924. Shewhart continue to develop the idea and in 1931 he published a
book on statistical quality control [1].
Shewhart recognised that within a manufacturing process there were not only
natural variations inherent in the process, which affected quality but there
were also variations that could not be explained. Shewhart recognised that it
is possible to set limits on the natural variation of any process so that
fluctuations within these limits could be explained by chance causes, but any
variation outside of these limits, special variations, would represent a change
in the underlying process.
Shewhart’s concept of natural and special variations is clearly relevant to the
production of concrete at a ready-mixed plant or precast factory and the
requirement to achieve a specified compressive strength. Natural variations
exist in the process due to variation in the raw materials (aggregate grading,
chemical composition etc), batching accuracy, plant performance, sampling
and testing etc. Special causes of variation, outside of the natural variations
could be due to changed constituent materials being used, weigh-scales losing
accuracy, a new batcher, problems with testing equipment etc.
Control charts have found widespread use in the concrete industry in both
ready-mixed concrete and precast concrete sectors as a tool for quality
control. Control charts can be applied to monitor a range of product
characteristics (e.g. cube/cylinder strength, consistence, w/c ratio),
constituent materials (aggregate grading, cement strengths etc.) or production
(batching accuracy).
Their most common application of control charts is as a means of continuously
assessing compressive strength results in order to:
- check whether target strengths are being achieved;
- measure the variations from target (all products vary);
- identify magnitude of any variation;
- objectively define action required (e.g. change w/c ratio) to get the
process back on target;
- identify periods and concretes where the strength was less than
specified, so that investigations can be carried out and corrective action
taken.
The use of control charts should not be treated in isolation from the rest of
production control. For example routine checking and maintenance of weigh
equipment will minimise the risk of a weigh-scale failure. Control charts
provide information about the process, but the interpretation of the
information is not a mechanical process. All the information available to the
concrete producer should be used to interpret the information and make
informed decisions. Did a change in quality occur when a new batch of
constituent was first used? Is all the family showing the same trend? Are other
Page 3 of 53
plants using similar materials showing a similar trend? Such information leads
to the cause of the change in quality being identified and appropriate action
being taken. For example a loss of accuracy in the weigh-scales should lead to
repair, maintenance and re-calibration and not a change in mix proportions.
Where a change in mix proportions is required, the use of control charts can
lead to objectively defined changes in proportions.
Effective control of concrete production is more easily achieved when there
are good relationships with the constituent material suppliers, particularly the
suppliers of cementitious materials. Early warning of a change in performance
from the constituent material supplier should be part of the supply agreement,
e.g. that stock clinker is being used during the maintenance period, and on the
basis of this warning, the producer will decide the appropriate action.
Some producers use changes in cement chemistry to predict changes in
concrete strength. Effective production control is about using all this
information to produce concrete conforming to its specification. Effective
production control, which includes the use of control charts, significantly
reduces the risk of non-conformity benefiting both users and producers of
concrete.
There are drawbacks to the existing method of assessment of conformity of
mean strength adopted in EN 206-1 including not following the CEN Guidance
on the evaluation of conformity [2]. It is believed that control charts (already
widely used as a quality assurance tool in factory production control) would
provide an alternative and better means of ensuring the characteristic strength
is achieved and it is a method that follows the CEN Guidance.
This publication will review various control systems that are currently used in
the concrete industry and, by the use of examples, show how the principles are
applied to control the production of concrete.
Page 4 of 53
2 Statistics for Concrete

2.1 Normal distribution of strength
Compressive strength test results tend to follow a normal distribution as
illustrated in Figure 1. A normal distribution is defined by two parameters, the
mean value of the distribution and the standard deviation (σ), which is the
measure of the spread of results around the mean value. A low standard
deviation means that most strength results will be close to the mean value; a
high standard deviation means that the strength of significant proportions of
the results will be well below (and above) the mean value. The area under the
normal distribution between two values of ‘x’ represents the probability that a
result will fall within this range of values. The term ‘tail’ is used to mean the
area under the normal distribution between a value, e.g. a compressive
strength, and where the frequency is effectively zero. For strength it is the
lower tail, i.e. low strength results, that is important but for other properties,
e.g. consistence, both the lower and upper tails are important.
Figure 1: Illustration of concrete strength distribution
At the extremes of the strength range for a given set of constituent materials,
the assumption of a normally distributed set of data may not be valid. It is not
possible to have strengths less than zero and most concretes have a ceiling
strength beyond which they cannot go. In these situations the data set is
skewed. However as low strengths are of concern to specifiers, an assumption
of normally distributed data does not lead to problems in practice.
2.2 Characteristic strength and target strength

EN 206-1[3] specifies the characteristic compressive strength of concrete in
terms of a standard cylinder test or a standard cube test carried out at 28 days.
The characteristic strength is defined in EN 206-1 as the “value of strength
below which 5% of the population of all possible strength determinations of the
volume of concrete under consideration, are expected to fall”. Put simply this
means that if every single batch was tested, 5% of the results would fall within
the lower ‘tail’ of the normal distribution that starts 1.64σ below the actual
Page 5 of 53
mean strength. However the actual mean strength will not be known until the
concrete has been produced and tested and therefore the target mean strength
(TMS) is usually set at some higher value to ensure the concrete achieves at
least the specified characteristic strength.
The target mean strength is given in Equation 1.
TMS = fck + k x σ equation 1
Where TMS = target mean strength
fck = characteristic compressive strength
σ = estimate for standard deviation of population
k = statistical constant
k x σ = the margin
The fixed point in the distribution is the specified characteristic strength and
as the margin increases and/or the standard deviation increases, the target
mean strength increases, see Example 1.
Example 1
The target mean strength for a specified characteristic strength of C25/30 is
given in Table 1. A standard deviation (σ) of 3 N/mm2 is typical of a concrete
with low variability and a value of 6 N/mm2 represents high variability.
Table 1: Target mean strength for specified characteristic strength of
30N/mm2 (cube)
Margin Area in lower tail Target mean strength (cube),
(i.e. percentage N/mm2
2
below characteristic σ = 3 N/mm σ = 6 N/mm2
strength)
1.64σ 5% 35 40
1.96σ 2.5% 36 42
2.00σ 2.28% 36 42
2.33σ 1.0% 37 44
3.0σ 0.13% 39 48
The numbers in this table have been rounded.
A concrete strength below the characteristic strength is not a failure as

statistically 5% of the results are expected and accepted as to fall below this
value. However for structural safety reasons, a batch with a concrete strength
significantly below the characteristic strength is excluded, even though it
forms part of the expected population. Consequently EN 206-1 specifies a
minimum strength requirement for individual results (fci) of (fck — 4). Any batch
below this strength is a non-conforming batch.
The risk of non-conformity decreases as the margin increases. Statistics are
used to quantify that risk. For a given margin the probability of a test result
falling below the specified characteristic strength or failing the individual
strength criterion is given in Table 2. Table 2 shows that the probability of
having a result below the specified characteristic strength is independent of
the standard deviation (as the margin is based on the standard deviation) but
the risk of failing the criterion for individual batches increases as the standard
deviation increases.
Page 6 of 53
Table 2: Effect of margin on proportion of concrete below characteristic

strength; and risk of failing the strength criterion for individual batches
Margin Probability of a test Risk of failing the strength criterion
result being below for individual batches
the characteristic σ = 3 N/mm2 σ = 6 N/mm2
strength
1.64σ 1 in 20 (5%) 0.1% 1%
1.96σ 1 in 40 (2,5%) 0.05% 0.4%
2.33σ 1 in 100 (1%) 0.01% 0.1%
3.08σ 1 in 1000 (0,1%) 0.0005% 0.01%
The definition of ‘characteristic strength’ in EN 206-1:2000 has its

complications. For a structural engineer the phrase ‘the volume of concrete
under consideration’ may be applied to all the concrete in their structure and
to the concrete in a single element of that structure even if this comprises a
single batch. For conformity to EN 206-1, the ‘volume under consideration’ is
all the concrete in an assessment period. Neither of these interpretations of
this phrase is suitable for use in control systems as the process is continual.
Caspeele and Taerwe [5] have proposed that if the production achieves an
average outgoing quality limita (AOQL) of 5%, the production can be accepted
as having achieved the characteristic strength.
2.3 Standard deviation

The standard deviation of a population will only be truly known if every batch
of concrete is tested. However if 35 or more results are available, the
estimated standard deviation is likely to be very close to the true standard
deviation. This is the reason why EN 206-1 requires 35 results to calculate the
initial standard deviation
When n ≥35, the standard deviation may be estimated using the equation:
∑ (x − x )
2
Standard deviation, σ = i
(n − 1)
Alternatively it can be determined through a range of pairs approach where
Mean range of successive pairs = 1.128 x standard deviation equation 2
or,
Standard deviation = 0.886 x mean range of successive pairs of results
The range is the numerical difference between successive results and the
difference is always taken as a positive number, e.g. |2-3|=1. The range of
pairs method of calculating the standard deviation is particularly suited for
populations where there are step changes in mean strength in the data set,
e.g. concrete, as the effect of the step change will be limited to a single pair
of results. With concrete production, step changes in mean strength (usually
a
From the operating-characteristic curve for the selected sampling plan, the average outgoing
quality (AOQ) curve is determined by multiplying each percentage of all possible results below
the required characteristic strength in the production by the corresponding acceptance
probability.
Page 7 of 53
due to a change in a constituent) are more common than drifts in mean

strength.
Example 2
Table 3: Calculation of the standard deviation using mean range

Result Transposed Range,
cube N/mm2 Calculation of standard deviation
strength,
N/mm2
1 54.5
2 52.5 2.0
3 49.5 3.0
4 47.5 2.0 Estimation of the standard deviation
5 49.0 1.5
6 43.5 5.5 = 0.886 x 51/14
7 54.5 11.0
8 46.5 8.0 = 0.886 x 3.64 = 3.0 N/mm2 (rounded to the
9 50.0 3.5 nearest 0.5 N/mm2)
10 50.5 0.5
11 47.0 3.5
12 48.5 1.5
13 53.0 4.5
14 51.5 1.5
15 48.5 3.0
Sum of ranges 51.0
Mean of ranges 3.64
Example 3 (copied from reference [4])

15 random data have been generated assuming a mean strength of 37.0 N/mm2
and a standard deviation of 3.5 N/mm2. These have been repeated to give a
total of 30 data, see Figure 2a. The standard deviation of the 30 data given in
Figure 2a is:
3.6 N/mm2 when determined by the standard method;

3.7 N/mm2 when determined from 0.886 x mean range.
To illustrate the effect of a change in mean strength on the standard

deviation, an extreme reduction in mean strength of 5.0 N/mm2 is introduced
at result 16 i.e. data 16 to 30 are all 5.0 N/mm2 less than in Figure 2a. The
dispersion of the data around these mean strengths is unchanged. The
standard deviation of the 30 data given in Figure 2b is:
4.4 N/mm2 when determined by the standard method;

3.8 N/mm2 when determined from 0.886 x mean range.
This shows that the standard deviation calculated from the mean range has
been less affected by the change in mean strength.
Page 8 of 53
45
Compressive strength - N/mm2 40
35
30
25
20
15
10
5
0
0 5 10 15 20 25 30
Result number
Figure 2a. Fifteen random data generated assuming a mean strength of 37.0 N/mm2
and a standard deviation of 3.5 N/mm2 (the first group of 15 results are the same
as the second group of 15 results).
45
Compressive strength - N/mm2
40
35
30
25
20
15
Reduction of 5 N/mm2
10
in the mean strength
5
0
0 5 10 15 20 25 30
Result number
Figure 2b. The same data as in Figure 6a, but with a reduction in mean strength of
5.0 N/mm2 introduced at result 16.
The true standard deviation of a population, σ, can only be determined if all

the population were to be tested, which is impractical. In practice the
population standard deviation is estimated by testing samples. The more
samples that are tested, the more reliable will be the estimated population
standard deviation. EN 206-1 requires at least 35 results to initially estimate
Page 9 of 53
the population standard deviation. Prior to obtaining the estimated population

standard deviation, concrete is controlled by more conservative initial testing
rules. Without an estimated population standard deviation, it is not possible to
use control charts to control the concrete production.
Once the initial population standard deviation has been estimated, EN 206-1
permits two methods for verifying the initial estimate. The first method
involves checking that the standard deviation of the most recent 15 results
does not deviate significantly from the adopted value. The second method
involves the use of continuous control systems.
The standard deviation for strength tends to be constant for medium and high
strength mixes but for lower strengths it tends to increase proportionally with
mean strength [6] and the relationship illustrated in Figure 3 may be assumed.
In practice this means that the standard deviation for concretes that have a
characteristic strength of 20 N/mm2 or more is determined by testing and
calculation, while the standard deviation for concrete with a lower strength is
interpolated.
Standard Deviation (N/mm 2)
Plant SD
27
Strength (N/mm2)
Figure 3: Simplified standard deviation to mean strength relationship
2.4 Setting the target strength

The target strength is set to achieve a balance between the following
requirements:
- high probability of achieving a population with at least the specified
characteristic strength;
- low risk of failing the minimum strength criterion;
- low consumers risk;
- low producers risk;
- competitive and economic.
The target strength is selected by the producer, but the producer may have to
comply with certain minimum values. The target strength should never be
lower than (fck +1.64σ), but it is normally higher than this value. National
Page 10 of 53
requirements, the requirements of a certification body or other requirements

(see 9.4) may impose minimum target strengths.
UK experience is that a minimum target strength of (fck +1.96σ) at a test rate
of at least 16 results per month is a good balance between these conflicting
demands. With a concrete family this gives about a 3σ margin, i.e. a 1:1000
risk of failing the minimum strength requirement (fck — 4). Data collected by a
UK certification body on individual batch non-conformities shows that the
actual rate of non-conformity is an order of magnitude lower and this is due to
the active control of the production.
Page 11 of 53
3 Simple Data Charts

Simple data control charts are used to routinely monitor quality. There are two
basic types of control charts.
Univariate – a control chart of one quality characteristic (e.g. mean strength)
Multivariate - control chart of a statistic that summarises or represents
more than one quality characteristic (e.g. coefficient of variation)
If a single quality characteristic has been measured or computed from a
sample, the control chart shows the value of the quality characteristic versus
the sample number or versus time.
Simple data charts are useful in providing a visual image of production and
unusual results. Simple charts may also give an indication of trends but the
general scatter of the data may also mask trends that can be identified only by
more in-depth analysis of the data.
Consider the data in Table 4 and illustrated in Figure 4. A review of the data
shows that all the results are within +/- 8 N/mm2 of the target. The results are
fairly evenly distributed around the target (2 on target, 9 results above and 7
below) so it is not immediately obvious what conclusions can be drawn from
the data.
Table 4: Example data for mean strength with a target strength of 40N/mm2
Result 28 day strength, Result 28 day strength,
N/mm2 N/mm2
1 37 10 40
2 42 11 34
3 36 12 44
4 35 13 46.5
5 42 14 42
6 38 15 44.5
7 39.5 16 45
8 40 17 44
9 35 18 48
Figure 4: Simple univariate control chart for strength
Page 12 of 53
4 Shewhart Charts
4.1 Introduction
While graphical plots can give useful information about the pattern of a
production process, the control chart becomes a much more powerful tool if
statistical rules are also applied to the data. Shewhart control systems
measure variables in the production processes (e.g. target mean strength).They
make use of calculated control limits and apply warning limits based on the
measured variation in the production process.
ISO 8258 [7] gives general information on Shewhart control charts and ISO 7966
[8] gives general information on Shewhart control charts for acceptance
control.
The Shewhart chart will have a horizontal central line which represents the
expected mean value of the test results on the samples taken from production;
in the case of concrete, the Target Mean Strength for a chart controlling
compressive strength. Lines representing the upper control limit (UCL) lower
control limit (LCL), upper warning limit (UWL) and lower warning limit (LWL)
may also be added. Generally action is required if a result is beyond either of
the control limits.
The UWL and LWL are set at a level so that most of the results will fall
between the lines when a system is running in control. These are not
specification limits but ‘warning’ limits based on the variability of the
production process. Given that concrete strengths follow a normal distribution
(Figure 1), it follows that there is a 50% chance that a result will be above the
TMS and a 50% chance that it is below the TMS. In chapter 2 it was shown that
a margin of 1.96 x σ will lead to 2.5% of results being below the characteristic
strength. Some variables, e.g. consistence, have both upper and lower limits
and in these cases it is essential to have both an UWL and a LWL. While for
conformity to a specified characteristic strength a high value is not significant,
from the viewpoint of economic production it does matter. Therefore in
practice, both upper and lower warning limits are used even for a variable that
has a single limit value, e.g. concrete strength. Setting upper and lower
warning limits at 1.96σ leads to the expectation that 95% of the results will fall
within these limits and 2.5% in each of the ‘tails’ of the normal distribution. If
a margin of 3.0 x σ is adopted, there is very little chance of a result falling
outside this limit due to natural variation (0.3% for two-tailed test). A
Shewhart control chart can be constructed with
UCL = TMS + 3 x σ
LCL = TMS – 3 x σ
UWL = TMS + 2 x σ
LWL = TMS – 2 x σ
The probability of a single result falling outside of either the UWL or LWL is
4.56%, i.e. 2.28% above the UWL and 2.28% below the LWL (see Table 1Table
2).
The probability of two consecutive results falling outside the limits purely by
chance is
= 0.0456 x 0.0456 = 0.002079 or 0.21%
Page 13 of 53
The probability that the two results are either both above or below the line
(i.e. in the same direction) is only 0.05%. Such an outcome is very strong
evidence that the expected outcome is not being achieved.
4.2 Shewhart action criteria

4.2.1 Points beyond UCL or LCL
The presence of one or more points lying outside of the UCL or LCL is primary
evidence that the system is out of control at that point. Since there is only a
0.3% chance that this result is due to natural variation, it is probable that
special variation will account for the extreme value and an immediate
investigation into the cause should be undertaken.
4.2.2 Points beyond UWL or LWL

The presence of two consecutive, or more than 1 in 40, points beyond either
warning line is evidence that the process is out of control and an investigation
of the data should be undertaken.
4.2.3 Patterns within control limits

It is also possible to analyse data that doesn’t breach either the control or
warning limits to evaluate whether any trends are significant. Runs analysis
can give the first warning of a system going out of control before points are
seen beyond the warning limits.
The following simple rules of thumb have been proposed for sequences of
results that remain within the warning limits [9]:
1. Seven or more consecutive results on the same side of the target mean
strength
2. At least 10 out of 11 results on the same side of the target mean
strength
strength
strength
4.3 Control of standard deviation

The control and warning limits are determined by the standard deviation of the
process; it is therefore important to monitor the standard deviation. As the
calculation to determine standard deviation is relatively complex, the
alternative calculation in equation 2 is used linking standard deviation to the
range of pairs of results. Plot the running mean range of the last n successive
results where n ≥ 15 against test result number. Select the change in standard
deviation that will prompt action (Δσ) and set action lines at:
1.128*current standard deviation ± 1.128* Δσ
4.4 Example Shewhart chart

Consider again the strength data in Table 4 and subject it to a Shewhart
analysis using the rules stated in 4.1.3. Figure 5 shows the data with a UCL,
Page 14 of 53
LCL, UWL and LWL applied. Immediately it is apparent that point 18 has
exceeded the UWL This does not breach the rule defined in 4.1.3 (requiring 2
consecutive points above UWL) but also at this point there is a sequence of 7
points on the same side of the target mean strength (see 4.2.3). The Shewhart
chart is showing that the process is out of control, i.e. the actual mean
strength is higher than the mean strength required.
Figure 5: Control levels applied to data
4.5 Modified application of Shewhart control chart

If the aim is to assess whether the level of production is higher than a specified
characteristic value, a modified application of Shewhart control charts can be
used, with the use of specific modified variables. This application comprises
checking that the average of n measured strength results is greater than a
lower line Ll situated at a given distance from fck with the following variables:
Ll ≥ fck + (qn x s) where
- qn depends on n and on the AOQL chosen,
- s is an updated evaluation of the standard deviation of the relevant
production.
In the case where n ≥ 15 and qn ≥ 1.48, the Shewhart charts will satisfy the
requirement for an AOQL of 5%. This criterion also satisfies the conformity
criteria for mean strength in EN 206-1. A warning line at some higher value may
also be appropriate.
Example 4
A precast concrete factory intends using a Shewhart chart system to show
conformity to the mean strength criterion in EN 206-1. Due to process
requirements the strengths tend to exceed the characteristic strength within a
few days and therefore they opt to test at a real age of 7 days to verify that
the specified 28 day strength is already being achieved at 7 days. As the
compressive strengths are expected to be well above the specified strength,
they opt not to have a warning line.
Page 15 of 53
To do this task the running average strength of the last ‘n’ consecutive
results, where ‘n’ is a predetermined number that is at least 15 are plotted on
a Shewhart chart with one limit line with a value of (fck + 1.48s). If the
running mean strength below this line this indicates that an AOQL of 5% is not
being achieved. A warning line at some value higher than (fck + 1.48s) may be
added.
The specified compressive strength class is C25/30 and they use cubes for
assessing the production and conformity. The current standard deviation is
2.5 N/mm2. The limit value Ll is:
30 +1.48*2.5 = 33.7 N/mm2
For control purposes, rather than using non-overlapping groups of results, they
opt to use the running mean of the last 15 results. This is shown in Figure 6a),
which shows that the mean strength is consistently above the limit value.
Figure 6a): Control of conformity of mean strength using a running mean of

15 results
It is also necessary to check that the sample standard deviation has not
changed significantly. Though clause 8.2.1.3 of EN 206-1:2000 states that the
present value is still applicable if the calculated value based on the last 15
results is within about ± current value/3, this is not a very sensitive indicator
of change and most producers would regard a 0.5 N/mm2 change in standard
deviation as significant and the precast company uses this value of 0.5 N/mm2,
which is controlled with another modified Shewhart chart. On this chart a
horizontal line is drawn at the current mean range value (1.128σ) and action
lines ± 1.128 x 0.5 from this value.
Again a running mean range of the last 15 consecutive and overlapping pairs of
results is used. When the running mean value crosses one of these action lines,
this indicates that the standard deviation has changed by 0.5 N/mm2 and a
new value should be applied.
In this example the current standard deviation is 2.5 N/mm2 and this equates
to a mean range of 1.128 x 2.5 = 2.82 N/mm2 and upper and lower action lines
at 3.38 N/mm2 and 2.26 N/mm2 (2.82 ± 1.128 x 0.5). These are shown in Figure
6b). At test result 26, the mean range crosses the upper action line indicating
Page 16 of 53
that the standard deviation has increased by 0.5 N/mm2. The limit value is
increased in Figure 6a) to 34.4 N/mm2 and in Figure 6b) a new mean range is
set at 3.38 N/mm2 with upper and lower action lines set at 3.94 N/mm2 and
2.82 N/mm2 respectively. As the running mean strength is still well above the
limit line, the mix proportions are not changed, i.e. the appropriate action is
to take no action other than change the values on the control charts.
Figure 6b): Control of standard deviation using the running mean range of
the last 15 ranges
Page 17 of 53
5 CUSUM
5.1 Introduction
CUSUM control systems (short for cumulative sum) were developed in the
1950s, initially for quality control of continuous manufacturing processes. They
have found widespread use in the concrete industry. In CUSUM charts, the
central line does not represent a constant mean value but is a zero line for the
assessment of the trend in the results. In concrete production three CUSUMs
are used:
• CUSUM M, for the control of mean strength;
• CUSUM R (range), for the control of standard deviation;
• CUSUM C, for the control of correlation.
The CUSUM method, described in more detail in BS 5703[9] and Concrete
Society Digest No. 6[10] and ISO/TR 7871[11], involves subtracting the test
result from a target value then producing an ongoing running sum (the CUSUM)
of the differences. If the process is in control, the points on the CUSUM plot
are distributed randomly (positive and negative differences cancelling each
other out), to give an accumulative sum that is close to zero, but if the process
slips out of control, this will be quickly illustrated by the CUSUM plot moving
towards the UCL or LCL.
BS 5700[12] describes the following advantages of the CUSUM system:
a) for same sample size it gives a more vivid illustration of any changes;
b) uses data more effectively therefore produces cost savings;
c) gives clear indication of location and magnitude of change.
CUSUM charts have been found to be more sensitive at detecting small shifts in
the mean of a process than Shewhart, whereas Shewhart charts are superior at
detecting large shifts [13]. When the CUSUM reaches the UCL or LCL, it is
possible to use the plot to determine at what point the process went out of
control and what scale of corrective action is required.
Historically, CUSUM control charts were plotted manually and to determine
whether a trend in the plot was significant or not, a transparent mask in the
shape of a truncated (cut-off) V is placed on its side over the plot, with the
lead point placed on the most recent result, not on the central line. The
transparent V-mask overlay in Figure 7 is shown in red. Each limb is marked
with its standard deviation and in this example the values run from 3.0 N/mm2
in 0.5 N/mm2 steps to 5.0 N/mm2. The limbs have no limitation on their length.
The ‘arms’ of the V-mask represent the upper and lower control limits. If the
plot crossed either the upper or lower arm of the V, a significant change is
deemed to have occurred, see Figure 7. Computerized systems have taken
over the analysis of CUSUMs but the concept is easier to understand using the
more visual V-mask method.
Page 18 of 53
Figure 7: Illustration of V-mask placed over lead point to detect change (the
current standard deviation is 4.5 N/mm2)
Consider the data for compressive strength in Table 4 for a plant operating on a
standard deviation of 3.5 N/mm2 and with a target mean strength of
40 N/mm2. The CUSUM can be calculated and is tabulated in Table 5. A plot of
the CUSUM with UCL and LCL is shown in Figure 8. When the V-mask is placed
on the lead point (point 18), the CUSUM crosses the LCL at point 11 indicating
that a change in the process occurred at this point. While the Shewhart
analysis of the data (see 4.4) also indicated a change at point 18, the plot of
the CUSUM gives a clear visual picture of the trend and shows that it has been
present since point 11.
Page 19 of 53
Table 5: CUSUM data

Result 28 day Difference CUSUM
strength from 40N/mm2 (N/mm2)
(N/mm2) target (N/mm2)
1 37 -3 -3
2 42 2 -1
3 36 -4 -5
4 35 -5 -10
5 42 2 -8
6 38 -2 -10
7 39.5 -0.5 -10.5
8 40 0 -10.5
9 35 -5 -15.5
10 40 0 -15.5
11 34 -6 -21.5
12 44 4 -17.5
13 46.5 6.5 -11
14 42 2 -9
15 44.5 4.5 -4.5
16 45 5 0.5
17 44 4 4.5
18 48 8 12.5
Figure 8: Example CUSUM plot for data in Table 5
Page 20 of 53
5.2 Controlling mean strength

Compressive strength of concrete is the simplest parameter to monitor. The
test result is compared with a target mean strength based on the specified
characteristic strength plus the margin (see equation 1).
CUSUMs controlling compressive strength are generally based on a concrete
family and all test results are converted to the equivalent value of a selected
reference concrete, e.g. C32/40. The analysis is usually based on early age test
results (see 7.1), e.g. on 7 day strength data, as the risk of waiting until 28
days to identify a loss of control is unacceptable. The predicted 28 day
strength is calculated from the early age test result (see 7.1) and this is used in
the CUSUM until the actual 28 day strength is available. Some systems then
substitute the measured 28-day strength and re-calculate the CUSUM while
other systems continue to use the predicted strength. The current trend is to
replace the predicted strength with the measured 28 day strength as since 2000
the data have also been used for conformity and conformity is based on the
measured 28 day strength.
A worked example of a CUSUM controlling target mean strength (CUSUM M)
using the concrete family approach is detailed in chapter 11.
5.3 Controlling standard deviation

CUSUM may be used to monitor standard deviation by using the relationship
between the range of successive pairs of results given in equation 2.
The expected range (current standard deviation x 1.128) of a pair of results is
monitored against the actual range and a CUSUM of the differences calculated.
The results are plotted and monitored using a V-mask or computer model.
It should be noted that range is not a normally distributed variable, therefore if
a symmetrical V-mask is adopted; longer run lengths will be required to detect
a decrease in the standard deviation. However this conservative approach –
the use of a symmetrical mask- is adopted for simplicity in many systems.
When a significant change is detected, the standard deviation is adjusted and
results corrected to the new target mean strength – higher if the standard
deviation has increased, lower if the standard deviation has decreased. The
result immediately prior to the change of target strength should also be
adjusted to the new target mean strength to calculate the range of pairs,
otherwise an additional element of variation would be built into the data.
An example CUSUM on standard deviation (CUSUM R) is illustrated in
chapter 11.
5.4 Controlling correlation

Control systems for concrete are generally based on early age strength as the
consequences of waiting for the 28 day compressive strength results lead to an
unacceptable level of risk. Seven day compressive strength results are
normally used in the control system and the 28 day strength is predicted using
an estimated correlation (the relationship between 7 and 28 day strength). In
order to confirm that the correlation factor is correct, a CUSUM C may be run
on the differences between actual and predicted 28 day strengths. If the
Page 21 of 53
CUSUM C is positive then the system is underestimating the 28 day strength and
if negative it is overestimating the 28 day strength.
When a significant trend is detected, a new correlation relationship is
determined. The CUSUM M for mean strength using predicted results will need
to be recalculated as the system has effectively been under or over estimating
for a period of time and it may be significantly adrift. The plot of range need
not be re-determined because the correlation change will affect all results
similarly (except for the range straddling the point of correlation change).
The relationship between 7 and 28 day strength is affected by the cement type
and source, e.g. the strength gain between 7 and 28 days of a concrete made
with a CEMI cement will be less than for an equivalent concreteb made with a
CEMIII/B cement. Concretes with different cement types should therefore be
either controlled by separate control systems or the difference in correlation
between different cement types considered in the corrections that are applied
within the concrete family [14]. Different sources of the same type and class of
cement may have different 7:28 day strength ratios and if a source of cement is
changed, the validity of the current 7:28 day strength ratio should be
reviewed.
It is important to appreciate that calculating the correlation as a simple
straight-line relationship or as a simple percentage addition will not be
accurate at the limits of the mix design. This is because concrete mixes have
ceiling strengths that may be due to a failure of aggregate bond, failure of the
aggregate particles themselves or, as the cement content increases, the
voidage is increasing in proportion, i.e. the (w + air)/c ratio remaining
constant. Therefore at the ceiling strength, increasing the cement content of
the mix or using plasticisers to reduce w/c ratio will have little effect on
compressive strength. As the 28 day strength approaches the ceiling strength,
the ratio between 7:28 day strength will change. Clearly as the 28 day strength
approaches the ceiling strength, there will be a reduced strength gain from 7
to 28 days. Similarly, at lower strengths, the results will tend to converge
towards zero.
The initial correlation may be established in different ways including:
• Initial trial mixes at a range of cement contents;
• Historical data.
Once the initial correlation is established, it needs to be checked routinely to
check that it is still valid. This is the purpose of CUSUM C.
An example CUSUM for correlation (CUSUM C) is shown in chapter 11. There
are fewer data in the correlation CUSUM compared to the CUSUMs for mean
and standard deviation because of the 21 day gap between 7 and 28 day results
becoming available.
In the example, the actual 28 day results have been used in place of the
predicted 28 day data when they become available. Some systems will only use
predicted data. However, all CUSUMs will recalculate the predicted data when
the CUSUM shows a change in the correlation.
b
A concrete designed to produce the same 28 day strength
Page 22 of 53
5.5 Design of V-mask

The design of the mask (i.e. the appropriate gradients and decision intervals) is
based on statistical probabilities and they are linked to the standard deviation
of the plant as illustrated in Figure 9 for a typical CUSUM system.
Figure 9: Design of V-mask
A number of concrete producers have opted with their CUSUM systems to

follow the ‘double-decision approach’ used in Shewhart charts, namely a
warning line and a control limit, and take action when the warning line is
crossed. In this case the warning lines are parallel to the X-axis and offset by
the same decision interval as used in the traditional V-mask. Such an approach
results in action being taken sooner than with the traditional V-mask. As action
is being taken when the warning line is crossed, a control limit serves no
practical purpose and they are usually omitted.

When the CUSUM indicates that the target mean strength or assumed standard
deviation is not being achieved, adjustments will be made to the concrete
production. Often, but not always, this requires a change in cement content or
w/c ratio.
The magnitude of the change required when the CUSUM M indicates need for
action is effectively determined by the number of results over which the
process went out of control (a few points would give a steep gradient and
represent a large change, while conversely many points would indicate a
gradual change and a shallow gradient).
The change in cement content can be calculated using the parabolic equation:
dc = 0.75 x Cmra x [(DI/n) + G]
equation 3
Where:
• dc = change in cement content in kg/m3
Page 23 of 53
• 0.75 is an anti-hunting factor to prevent oscillatory and complementary

changes occurring, i.e. it minimises the risk of over-correcting and
having to apply a second adjustment in the opposite direction
• Cmra is the relationship between strength and cement content from the
main relationship (usually in the range 5 to 6 kg/m3 to give a strength
increase of 1N/mm2)
• DI is the decision interval of V-mask (see Figure 9)
• G is the gradient of V-mask (see Figure 9).
Both DI and G are a function of standard deviation and have units of N/mm2.
This relationship is shown graphically in Figure 10 for a change in the CUSUM M:
the smaller the number of results the greater is the change required in cement
content.
Figure10: Action following change in mean for different SD and run length
A change in the standard deviation would result in a new target mean strength
being adopted and therefore a corresponding increase or decrease in the
cement content of the control mix and all other mixes in the concrete family.
The information from a CUSUM that a change has occurred is only part of the
information available to the concrete producer. The aim of the producer is to
identify the cause of the change and take appropriate action at all relevant
plants. For example if the change is identified as being due to a change in
constituent A, the CUSUMs of all plants using constituent A should be checked
to show if they are showing the same trend. If so it would be prudent to take
action even if the CUSUMs at one or more of these plants have not yet
indicated a change sufficient to cross the control or warning lines.
A change in cement content or w/c ratio is not always required. For example if
the cause of the change is known and it has been corrected, adjustment of the
mix proportions may not be needed. A typical example of this is where the
cement strength is known to have gone down, but subsequently recovered.
Page 24 of 53
6 Multivariable and Multigrade Analysis

Multigrade, multivariable CUSUM was developed by Ken Day [15] as a
development of the univariable CUSUM system discussed in chapter 5.
6.1 Multivariable
Multivariable CUSUM systems not only monitor a single property of concrete,
e.g. compressive strength, but instead simultaneously monitor a number of
properties, for example the Day’s CoNAD system plots;
• Compressive strength;
• Density;
• Temperature;
• Slump.
Early action can be taken before the plot reaches the UCL or LCL because the
downward trend in compressive strength may be accompanied by a downturn in
density and either (or both) an upturn in slump or temperature. If this pattern
occurred over two or three results it indicates a genuine change and a probable
cause (water addition). The temperature, slump and density information is
available immediately and therefore trends can be detected prior to the 7 day
strengths being available. An illustration (without control limits) of a plot of
multivariables is shown in Figure 11.
Figure 11: Plot from Multivariable system
6.2 Multigrade
Multigrade CUSUMs are designed to incorporate concretes from different
compressive strength classes in the same CUSUM system. One approach is to
Page 25 of 53
adopt the family of mixes concept discussed in 7.2. However Day argues that
it is possible to sum differences from the current average value of any variable
in each individual strength class of concrete, as though they were all from the
same average value. In this way results from high or low strength mixes and
even lightweight aggregate mixes can be incorporated into the CUSUM [16].
Day argues [17] that the advantages of this technique are:
1. There is essentially no limit to the range of individual mixes that can be
treated as a single family.
2. There is no requirement for adjustment formulas.
3. There is no requirement for checking that constituent mixes remain as
acceptable family members (except where a change point is detected).
4. As a consequence of the above, change point detection is much more
rapid and multi-variable CUSUMs become more effective in cause
detection.
An overall CUSUM can be plotted incorporating all the results. Additionally,
the data for individual compressive strength classes can be analysed to
determine whether effects are across the range of mixes or are concentrated in
specific compressive strength classes.
As the trend is to hold all information on the computer system and interrogate
it for trends, differences between the operational practice of using CUSUM and
multivariable and multigrade analysis are less than they appear. In both
systems, the key is the operator understanding the information being presented
by the control system, interrogating the data to determine the cause and
taking appropriate action.
Page 26 of 53
7 Speeding the Response of the System

7.1 Early age testing
By using early strength data to predict 28 day strength, an adverse trend will
be detected more rapidly. If seven day test data are used, an adverse trend
will be detected three weeks earlier than waiting for 28 day strength data.
Generally control systems for mean strength are based on 7 day strengths as
the risk of waiting for 28 day test results is unacceptable to both the consumer
and producer. However with certain precast elements, process requirements
lead to strengths that exceed the specified characteristic strength within a few
days and in this case testing may be undertaken at a single age less than
28 days and the process controlled using Shewhart charts. In this case there is
no need to predict 28 day strengths as they already exceed what is specified
and the sophistication of the CUSUM system is an unnecessary complication.
Systems utilising accelerated tests (e.g. 1 or 2 days) have had limited success
[6]. These systems generally entail heating the concrete samples but this
process introduces a greater variability than is found in the normally cured 7
day results.
Testing standard cured specimens at three days may introduce greater
variability than testing at 7 days. This is because at three days the initial
temperature of the concrete, which is controlled by the ambient conditions,
has a greater impact on three day strength compared to 7 day strength. In
climates that are similar to the UK, the 7:28 day strength ratio is likely to be
more stable than a 3:28 day strength ratio and it is the better choice for the
control system. The complications of modern concretes, e.g. additions,
admixtures, are also best handled using a 7:28 day strength relationship.
The ratio between the early strength test data and the 28 day test data for the
same mix is also monitored by a CUSUM to detect any significant change in the
relationship. This is the CUSUM C described in 5.4. However, if one of the
other control systems is used with early strength data, it is necessary to
continually check that the correlation has not changed. This may be done on a
continual basis using, say, the last 10 results or periodically, say, every fifth
result.
The multivariable CUSUM approach utilising data from temperature, density
and slump (discussed in chapter 6) claims to allow early interpretation of data,
i.e. a detection of a change in quality, than is possible with to a single variable
CUSUM.
While predicted 28 day strengths are used to determine if action is needed to
keep the process under control, any conformity assessment is based on the 28
day strength data and not the predicted data.
7.2 Family of mixes concept

Changes in the performance of concrete are usually the result of a step change
in a constituent material, e.g. between deliveries. Such a change will affect all
concrete made with that constituent. A change in the functioning of a plant
will affect all concrete produced in that plant. To detect an adverse change in
Page 27 of 53
quality requires sufficient data, as a real change has to be separated from

normal variability. Controlling each concrete separately will delay the time
taken to detect an adverse change. Utilising the family of concretes approach
enables more test results to be incorporated into the system. This speeds up
the time taken to detect a significant change. For example Figure 5, the
Shewhart analysis of the data in Table 4, shows 7 results above the target value
and this indicates a significant trend. To identify this trend clearly it needs 7
results to be entered into the system. If each compressive strength class of
concrete were monitored separately, it is possible that these 7 results could
have been at 7 different compressive strength classes and therefore each
control system would only have a single result entered, and no trend would
have been detected.
Clearly not all the concrete produced at a ready-mixed concrete plant is the
same compressive strength class; in fact a huge variety of mixes can be
produced. Concretes can be produced with various compressive strength
classes, with different consistences; with and without a variety of admixtures
on their own or in combination; with different aggregate sizes, types and
proportions of fine aggregates; with different minimum cement content and
maximum water cement ratio specifications; with or without fibres; with
different cements and additions etc. The range of mixes produced at a precast
factory will be significantly smaller. Even with only two concrete types being
tested at the same rate, combining the data into a family will halve the time
taken to detect an adverse change.
Each mix may be controlled individually (particularly at a precast works), or by
the specified compressive strength class. Alternatively, EN 206-1 allows similar
concretes to be grouped together into a familyc of mixes and the control
system can be applied to the family.
Annex K of EN 206-1:2000 recommends where there is little experience of the
use of concrete families the following may be included in a family
• cement of one type, strength class and source;
• demonstrably similar aggregates and type I additions;
• concretes with or without a water reducing/plasticizing admixture;
• full range of consistence classes;
• concretes with a limited range of strength classes.
Valid correction factors are established for these variables and for the various
members of the family, to make results comparable, and to enable them all to
be used in a single CUSUM control system.
The starting point for any family is the selection of a consistence, maximum
aggregate size and a set of constituent materials. For this selection the
relationship between strength and cement content (or water/cement ratio) is
determined and this is called the main relationshipd. The main relationship is
usually established by a series of trial mixes at varying cement contents,
c
Defined in EN 206-1[3] as “group of concrete compositions for which a reliable relationship between
relevant properties is established and documented”
d
Defined in the QSRMC regulations [18] as “The relationship established between cement content and
strength. It will normally be the relationship arising from fitting a curve to data arising from a series of trial
concrete mixes”
Page 28 of 53
computer modelling plus a limited number of trial mixes or be determined by

analysing production data. From this relationship, the mix proportions to
achieve any target strength can be interpolated for this set of materials,
consistence and maximum aggregate size. Simple numerical adjustments to the
cement and fine aggregate contents (called secondary relationships) are made
to convert concretes covered by the main relationship to other concretes in the
family or vice versa. The reference concrete to which all other tested
concretes are converted lies on the main relationship.
The main and secondary family relationships are used in two directions. They
are used to determine the mix proportions needed to satisfy a concrete
specification, i.e. for concrete mix proportioning, and they are used to convert
the strength of a tested concrete to the value it would have been if it had been
the reference concrete. These equivalent strength values are used as input
data to control charts.
Where reliable relationships are established to the reference concrete, other
concretes may also be included in the family. With the power of computing,
the trend is to make use of the option of either including or excluding a wide
range of mixes from the family and to analysis the data in different ways. The
multigrade system and CUSUM system are capable of supporting such
sophisticated analyses. For example a number of plants using the same
materials could be included within a family, but while the combined data are
analysed for trends, each plant is also reviewed separately. Even when data
are not combined into families, the data can be examined to confirm trends.
For example if Family A is indicating a particular trend and it is believed to be
due to a change of constituent X, the data from other families/mixes
containing this constituent should also be reviewed for signs of the same trend
and, if appropriate, action.
In general air entrained concrete is not included in a family and it is controlled
as a separate concrete, because of the additional variability caused by
variations in the air content. However if it is being controlled by a CUSUM
system, it is also normal to adjust the strength on the basis of the measured air
content. This is to avoid making unnecessary changes to the mix proportions
when the changes in strength are simply the result of variable air content.
When controlling a family of mixes, the control strength of the reference
concrete (to which other compressive strength classes of concrete are
corrected) is either a concrete which represents the most common mix
produced at a plant or is mid-range in the family. It has to be a concrete on
the main relationship. A number of adjustments may need to be applied to a
test result to ensure that it can be compared to the reference concrete, e.g.
for slump or for the inclusion of a plasticiser. The first steps are to convert the
measured strength of the tested concrete into an equivalent strength of a
concrete on the main relationship and then this equivalent strength is moved
along the main relationship to become the equivalent strength of the reference
concrete. See the CUSUM example in chapter 11.
In practice, the adjustments for consistence and constituents are applied to
the cement content of the tested concrete so that it is deemed to be
equivalent to a concrete on the main relationship in terms of constituents and
consistence. The measured strength of the tested concrete may be above,
Page 29 of 53
below or the same as the strength value on the main relationship for the
adjusted cement content. A further adjustment needs to be made to the
predicted test result to allow for the difference between the strengths at the
adjusted cement content and the cement content of the reference concrete.
This is in order to determine what strength the tested concrete would have
achieved if it had had the same cement content as the reference concrete. It is
this twice adjusted result that is compared with the target mean strength of
the reference concrete, and the difference between these values is then
applied in the CUSUM.
Page 30 of 53
8 Guidance on Control Systems

8.1 Abnormal Results
From time to time, results will occur that are outliers to the bulk of the data.
These results could be due to a number of reasons, for example, a testing error
or misreporting. If an outlier is included in the main data analysis, it could
disproportionately affect the control system and result in actions being taken
to adjust the process that are unnecessary (which could increase the risk to the
user of the concrete). However, except where it is technically justifiable and
documented, each outlier is checked for conformity against the EN 206-1
individual batch criterion.
A general rule adopted is that results more than 3σ (3 standard deviations)
away from the mean should be considered as outliers and therefore excluded
from the analysis. The 3σ value equates to 3 results in a thousand (see Table
2). However if a result that breaches the 3σ limit is followed by one that is
greater than 2σ in the same direction, then both results should be included in
the control system and an immediate investigation undertaken to identify
whether a significant change has occurred.
8.2 Handling mixes outside the concrete family

The control system will provide information on concrete in the family of mixes
that are used within the system. The main relationship for a concrete family
can also be used to control mixes that are outside the family. This can be done
in a number of ways:
• The cement content of non-family concrete can be linked to a
conservative value for that of the control mix, and if the cement content
for the control mix is increased, a review is undertaken of the mix
design of the non-family concrete;
• Establishing a safe cement content differential (may be higher or lower)
to be applied to different cement types or cement blend levels. These
differentials can be established through trial mixes in the laboratory or
from other control systems. This procedure is particularly useful when a
critical concrete is produced infrequently. In effect this increases the
margin between the specified characteristic strength and the target
strength.
8.3 Handling mixes not controlled by compressive strength

requirements
Where mix proportions, particularly cement content per cubic metre, are
controlled by the strength criteria, setting the target strength for the mix is
relatively straightforward. It is the specified characteristic strength plus the
margin. However, strength does not always determine the mix design. For
example a foundation to be placed in aggressive ground conditions may require
a structural strength of only C25/30, but due to the need to resist the
aggressive ground the maximum w/c ratio may be 0.45, which in reality will
produce a strength well above that needed for achieving the specified
Page 31 of 53
characteristic strength. The target strength should be determined from the

higher of the (possibly) different cement contents required to conform to the
specified strength, maximum water/cement ratio, and minimum cement
content.
If this procedure is not followed, i.e. if the target strength is determined from
the specified strength alone, the standard deviation will increase significantly
and the result may appear as an outlier (on the high side). However the most
serious consequence is that an adverse trend may take much longer to cross
the V-mask due to the step change in trend caused by the result.
Example 5
A concrete has been specified as C25/30, maximum w/c ratio of 0.55,
minimum cement content of 300 kg/m3. Assuming a standard deviation of
4 N/mm2 within a family of concretes a C25/30 concrete has a target strength
of 38 N/mm2 (cube). However the target w/c ratio of 0.53 equates to a target
strength of 48 N/mm2 and this higher value should be used as the target.
The measured strength is 44 N/mm2. As there is an existing adverse trend in
the results this result triggers action, Figure 12a). However if the target
strength of this concrete had been incorrectly entered as 38 N/mm2, instead of
the result being 4 N/mm2 below the target strength of 48 N/mm2, it would
appear 6 N/mm2 above the target strength of 38 N/mm2. This incorrect entry
would not trigger action nor would it be identified as an outlier (current
standard deviation is 4 N/mm2 and the actual value is less than 3σ from the
target value, see Figure 12b).
Page 32 of 53
a) Correct strength entry for w/c controlled concrete
b) Incorrect strength entry for w/c controlled concrete

Figure 12: Example of the effects of incorrectly entering the target strength for a
mix controlled by the maximum w/c ratio
For clarity, the transposition of these results to the equivalent value of the
reference concrete has not been described.
8.4 Test rates

It is important that the appropriate level and frequency of testing is
maintained in the control system. If the testing rate is too low, it will take too
long to recognise that the system is no longer in control and there will be an
unacceptable risk that non-conforming product is supplied to the customer. It
would be uneconomic to test every single batch of concrete that is produced,
so a test rate is required to produce a situation where the risk is shared
reasonably between producer and user.
Page 33 of 53
For a given rate of production, as the rate of testing increases the level of
auto-correlation increases. Auto-correlation is the measure of the relationship
between a result and previous results. A high level of auto-correlation indicates
that a result is unlikely to be significantly different from the previous results
(e.g. when two successive deliveries of the same type of concrete are sampled)
and testing at a lower rate will provide the same level of control at a lower
cost. However it has to be recognised that plants with low rates of production
may not be able to achieve the optimal rate of testing. One way of handling
low test rates is to have an additional cement differential, i.e. more cement
than that indicated by the mix proportioning, to protect the user from the risk
associated with low testing rates.
A test rate of about 16 results per month provides enough data to run an
effective control system without an excessive level of auto-correlation.
The effect of auto-correlation is minimised in the calculation of the effect of a
change in the control system by the use of an anti-hunting factor (see 5.5).

An investigation should be undertaken to determine the cause of the change. A
CUSUM system will indicate the approximate time when the system started
going out of control, aiding the investigation.
The over-riding principle is to protect the user of concrete. If the cause of an
adverse change cannot be identified, all concretes in the family need to be
adjusted. The appropriate increase or decrease in control cement content is
established from the control system, the general rule is that upward changes in
cement content are obligatory and downward changes are optional.
As explained in 5.5, the appropriate action may be not to change the mix
proportions, as the cause of the change has been already identified and
corrected, e.g. the weigh scales have been repaired and re-calibrated. Good
practice has been to have a minimum value of standard deviation, typically in
the range 3.0 N/mm2 to 3.5 N/mm2. This means that when the control system
shows the standard deviation is lower than this minimum value it is not
adopted and the higher minimum value is assumed. With single concretes and
short production runs, it is possible to achieve a standard deviation in the order
of 2 N/mm2. However, changes between batches of cement can result in
changes of concrete strength in the order of 1 N/mm2 and this equates of 0.5σ
with a standard deviation of 2N/mm2 and an increased risk of non-conformity.
Having a minimum standard deviation protects the user and producer.
Page 34 of 53
9 EN 206-1 Conformity Rules for Compressive Strength

9.1 Basic requirements for conformity of compressive strength
EN 206-1[3], the European standard for concrete, states that conformity
assessment shall be made on test results taken during an assessment period
that shall not exceed the last twelve months. Conformity of concrete
compressive strength is assessed on specimens tested at 28 dayse by applying
two criteria:
Criterion 1 — applies to groups of n non-overlapping or overlapping
consecutive test results fcmf ;
Criterion 2 — applies to each individual test result fcig .
The criteria are given in Table 6.
Table 6: EN 206-1 Conformity Criteria

Production Number n of test Criterion 1 Criterion 2
results for Mean of n results Any individual
compressive ( fcm) test result ( fci)
strength in the N/mm2 N/mm2
group
Initial 3 ≥ fck + 4 ≥ fck — 4
Continuous Not less than 15 ≥ fck + 1.48σ ≥ fck — 4
Depending upon the shape of the test specimens, the appropriate

characteristic strength, fck, is selected for the specified compressive strength
class. The same criteria apply regardless of whether cylinder or cube strengths
are measured.
Based on criteria defined in EN 206-1, the production is defined as either
‘initial’ or ‘continuous’. Based on this classification of production, conformity
is confirmed if both the criteria given in Table 6 are satisfied.
Where conformity is assessed on the basis of a concrete family (see 7.2),
Criterion 1 is to be applied to the reference concrete taking into account all
transposed test results of the family; Criterion 2 is to be applied to the original
test results, i.e. for conformity of the individual batch, the test result cannot
be less than the characteristic strength associated with the specified
compressive strength class minus 4 N/mm2.
The UK Quarry Products Association (now part of the UK’s Mineral Products
Association) has produced detailed Guidance on the application of the EN 206-1
conformity rules [4], and it is demonstrated that the risk to the concrete
producer from the mean strength (≥ (fck + 1.48σ)) criterion is unacceptably
high for 15 results even when using a production margin of 1.96σ. Such a
margin may be adequate for the individual batch criterion (equivalent to a 3 in
a 1000 chance of failing when the standard deviation is 4.0 N/mm2) but the
e
If the strength is specified for a different age the conformity is assessed on specimens tested at the
specified age
f
fcm is the mean compressive strength of concrete
g
fci is the individual test result for compressive strength of concrete
Page 35 of 53
producer’s risk of failing the mean strength criterion remains high. Simply
increasing the rate of testing to some very high level does not solve the
problem due to the increased level of auto-correlation off-setting the benefit
from the increased number of test data.
9.2 Assessment period

Across Europe, the actual rate of testing in practice is variable. Test rates and
lengths of conformity assessment periods have significant implications for
conformity. At a rate of testing of 15 results a month, it would take about one
month to detect that a change of 0.5σ lower than target had occurred, i.e. for
this month the mean strength would have been, for example (1.96 — 0.5)σ,
which would be a failure of the mean strength criterion if the assessment
period was one month. By taking immediate action, the producer can adjust
the mix proportions, e.g. increase the cement content, to give the expected
target strength, and when the results are averaged over a longer period, the
concrete will be in conformity for the whole of the assessment period. The fact
that the same rules applied to the same data over different assessment periods
can produce different conformity results, does not inspire confidence in the
process.
The statistical basis of the mean strength criterion in EN 206-1 is the
achievement of an AOQL of 5%. Failure to conform to the mean strength
criteria shows that the average outgoing quality limit is more than 5% and not
that the concrete is unfit for purpose. However it is an indication that the
producer needs to take action to achieve an AOQL of 5%
9.3 Conformity rules for compressive strength

If the assessment period selected by the producer is one year and all the
results in a family are used, the risk of non-conformity over the assessment
period is close to zero. However over this year there could be periods where
the production was not achieving the specified characteristic strength, see
Figure 13.
Example 6
This example is based on conformity to the mean strength criterion of a
C25/30 concrete with a target strength is 38 N/mm2 (cube), i.e. a margin of
2σ. The standard deviation is a constant 4 N/mm2. Figure 13 comprises
randomly generated data with a mean strength of 38 N/mm2 except for results
77 to 109 where the mean strength is set at 35 N/mm2 for the number
generation. However the data set was manually adjusted to exclude individual
failures, as in practice such results are declared as non-conforming and
excluded from the analysis of conformity of mean strength, i.e. they no longer
form part of the population claimed to be in conformity .
The running mean of 15 indicates non-conformity between results 86 to 100
and this transposes to non-conformity of concrete represented by test results
((86 — 15) = 71 to 100) yet this period of ‘non-conformity’ is disguised when
the assessment period is one year.
For the purposes of discussion, it is assumed that the data, and the concrete
production, were not subject to production control using one of the control
Page 36 of 53
chart systems described in this publication. If such a control system had been
in operation, the change in mean strength would have been detected and the
production would have been changed well before result 109.
Figure 13: Illustration of the potential problem with using a long assessment
period
Finding a solution that is fair to both consumers and producers is not easy as
the solution has to cover normal, low and high production plants. A proposed
solution is:
For plants with low rates of production of designed concrete
Where the number of test results per three months for designed concrete is not
greater than 35, the assessment period shall comprise at least 15 and not more
than 35 consecutive results taken over a period not exceeding 12 months.
For plants with high rates of production of designed concrete
Where the number of test results per three months for designed concrete
exceeds 35, the assessment period shall not exceed three months.
Such a solution does not resolve all the issues related to conformity of
strength, for example the criteria are based on the assumption that the
concrete mix proportions are controlled by strength and this is not always the
case (see 8.2).
It is also proposed that the use of control charts be accepted as an alternative
to the mean strength rule. This is conditional on the concrete being subject to
third party certification or an agreement between the parties. As a control
chart comprises successive sampling plans (with a known standard deviation),
the operating-characteristic curve of the individual sampling plan may be
established. The average outgoing quality (AOQ) curve is then determined by
multiplying each percentage of all possible results below the required
characteristic strength in the production by the corresponding acceptance
probability.
Page 37 of 53
Section 4.4 shows how this approach can be applied to control systems based
on the Shewhart chart and 9.4 shows how it may be applied with the CUSUM
system.
9.4 Achieving an AOQL of 5% with CUSUM

Caspeele and Taerwe [5] have developed a system where for a selected margin
an upper limb V-mask for CUSUM M is used to determine when an AOQL of 5% is
not being achieved. Parameters for the V-mask are given for n =15 and 35 and
for independent and auto-correlated results. Production data show that
concrete test results have some auto- correlation and therefore these are the
appropriate values. For the reasons given in 9.2, the producer should select an
assessment period based on 35 results. The paper [5] is focussed on achieving
an AOQL of 5% and it includes margins less than 1.64σ. The use of a margin less
than 1.64σ means that the producer is deliberately targeting the production
not to achieve the specified characteristic strength. This is unacceptable and
the margin should never be less than 1.64σ. Table 7 gives the V-mask
parameters for selected margins based on n =35 and auto-correlated results.
Table 7: Selected conformity V-mask parameters

Margin Decision interval Slope
1.64σ 3σ σ/2
1.66σ 8σ σ/4
1.70σ 9σ σ/4
1.71σ 4σ σ/2
1.74σ 10σ σ/4
1.76σ 5σ σ/2
1.82σ 6σ σ/2
1.86σ 7σ σ/2
1.91σ 8σ σ/2
1.95σ 9σ σ/2
1.99σ 10σ σ/2
2.06σ 2σ σ/1
2.26σ 5σ σ/1
2.49σ 10σ σ/1
The conformity V-mask only applies to the upper limb, i.e. to the actual
strength being less than the target strength, and its length is the selected value
of n; 35 in the case of Table 7. In effect, the V-mask tests whether the last n
results have achieved an AOQL of 5%. Figure 14 for CUSUM M shows the
conformity V-mask with the V-mask that shows when a significant change has
happened (see 5.4).
Page 38 of 53
Figure 14: Conformity and action V-masks for a target strength of (fck + 1,96σ)
When the CUSUM crosses the significant change line, the producer takes
immediate appropriate action. However there will be a number of test
specimens that have been cast but not tested. If a conformity V-mask is being
used, the CUSUM M should be re-set once these results have been tested. This
means that the adverse trend may move towards the conformity line. At a test
rate of 16 per month and the use of 7-day strength data, there would be about
4 results between crossing the significant change line and re-setting CUSUM M
and therefore the probability of crossing the conformity line is very low.
A decade of experience using CUSUM with the target strength and significant
change limits given in Figure 14 and checking conformity of mean strength to
EN 206-1:2000 using a running mean of 35 results [19] has never resulted in
non-conformity of mean strength. Under these conditions the conformity V-
mask is highly unlikely to be crossed and if it was crossed, it would indicate
that the producer must take action to achieve the target strength, but the
producer would have taken this action already, i.e. when the significant change
line was crossed.
9.5 Non-conformity
If the control system shows that an AOQL of 5% is not being achieved, the
producer is required to take immediate action to achieve the target strength.
In addition the producer should identify any concretes that are not fit for
purpose and inform the user and specifier. The certification body will check
that the investigation was undertaken in the appropriate way and the users and
specifiers correctly informed.
Page 39 of 53
10 Implementing Control Systems

Once continuous production is established the producer should have the choice
of controlling the mean strength using control charts. The use of control charts
should be limited to producers with third party certification (e.g. the majority
of the ready-mixed concrete and precast concrete industries in Western
Europe) or to where there is agreement between the producer and user.
The following are recommended minimum requirements on the control system:
- achieve a maximum average outgoing quality (AOQ) not exceeding 5,0%;
(Comment: This ensures that no more than 5% of the production is
below the specified characteristic strength).
- aim to ensure conformity of the production with the required
characteristic strength;
- include regular monitoring of strength and standard deviation and
deviations from target values;
- where applicable, include one or more procedures for speeding the
response of the system (e.g. use of early strength data, use of concrete
families);
- define and apply clear decision rules for conformity and warning limits;
- document how the system achieves a maximum average outgoing quality
(AOQ) not exceeding 5.0% (unless one of the rules of application given in
an informative annex to EN 206-1 is used);
- when the control chart shows that the standard deviation is ≥ 0.5 N/mm2
above the currently applied value, change the applied value.
Page 40 of 53
11 CUSUM Example
11.1 Reference mix and concrete family
The following example is used to simply illustrate the CUSUM process applied to
a concrete plant controlling its production based on a family of mixes, as is the
common practice in the UK.
The control system is based on a reference concrete described in Table 8. This
concrete is representative of the main concrete produced at the plant. The
control cement content is the current level that the CUSUM identifies as
necessary to produce the target strength of the reference concrete. Note that
concrete with the same compressive strength class as the reference concrete
might be actually produced with higher cement contents than the reference
concrete. For example, specification requirements for durability may show that
the minimum cement content or the cement content to satisfy the specified
maximum w/c ratio is higher than the CUSUM control cement content of the
reference concrete. How these mixes are handled is described in 8.3.
Table 8: Reference Mix Details

Compressive strength C32/40
Aggregate size and type 20mm gravel
Cement Type CEM III/A
Slump 70mm (S2)
Admixture None
Control cement content 320 kg/m3
The control cement content is that expected to give the target strength of the
reference concrete. Not all concretes will be included in the family of mixes
used to control the main production. The parameters of the family are shown
in Table 9.
Table 9: Family parameters

Compressive strength C16/20 to C45/55 inclusive
Aggregate size and type Gravel only 20mm or 10mm
Cement Type CEM III/A only
Slump 25mm to 150mm inclusive
Admixture With or without water reducing admixture
11.2 Main relationship

A key relationship needed in the CUSUM analysis is the main relationship
between cement content and strength (Figure 15), which not only allows mixes
to be transposed to an equivalent to the reference concrete, but is also used to
determine the size of correction to be applied when the CUSUM indicates
change has occurred (see 5.6).
Page 41 of 53
80
70
60
Strength (N/mm2)
50
40
30
20
10
0
0 100 200 300 400 500 600
Cement Content
Figure 15: Main relationship between cement content and strength
In addition to the main relationship, relationships are also needed to be

established for the effect on cement content of changing the aggregate size,
slump and the effect of the water-reducing admixture (WRA). These effects
are expressed as adjustments and are made to the cement content. The
magnitude of these adjustments will normally be established by laboratory
work. The values for the family of mixes in this example are shown in Table
10. The adjustments are those needed to convert the cement content of the
tested concrete into an equivalent cement content on the main
strength/cement content relationship (Figure 15). For example a 10mm
maximum aggregate size concrete would have a higher cement content than a
20mm maximum aggregate size concrete (the aggregate size used in the main
relationship) and so a 10mm concrete has to be adjusted by reducing the
cement content used to produce that concrete to what it would have been if a
20mm concrete had been produced. If a WRA is used, the cement content is
adjusted upwards. In this example the WRA is added as a constant percentage
by mass of cement.
Once these changes have been made to adjust the actual cement content of
the tested concrete to the equivalent value on the main relationship, a second
adjustment is needed to further correct that new cement content to that of
the reference concrete. This strength adjustment is the difference in strength
on the main relationship between that at the equivalent cement content of the
tested concrete and the target mean strength of the reference concrete.
Page 42 of 53
Table 10: Adjustments per cubic metre to convert the cement content of
the tested concrete to a concrete on the main relationshipA)
Adjustments where the tested concrete contained a WRA
Cement content of 200 to 380 380+
tested concrete, kg/m3
Adjustment to cement At these higher cement contents it is

content, kg/m3 +25 kg more effective to use a
superplasticizing admixtureB).
Adjustments where the tested concrete contained 10mm maximum aggregate sizeC).
Cement content of 200 to 380 380+
tested concrete, kg/m3
Adjustment to cement —10

—15
content, kg/m3
Adjustments where the tested concrete has a target slump that is not 70mm
Target
20 (S1) 50 70 (S2) 100 120 (S3) 150
slump, mm
Slump
+15 kg +10 kg 0 —5 kg —10 kg —15 kg
adjustment
A)
Numerically equal but opposite in sign (+ becomes -) adjustments are used in mix
proportioning.
B)
As the following example does not need this adjustment for superplasticizer, a value is not
given.
C)
As the following example does not include aggregates larger than 20mm, no adjustments are
given.
11.3 Applying adjustments

Consider the following concretes tested for inclusion in the CUSUM system
Mix Ref 1
C25/30 20mm aggregate CEM III/A 100 mm slump no WRA
The mix was produced with a cement content of 275 kg/m3. This mix meets the
family criteria in Table 9.
To relate the mix to the reference concrete described in Table 8, the mix must
first be adjusted for the fact it was specified as 100mm slump rather than
70mm slump. Table 10 indicates that for 100mm slump an adjustment of —5
kg/m3 is required. The adjustment is negative as additional cement would
have been added to the mix to maintain the w/c ratio after the consistence
had been increased from 70mm to 100mm slump.
The adjusted cement content becomes: 275 — 5 = 270 kg/m3
From the main relationship in Figure 15, we would expect a 270kg/m3 cement
content to achieve a strength of 37.3hN/mm2 but the reference concrete is
h
This is a value calculated from the equation of the line in Figure 15.
Page 43 of 53
C32/40. As the plant has a standard deviation of 3.5N/mm2 and a design

margin of 2.0σ, the target mean strength of the reference concrete is
47 N/mm2. A difference of 9.7 N/mm2 (47 — 37.3 N/mm2) in the CUSUM system
has to be introduced to transpose the strength at the equivalent cement
content to the target mean strength of the reference mix. For example in
Table 11 mix reference 1, the predicted and actual 28-day strengths are 42.5
N/mm2 and 39.5 N/mm2 respectively. After adjustment to equivalent cement
content values of the reference concrete, these become 52.2 N/mm2 (42.5 +
9.7) and 49.2 N/mm2 and the change in CUSUM M is 52.2 — 47 = 5.2 N/mm2 if
the predicted strength is being used and 49.2 — 47 = 2.2 N/mm2 when the
predicted 28-day strength is replaced with the actual 28-day strength.
In this example the lowest compressive strength class in the concrete family is
C16/20. However, if a family were to include concrete with strength classes
lower than C16/20, i.e. a target mean strength of less than 27 N/mm2, any
results are re-calculated using the standard deviation obtained from Figure 3.
Mix Ref 2
C32/40 20mm aggregate CEM III/A 150mm slump with a WRA
The mix was produced with a cement content of 310kg/m3. This mix meets the
family criteria in Table 9.
To relate the mix to the reference concrete described in Table 8, the mix must
first be adjusted for the fact that it contained a WRA; and then adjusted
because it was specified as 150mm slump rather than 70mm slump. For the
WRA the cement content for CUSUM purposes will need to be increased to
correct for the water reduction effected by the addition of the WRA. For the
higher-than-reference slump the adjustment in cement content is negative.
Total adjustment to apply is +25 —15 = 10kg (from Table 10)
The adjusted cement content becomes:
310 + 10 = 320 kg/m3
Note when undertaking mix proportioning, the adjustments from the main
relationship always have the opposite sign, but the same numerical value.
The second adjustment is to correct the recorded strength at this increased
cement content to the strength expected at the cement content of the
reference mix. From the main relationship in Figure 15, a 320 kg/m3 cement
content is expected to achieve a strength of 46.8hN/mm2 but the reference
concrete is C32/40. As the plant has a current standard deviation of 3.5N/mm2
and a design margin of 2.0σ, the target mean strength of the reference
concrete is 47 N/mm2, a difference of 0.2 N/mm2, which is the adjustment that
will be made to the predicted and actual cube strengths in the CUSUM system.
Mix Ref 3
C32/40 20mm aggregate CEM III/A 70mm slump no WRA
The mix was produced with a cement content of 320 kg/m3. This mix is the
reference concrete and as it was batched at the control cement content, no
adjustments need to be applied.
These concretes are the first three included in the CUSUM analysis tabulated in
Table 11.
Page 44 of 53
11.4 CUSUM calculation

Once the adjustments have been made and the adjusted 28 day strength
calculated, the data may be used in any control system (e.g. CUSUM or
Shewhart). The following example analyses the data by CUSUM techniques. A
CUSUM is run on mean (CUSUM M), standard deviation (CUSUM R) and
correlation (CUSUM C). For control purposes the mixes include a prescribed
concrete (P300, sample reference 13) and a nominal mix (1:2:4, sample
reference 14).
The results are plotted on the CUSUM and the V-mask overlaid on each result.
This is a manual system used to illustrate the principles of the CUSUM
technique. Clearly, this process would normally be carried out on computer,
either through a spreadsheet, a commercially available CUSUM programme or
by the development of a company specific computer programme.
The current plant standard deviation is 3.5 N/mm2 which gives a target range
of (1.128 x 3.5) = 3.9 N/mm2 (see 2.3). The margin is 2.0σ. The target cube
strength of the reference concrete is (40 + (2.0 x 3.5)) = 47 N/mm2.
From the three plots in Figure 16, Figure 17 and Figure 18 it may be seen that
the correlation and standard deviation are running in control, but there has
been a change in the mean from point 7 to point 17. This is evidence that the
current control cement content of 320 kg/m3 is not giving the required average
strength of 47 N/mm2. As it is the upper arm of the V-mask that has been
crossed, this indicates that the average mean strength is lower than the target
strength. The mix proportions are immediately adjusted (see 11.5) and CUSUM
M is reset at zero. Note that this change has been made on the basis of a set of
actual 28 day strengths (data points 1 to16) and predicted 28 day strengths
(data point 17).
Page 45 of 53
Table 11: CUSUM calculation

Mix description (all CEM III/A) Results Adjustments CUSUM M CUSUM R CUSUM C
Reference strength
Expected strength
Actual — predicted
Adjusted strength
Cement/Strength
Adjusted cement
Predicted 28 day
Batched cement
Difference from
Difference from
From predicted
Target strength
Aggregate size
Mix reference
Strength class
Actual 28 day
Total cement
Target Range
Target slump
28 day
Actual 7 day
From actual
adjustment
adjustment
Plasticiser
CUSUM M
CUSUM R
CUSUM C
Strength
content
content
Range
target
target
Code
CURRENT STANDARD DEVIATION = 3,5 N/mm2; TARGET RANGE = 1,128 x 3,5 = 3,9 N/mm2
1 C25/30 20 100 No 275 31.1 42.5 39.5 -5 270 A 37.3 40 47 9.7 49.2 49.2 2.2 2.2 -3.0 -3.0
2 C32/40 20 150 Yes 310 33.8 45.3 46.3 10 320 A 46.8 40 47 0.2 46.5 46.5 -0.5 1.7 2.7 3.9 -1.2 -1.2 1.0 -2.0
3 C32/40 20 70 No 320 35.2 46.8 46.8 0 320 A 46.8 40 47 0.2 47.0 47.0 0.0 1.7 0.5 3.9 -3.4 -4.6 0.0 -2.0
4 C32/40 20 70 No 320 37.2 48.8 49.3 0 320 A 46.8 40 47 0.2 49.5 49.5 2.5 4.2 2.5 3.9 -1.4 -6.0 0.5 -1.5
5 C25/30 20 70 Yes 245 26.7 37.5 39.5 25 270 A 37.3 40 47 9.7 49.2 49.2 2.2 6.4 0.3 3.9 -3.6 -9.6 2.0 0.5
6 C32/40 20 150 Yes 310 41.5 52.8 53.8 10 320 A 46.8 40 47 0.2 54.0 54.0 7.0 13.4 4.8 3.9 0.9 -8.7 1.0 1.5
7 C32/40 20 70 No 320 42.6 53.8 53.3 0 320 A 46.8 40 47 0.2 53.5 53.5 6.5 19.9 0.5 3.9 -3.4 - -0.5 1.0
12.1
8 C28/35 20 50 No 285 28.2 39.2 39.2 10 295 A 42.1 40 47 4.9 44.1 44.1 -2.9 17.0 9.4 3.9 5.5 -6.6 0.0 1.0
9 C28/35 20 50 No 285 30.9 42.2 40.7 10 295 A 42.1 40 47 4.9 45.6 45.6 -1.4 15.6 1.5 3.9 -2.4 -9.0 -1.5 -0.5
10 C40/50 20 120 Yes 360 40.4 51.8 48.8 15 375 A 57.3 40 47 - 38.5 38.5 -8.5 7.1 7.1 3.9 3.2 -5.8 -3.0 -3.5
10.3
11 C25/30 20 100 No 275 27.6 38.6 40.5 -5 270 A 37.3 40 47 9.7 50.2 50.2 3.2 10.3 11.7 3.9 7.8 2.0 1.9 -1.6
12 C25/30 20 70 Yes 245 24.1 34.5 35.0 25 270 A 37.3 40 47 9.7 44.7 44.7 -2.3 8.0 5.5 3.9 1.6 3.6 0.5 -1.1
13 P300 20 150 Yes 300 26.2 36.9 37.4 10 310 A 44.9 40 47 2.1 39.5 39.5 -7.5 0.5 5.2 3.9 1.3 4.9 0.5 -0.6
14 1:2:4 20 70 No 270 27.6 38.6 37.6 0 270 A 37.3 40 47 9.7 47.3 47.3 0.3 0.8 7.8 3.9 3.9 8.8 -1.0 -1.6
15 C40/50 20 120 Yes 360 38.3 49.8 47.3 15 375 A 57.3 40 47 - 37.0 37.0 - -9.2 10.3 3.9 6.4 15.2 -2.5 -4.1
10.3 10.0
16 C40/50 20 120 Yes 360 41.5 52.8 53.8 15 375 A 57.3 40 47 - 43.5 43.5 -3.5 - 6.5 3.9 2.6 17.8 1.0 -3.1
10.3 12.7
17 C25/30 20 100 No 275 21.7 31.5 -5 270 A 37.3 40 47 9.7 41.2 41.2 -5.8 - 2.3 3.9 -1.6 16.2
18.5
TARGET STRENGTH NOT BEING ACHIEVED; CEMENT CONTENT INCREASED; CUSUM M RESET TO ZERO
Page 46 of 53
Figure 16: CUSUM M
Figure 17: CUSUM R
Figure 18: CUSUM C
Page 47 of 53
11.5 CUSUM action following change

The CUSUM M has shown that there has been a decline in the performance;
therefore to bring the process back into control it is necessary to increase
the control cement content. The magnitude of the increase in cement
content is a function of the standard deviation of the plant, and the number
of results over which the change has taken place (see 5.6).
In this case, the plant standard deviation is 3,5 N/mm2 and the change
occurred at Mix reference 7, but the CUSUM M first crosses the V-mask at
Mix reference 9 giving a change over 9 results. From Figure 19 it can be
seen that a change over 9 results gives a change in cement content of
14 kg/m3. For simplicity, this would be rounded to 15 kg/m3 and therefore
the control cement content of the reference concrete would be increased
from 320 kg/m3 to 335 kg/m3.
A new main relation would also be adopted that relates to a control cement
content of 335 kg/m3 for a characteristic strength of 40 N/mm2 (target
strength 47 N/mm2). Table 12 shows the relationships in tabular form and
from Table 13, this can be seen as a change in cement/strength code from A
to B. The cement contents actually used at the plant would immediately be
increased to the amount shown by the new main relationship. This changed
main relationship will also lead to revised adjustments being applied to
obtain the predicted cube strength of the reference concrete. These
adjustments are applied from result 18 onwards (Table 13). They will also
be applied to the batching of new mixes, but there will be a period where
the concrete has already been batched at a cement content that is less than
that now known to be necessary. However for the control of the mean
strength, mix 18 onwards is adjusted to the new target cement content of
the reference concrete (335 kg/m3).
Page 48 of 53
25
Change in cement content kg/m3 20
15
SD = 3.5 Nmm2
10
0
0 5 10 15 20 25 30 35 40 45 50 55 60
No of results
Figure 19: Strength change against number of results
Table 12: Relationships between strength and cement content

Cube Cement content, kg/m3, for Cube Cement content, kg/m3, for
strength cement/strength codes strength cement/strength codes
N/mm2 A B N/mm2 A B
20 180.0 195.0 41 290.0 305.0
21 185.0 200.0 42 295.0 310.0
22 190.0 205.0 43 300.0 315.0
23 195.0 210.0 44 305.0 320.0
24 200.0 215.0 45 310.0 325.0
25 205.0 220.0 46 315.0 330.0
26 210.0 225.0 47 320.0 335.0
27 215.0 230.0 48 325.0 340.0
28 220.0 235.0 49 330.0 345.0
29 225.0 240.0 50 335.0 355.0
30 230.0 245.0 51 340.0 360.0
31 235.0 255.0 52 345.0 365.0
32 240.0 260.0 53 355.0 370.0
33 245.0 265.0 54 360.0 375.0
34 255.0 270.0 55 365.0 380.0
35 260.0 275.0 56 370.0 385.0
36 265.0 280.0 57 375.0 390.0
37 270.0 285.0 58 380.0 395.0
38 275.0 290.0 59 385.0 400.0
39 280.0 295.0 60 390.0 405.0
40 285.0 300.0
11.6 Further data and a change in standard deviation

Table 13 is the continuation of the CUSUM calculation with additional data.
During this period no more actual 28-day strength are available. The CUSUMs
with the additional data are shown in Figures 20 to 22.
Page 49 of 53
Following the change in cement content to achieve the target strength,

sample number 18, C32/40 70mm slump, which is the control mix and
therefore batched at the control cement content has previously not
required a correction to the strength (See Table 11, samples 3, 4 and 7).
However, the control cement content has now increased to 335 kg/m3; since
the mix was batched at 320 kg/m3 before the CUSUM M detected the need
for a change, an adjustment from the new main relation needs to be
applied.
The same concrete was batched at sample number 22 but in this case the
control mix cement content changes needed to compensate for the changes
in mean strength and in standard deviation have already been implemented.
For this reason there is no adjustment to make as the batched cement
content is now 345 kg/m3 (320 +15 +5)(+15 kg/m3 necessitated by the
change in strength/cement content relationship and 5 kg/m3 necessitated by
the increase in standard deviation, see Table 12).
The range over the adjustment between samples 17 and 18 in compressive
strength is large (56.3 — 41.2) = 15.1 N/mm2. The results immediately
before and after the change of mean strength are corrected on different
main relationships which will increase variability. Consequently an
amendment needs to be made to the result immediately prior to the change
mean strength in order to not introduce excessive variation into the CUSUM
R calculation. The result before the change of mean strength is adjusted
using the new main relation for the range calculation only. From the new
main relationship the expected strength is 44.2 N/mm2 and this reduces the
range from 15.1 N/mm2 to 12.1 N/mm2.
After Mix reference 18, a change in the plant standard deviation is also
detected, see Figure 21. A new standard deviation needs to be calculated
from the average current range. The average range is 5.3 N/mm2 and the
new standard deviation is 4.7 N/mm2 (5.3/1.128). To avoid over-correcting,
a decision is taken to change the standard deviation to 4.0 N/mm2. The
margin is increased to 1.96 x 4.0 = 7.8 N/mm2 rounded to 8 N/mm2. A
1 N/mm2 increase in the margin requires a 5 kg/m3 increase in the cement
content, see Table 12. The current control mix cement content is therefore
immediately increased from 335 kg/m3 to 340 kg/m3. The cement/strength
relationship is unchanged (relationship B); what has changed is the target
strength of the reference concrete has moved from 47 to 48 N/mm2.
Page 50 of 53
Table13: CUSUM table continued
Mix description (all CEM III/A) Results Adjustments CUSUM M CUSUM R CUSUM C
Reference strength
Expected strength
Actual — predicted
Adjusted strength
Cement/Strength
Adjusted cement
Predicted 28 day
Batched cement
Difference from
Difference from
From predicted
Target strength
Aggregate size
Mix reference
Strength class
Actual 28 day
Total cement
Target Range
Target slump
28 day
Actual 7 day
From actual
adjustment
adjustment
Plasticiser
CUSUM M
CUSUM R
CUSUM C
Strength
content
content
Range
target
target
Code
16 C40/50 20 120 Yes 360 41.5 52.8 53.8 15 375 A 57.3 40 47 - 43.5 43.5 -3.5 - 6.5 3.9 2.6 17.8 1.0 -3.1
10.3 12.7
17 C25/30 20 100 No 275 21.7 31.5 -5 270 A 37.3 40 47 9.7 41.2 41.2 -5.8 - 2.3 3.9 - 16.2
18.5 1.6
TARGET STRENGTH NOT BEING ACHIEVED; CEMENT CONTENT INCREASED; CUSUM M RESET TO ZERO
17 Adjusted 21.7 31.5 270 B 34.3 40 47 12.7 44.2 44.2 0.0 16.2
18 C32/40 20 70 No 320 41.8 53.1 0 320 B 43.8 40 47 3.2 56.3 56.3 9.3 9.3 12.1 3.9 8.2 24.4
STANDARD DEVIATION INCREASED TO 4,0 N/mm2; TARGET STRENGTH INCREASED TO 48 N/mm2; TARGET RANGE INCREASED TO 4,5 N/mm2; CEMENT CONTENT INCREASED
18 Adjusted 41.8 53.1 320 B 43.8 40 48 4.2 57.3 57.3 9.3 0.0
19 C25/30 20 100 No 290 26.2 36.9 -5 285 B 37.2 40 48 10.8 47.7 47.7 -0.3 9.0 9.6 4.5 5.1 5.1
20 C28/35 20 50 No 305 28.6 39.7 10 315 B 42.9 40 48 5.1 44.8 44.8 -3.2 5.8 2.9 4.5 - 3.5
1.6
21 P300 20 150 Yes 300 24.4 34.8 10 310 B 41.9 40 48 6.1 40.9 40.9 -7.1 -1.3 3.9 4.5 - 2.9
0.6
22 C32/40 20 70 No 340 39.5 51.0 0 340 B 47.6 40 48 0.4 51.4 51.4 3.4 2.1 10.5 4.5 6.0 8.9
Page 51 of 53
Figure 20: CUSUM M with additional data
Figure 21: CUSUM R with additional data
Figure 22: CUSUM C with additional data
Page 52 of 53
12 References
[1] SHEWHART, W A, Economic Control of Quality of Manufactured Product, 1931.
[2] CEN, Product standards and conformity assessment guideline, CEN/BT N6287, 2000-11-
09.
[3] BRITISH STANDARDS INSTITUTION, Concrete – Part 1: Specification, performance,
production and conformity, EN 206-1:2000.
[4] HARRISON, T et al, Guidance on the application of the EN 206-1 conformity rules,
Quarry Products Association, April 2001 (available for free downloading from the MPA-
BRMCA web site).
[5] CASPEELE, R and TAERWE, L, Combined production and conformity control of
concrete with acceptance CUSUM control charts, 2010.
[6] DEWAR, J D and ANDERSON, R, manual of Ready-Mixed concrete, 2nd edition, 2003.
[7] INTERNATIONAL STANDARDS ORGANISATION, Shewhart control charts,
ISO8258: 1991.
[7] INTERNATIONAL STANDARDS ORGANISATION, Acceptance control charts,
ISO7966: 1993.
[8] NEWMAN and CHOO, Advanced Concrete Technology – Testing and Quality, Chapter
9, 2003
[9] BRITISH STANDARDS INSTITUTION, Guide to data analysis and quality control
using CUSUM techniques, BS 5703:2003
[10] BROWN, B V, Monitoring concrete by the CUSUM system, Concrete Society Digest
No.6, 1984.
[11] INTERNATIONAL STANDARDS ORGANISATION, Cumulative sum charts –
Guidance on quality control and data analysis using CUSUM techniques, ISO/TR 7871:1997.
[12] BRITISH STANDARDS INSTITUTION, Process control using quality control chart
methods and CUSUM techniques, BS5700.
[13] NIST/SEMATECH, e-Handbook of Statistical Methods,
http://www.itl.nist.gov/div898/handbook/, June 2008-06-18.
[14] SEAR, L, A combined CUSUM system for controlling OPC and OPC/PFA concretes,
ACT Project 89/10.
[15] DAY, K W, Concrete mix design, quality control and specification, 2nd edition, 1999.
[16] DAY, KW, Multigrade, multivariable CUSUM quality control, 8th CANMET/ACI
International Conference on advances in concrete technology, Montreal, May 2006.
[17] DAY, KW, Concrete mix design, quality control and specification, 3rd edition, 2007.
[18] QUALITY SCHEME FOR READY MIXED CONCRETE, Quality and product
conformity regulations,2003.
[19] TROY, J, Private communication, April 2010.
Page 53 of 53

Use of Control Charts in The Production of Concrete: by Ian Gibb and Tom Harrison October 2010

Uploaded by

Copyright:

Available Formats

Use of Control Charts in The Production of Concrete: by Ian Gibb and Tom Harrison October 2010

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Use of Control Charts in The Production of Concrete: by Ian Gibb and Tom Harrison October 2010

Uploaded by

Copyright:

Available Formats

Use of control charts in the production of concrete

Ian Gibb and Tom Harrison

© 2010 The authors and MPA/BRMCA - ERMCO

AOQ Average outgoing quality

AOQL Average outgoing quality limit

fci Individual test result for compressive strength of concrete

fcm Mean compressive strength of concrete

σ Estimate for the standard deviation of a population

fck Specified characteristic compressive strength

qn Statistical constant that depends upon n and the selected AOQL

s Sample standard deviation

UCL Upper Control Limit

UWL Upper Warning Limit

LWL Lower Warning Limit

LCL Lower Control Limit

dc Change in cement content

Mean value of ’n’ test results

2 Statistics for Concrete

Figure 1: Illustration of concrete strength distribution

2.2 Characteristic strength and target strength

A concrete strength below the characteristic strength is not a failure as

Table 2: Effect of margin on proportion of concrete below characteristic

The definition of ‘characteristic strength’ in EN 206-1:2000 has its

2.3 Standard deviation

due to a change in a constituent) are more common than drifts in mean

Table 3: Calculation of the standard deviation using mean range

Example 3 (copied from reference [4])

3.6 N/mm2 when determined by the standard method;

To illustrate the effect of a change in mean strength on the standard

4.4 N/mm2 when determined by the standard method;

The true standard deviation of a population, σ, can only be determined if all

the population standard deviation. Prior to obtaining the estimated population

2.4 Setting the target strength

requirements, the requirements of a certification body or other requirements

3 Simple Data Charts

Figure 4: Simple univariate control chart for strength

4.2 Shewhart action criteria

4.2.2 Points beyond UWL or LWL

4.2.3 Patterns within control limits

4.3 Control of standard deviation

4.4 Example Shewhart chart

Figure 5: Control levels applied to data

4.5 Modified application of Shewhart control chart

Figure 6a): Control of conformity of mean strength using a running mean of

Table 5: CUSUM data

Figure 8: Example CUSUM plot for data in Table 5

5.2 Controlling mean strength

5.3 Controlling standard deviation

5.4 Controlling correlation

5.5 Design of V-mask

Figure 9: Design of V-mask

A number of concrete producers have opted with their CUSUM systems to

5.6 Action following change

• 0.75 is an anti-hunting factor to prevent oscillatory and complementary

6 Multivariable and Multigrade Analysis

Figure 11: Plot from Multivariable system

7 Speeding the Response of the System

7.2 Family of mixes concept

quality requires sufficient data, as a real change has to be separated from