Professional Documents
Culture Documents
EEMUA 175 - Code of Practice, Cal & Valid Process Analysers - 2013
EEMUA 175 - Code of Practice, Cal & Valid Process Analysers - 2013
PUBLICATION 175
Edition 2
About EEMUA
EEMUA Aims
The income and resources of EEMUA are applied solely towards the promotion of
the aims of the Association. No profit is transferred by way of dividend, bonus or
otherwise to members. The work of the Association is mainly carried out by
members’ representatives on EEMUA Technical Committees, Working Groups, and
Forums, supported by EEMUA staff.
For the purposes of conducting its business, EEMUA is incorporated in England and
Wales as a company, limited by guarantee, not having share capital. All usual
aspects of company management and business administration are dealt with by
the EEMUA executive directors, including policy implementation and leadership of
the executive team (the staff).
A list of EEMUA publications for sale is given at the end of this Publication. The full
list is also on the Association's website, including details of on-line shopping
facilities.
ii
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
Legal Aspects
All rights, title and interest in this Publication belong to EEMUA. All rights are
reserved. No part of this Publication may be reproduced, stored in a retrieval
system, or transmitted in any form or by any means: electronic, mechanical,
photocopying, recording or other, without the express prior written agreement of
the EEMUA Chief Executive. Infringement of copyright is illegal and also reduces
the Association's income thereby jeopardising its ability to fund the production of
future publications.
It has been assumed in the preparation of this Publication that: the user ensures
selection of those parts of its contents appropriate to the intended application; and
that such selection and application is correctly carried out by appropriately
qualified and competent persons for whose guidance this Publication has been
prepared. EEMUA and the individual members of the Work Group that prepared
this EEMUA Publication do not, and indeed cannot, make any representation or
give any warranty or guarantee in connection with material contained herein.
EEMUA and the individual members of the Work Group that prepared this EEMUA
Publication expressly disclaim any liability or responsibility for damage or loss
resulting from the use or reliance on this Publication. Any recommendations
contained herein are based on the most authoritative information available at the
time of writing and on current good engineering practice, but it is essential for the
user to take account of pertinent subsequent developments and/or legislation.
Any person who encounters an inaccuracy or ambiguity when making use of this
Publication is asked to notify EEMUA without delay so that the matter may be
investigated and appropriate action taken.
Comments are considered by the relevant EEMUA Technical Committees and may
be incorporated in future editions of this Publication. New editions are publicised
on the EEMUA website
iii
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
iv
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
Contents
Preface .................................................................................................... ix
1. Scope: calibration and validation .......................................................... 1
2. Definitions ............................................................................................ 3
2.1 Equipment ......................................................................................... 3
2.2 Measurement ..................................................................................... 3
2.3 Calibration and validation .................................................................... 4
2.4 Statistical terms ................................................................................. 5
3. General considerations: calibration and validation................................ 9
4. Preconditions for analyser calibration and validation ......................... 11
4.1 General ........................................................................................... 11
4.2 Factory Acceptance Testing (FAT) ....................................................... 11
4.3 Site installation ................................................................................ 11
4.4 Site Acceptance Testing (SAT) ............................................................ 12
4.5 Site precision ................................................................................... 12
4.6 Read-out instrumentation .................................................................. 12
5. Principles of initial calibration, correlation and initial validation......... 13
5.1 Initial calibration and correlation ......................................................... 13
5.2 Probationary/initial validation ............................................................. 14
5.3 Analyser result and accepted reference value comparison methods. ......... 14
6. Principles of routine validation and calibration adjustment ................ 15
6.1 General ........................................................................................... 15
6.2 Analyser result and reference test/primary test method comparison
techniques ............................................................................................ 16
6.2.1 Reference sample method ........................................................ 16
6.2.2 Paired sample method ............................................................. 16
6.3 Routine validation (scheduled check of continuing performance) .............. 17
6.4 Calibration adjustment (adjustment, if required, to re-establish analyser
results within performance limits)............................................................. 17
6.5 Routine validation and calibration adjustment intervals .......................... 18
6.6 Number of result values required for routine validation and calibration
adjustment ........................................................................................... 19
6.7 Routine validation and calibration adjustment of model based/chemometric
process analysers................................................................................... 20
7. Comparison methods: reference sample method and paired sample
method ................................................................................................... 21
7.1 Comparing analyser and reference method results................................. 21
7.2 Routine validation and calibration adjustment using the reference sample
method ................................................................................................ 21
7.2.1 Advantages and limitations ...................................................... 21
7.2.2 Choice of reference samples ..................................................... 22
7.2.3 Testing of reference samples .................................................... 23
7.3 Routine validation and calibration adjustment using the paired sample
method ................................................................................................ 23
7.3.1 Advantages and limitations ...................................................... 23
7.3.2 Collecting paired sample results ................................................ 25
8. Evaluation of routine validation and calibration adjustment results: use
of statistical tools and control charts ...................................................... 27
8.1 General ........................................................................................... 27
8.2 Use of statistical tools ....................................................................... 27
8.3 Routine validation and calibration adjustment: use of control charts......... 27
8.3.1 CUSUM approach .................................................................... 29
8.4 Control chart limits ........................................................................... 30
v
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
8.5 Control chart limits derived from analyser reference sample or paired sample
laboratory data ...................................................................................... 31
Appendix 1: Example interpretation of control chart readings ................ 33
Appendix 2: Determination of control chart limits by measuring standard
deviations of differences ......................................................................... 35
References and Bibliography .................................................................. 37
vi
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
Figures
Figure 1 Example of a control charting diagram……………..…………………..28
Figure 2 Examples of analyser results in a CUSUM chart………………………29
Figure 3 Example of accurately distributed control chart reading………….33
Figure 4 Example of biased control chart reading…………………….............33
Figure 5 Example of drifting control chart reading………………………………33
Figure 6 Example control chart reading; value outside warning limit…….34
Figure 7 Example determination of control chart limits by measuring
standard deviations…………………………………………………………………………36
Tables
Table 1 Sample validation uncertainty results…………………………………….19
Table 2 Example of the control chart limits for a distillation analyser.…..35
vii
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
viii
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
Preface
This Code of Practice has been prepared by the EEMUA On-Line Analysers
Committee. It replaces the First Edition of this Publication, previously named
‘Calibration and checking of process analysers’, which in turn replaced the
following historical IP Codes of Practice relating to analysers: IP 340/92, 341/82,
347/82, 348/82, 349/82 and 353/82.
The word analyser has been used throughout this Publication to refer to
instruments, variously known as on-line analysers, process stream analysers,
process analysers, quality analysers, quality measuring instruments and process
quality monitors.
ix
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
x
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
The content and presentation of this Publication recognises that the effort available
for routine validation and calibration adjustment is limited. The simplified
approaches recommended are therefore intended to require only a reasonable
expenditure of effort in order to maintain assurance that the initial calibration and
correlation of the analyser to the reference method is still valid.
In some cases the simplified approaches will produce procedures, which although
adequate, may be less rigorous than could be evolved on purely technical and
theoretical grounds. The approaches should be amended to ensure that they meet
the individual business needs of the manufacturing or other process to which they
are applied.
It should be noted that this Publication is primarily intended for use by those
directly involved with supplying, using and maintaining on-line/process analysers.
These individuals are generally not specialists in quality measurement principles
and the associated statistical calculations. Therefore, as far as possible, this
Publication avoids the explicit use of statistical methods and favours descriptions
and explanations of a more practical nature. In order to remain consistent with
the theoretical approach these practical approaches are assumed to exhibit
random statistical behaviour with a normal/Gaussian distribution.
1
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
2
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
2. Definitions
The following glossary aims to add further definition to the terms used in the text
in the context of their use when describing activities associated with process
analysers.
2.1. Equipment
Analyser time constant. For a continuous analyser this is the time interval
between the initial response and the time to reach 63 per cent of the final output
value with a step change in the measured property. For cyclically sampling
analysers the time constant is the analyser cycle time.
2.2 Measurement
3
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
Bias (ASTM E456(5)). The difference between the analyser results and the
accepted reference value. A consistent offset between the analyser result and the
accepted reference value.
Dead band (IEC 60050-311). The finite range of values within which a
variation of the input variable does not produce any measurable change in the
output variable.
Sensitivity. The value of the smallest measurement that can be made and seen
on the read out instrumentation.
4
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
Accuracy (ASTM E456). The closeness of agreement between a test result and
its accepted reference value.
5
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
6
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
Control limits (ASTM E456). A set of boundaries on a control chart that are
used as criteria for signalling the need for action or judging whether a set of data
does or does not indicate a state of statistical control. If all values are within the
control chart control limits this indicates a state of statistical process control.
Normally three standard deviations above and below the control chart centre line,
equivalent to a level of confidence of 99%. Only 1 result in 100 is expected to
exceed the limit.
Lower control limit (ASTM E456). Minimum value boundary of the control
chart statistic that indicates statistical control.
Lower warning limit. Minimum value boundary of the control chart statistic that
indicates statistical control.
Performance action limit. Condition derived from the control chart that requires
action. A value exceeds the Upper or Lower Control limit or a series of values
exhibit a pattern that indicates drift, bias, or a particular recognised analyser fault.
Random error (ASTM E456). A part of the error that in the course of a number
of test results remains constant or varies in an unpredictable way.
7
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
method on identical test items, in the same laboratory by the same operators,
using the same equipment, within short intervals of time.
Systematic error (ASTM E456). A part of the error that in the course of a
number of test results remains constant or varies in a predictable way.
Upper control limit (ASTM E456). Maximum value boundary of the control
chart statistic that indicates statistical control.
Upper warning limit. Maximum value boundary of the control chart statistic that
indicates statistical control.
Warning limit (ASTM E456). Boundaries on a control chart that are used as
criteria for signalling the potential need for action or judging whether a set of data
indicates a change in its state of statistical control. Set beneath the control limits.
Normally two standard deviations above and below the control chart centre line.
Equivalent to a level of confidence of 95%. Only 1 result in 20 is expected to
exceed the limit.
8
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
Process analysers have been developed and applied to measure continuously the
quality of material within or drawn directly from a flowing line. The benefits are
very considerable in terms of reduced sampling and testing time, improved
precision, an increase in the number of tests on each batch of product, the
opportunity for automatic control of quality, and reduction in laboratory testing
effort.
Of necessity, a process analyser differs from the standard test apparatus and even
the basic measurement technique may be different. For absolute properties made
by direct measurements (e.g. density, viscosity, etc.) the result obtained should
be independent of the type of equipment and the method used for its
measurement. Thus, on average the analyser and the standard test should give
identical results provided that due account is taken of any difference in the
precision of the two methods.
Empirical properties, made by indirect measurement (e.g. flash point and octane
number) are only definable in terms of the standard test. As the analyser is
different in some degree, it cannot produce independent meaningful values and
should be calibrated to the standard test.
Whatever the property measured and the type of analyser, users should be able to
assure themselves, and demonstrate to other interested parties that the analyser
result is the same within acceptable limits as would be obtained from the relevant
standard test.
In the terms of ASTM D3764 there are four key activity stages associated with
establishing the measurement performance of a process analyser:
9
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
If the control limits for the statistical control chart defined for each analyser are
exceeded, a fault/equipment failure is discovered, or confidence in the analyser
results is lost, a calibration adjustment is usually carried out. The calibration is
done after the repair or other correction has been completed and the analyser is
returned to its fully working condition.
10
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
Reliable and accurate calibration is usually only achieved using analysers which:
This Publication is not directly concerned with the selection and installation of
analysers, however a summary of the key aspects that may affect validation and
calibration performance are covered in this Chapter.
The test conditions and samples should simulate as closely as practicable the
expected site conditions. Particular attention should be paid to the relationship
between the full scale range of analyser results, the expected normal process
levels, and the chosen routine validation and routine calibration levels.
Actual plant samples from the process stream should be used at the FAT if
practical. Due to difficulties in the transportation of samples and the availability to
the supplier of suitable certified reference materials or laboratory analysis, it may
be necessary to do the initial calibration and acceptance of the analyser on site at
as part of a Site Acceptance Test (SAT).
Evidence of the performance of the analyser under standard test conditions should
be obtained. In particular figures for analyser repeatability over 24 hours in terms
of the analyser result standard deviation. Details should also be recorded for the
uncertainty components/contributions of the analyser and the validation reference
materials used and any other relevant factors, e.g. temperature and atmospheric
pressure.
Rigorous calibration and correlation as defined by the ASTM practices is usually not
justified at factory due to the number and diversity of reference samples that are
required and the need for special facilities to maintain stable sample and
environmental reference conditions for long periods of time.
11
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
The total analyser system, from sample take off to result display, should be
reliable. What constitutes adequate reliability is largely a matter of judgement by
the user. However, the system should only be regarded as satisfactory if it does
not require any kind of intervention to remain within its acceptable limits more
frequently than once per week. Examples for intervention might be viewing
operation conditions/software/alarms, adjusting flow rates, cleaning filters, etc.
The site precision (normal expected performance) of the installed analyser should
be established over a period of a few weeks using a constant reference sample
injected periodically into the analyser or by taking laboratory cross reference
samples. The site precision is then used to determine if the analyser is beginning
to develop a bias (drift) or becoming less stable by on-going regular validation
checks using the reference sample or laboratory cross checks. This is assessed
using control charts as described in the following Sections.
The performance of the total analyser system should be validated through to the
operational (normally remote) display. In this context the analyser result means
the reading as it is normally displayed and acted upon such as on a control system
screen or other process operator interface.
All signal converters, input cards, data transmission, and display devices should
also be periodically checked. For example, they should be calibrated and checked
for sensitivity, deadband and linearity, digital resolution, etc., in accordance with
the manufacturer's instructions. Their contribution to the overall uncertainty
should be recorded and included in any evaluation of the overall analyser system
performance. The analyser system result remote display should also show the
result units and name of the property or composition being measured.
12
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
For process analysers these activities are usually carried out by the equipment
suppliers as part of their product development activity. This may form part of a
factory acceptance test.
However, for process analysers establishing the scientific and statistical assurance
to validate this process can require a considerable amount of organisation and
resources even if conducted remotely from the process plant.
To establish initial calibration that remains consistent across a large part of the
analyser range involves a number of different samples each of which need to be
analysed many times. This could extend to over ten significantly different samples
analysed to produce 30 results each using both laboratory and analyser
equipment. This is only likely to be justified for the most critical or important
analysers associated with high financial incentives.
Performance across the analyser measurement range commonly relies upon the
equipment manufacturer’s internal calibration function which compensates for any
non-linearity of the particular detection technique used. There should also be
assurance that each individual analyser produced remains consistent with the
extensive calibration and correlation testing done at the time of that particular
model’s product development.
13
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
For process analysers these activities are usually carried out by the owners of the
equipment during the initial period of operation of a new analyser using plant
samples or specially purchased reference materials.
The data is reviewed/pre-treated and a control chart plotted. If the results on the
chart exhibit a state of statistical control the analyser is deemed to have passed
probationary validation.
See Chapter 8 for more detail on the use of statistical tools and control charts.
See Section 6.2 for a description and comparison of the two methods.
14
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
One of the key success factors in maintaining high process operational confidence
in process analyser results is an organised strategy for scheduled, routine
validation and calibration adjustment.
Analysers are fundamentally different from other process instruments in that they
measure the physical or chemical compositional properties of a material (e.g. pour
point or weight percentage) rather than measure its state (e.g. temperature or
pressure).
For some of the properties measured by process analysers there are no absolute
standards available. For these properties it is possible that the uncertainty
components associated with laboratory/primary test methods and certified
reference materials can be equal to or larger than the analyser equipment
uncertainty component as stated by the process analyser manufacturer and/or as
determined by initial validation. This means that the analyser itself, when
operating correctly, may not add significantly to the overall uncertainty of its
result.
The number of samples and composition of the samples needed for routine
validation and calibration adjustment should be determined during the initial
validation and correlation activity stages. Measurements with non-linear
calibration curves may require a routine validation and calibration adjustment that
gives an assurance that the shape of the calibration curve has not changed. This
may require more than one reference material at specified composition levels.
Multiple validation levels may need to be considered to both confirm the analyser’s
measurement performance, close to the full scale range, and also at a decision
level for the process operation, e.g. at a low alarm point.
15
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
Other important factors which effect routine validation and calibration adjustment
performance are as follows:
Routine validation and calibration adjustment, comparing analyser results and the
accepted reference values obtained using the reference test/primary test method
can be achieved using either the reference sample/validation reference material or
paired-sample/line sample methods.
In the paired sample method the process analyser result is recorded at the same
time as a reference sample is collected and stored. The reference sample should
be identical to the sample flowing through the process analyser. The accepted
16
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
reference value is then determined for the stored sample using the reference
method and recorded on a certificate of analysis. The accepted reference value is
then compared with the process analyser result obtained at the same time as the
reference sample was taken.
Routine validation is the scheduled comparison of the process analyser result with
the accepted reference value derived from a reference method/primary test
method. Both results are obtained by sampling the same material, at different
times, using the reference sample or paired sample comparison methods described
in Chapter 7.
If the analyser meets the normal operational performance variation defined for it
in the control chart, no adjustment should be made, even if it does not exactly
agree with the assigned reference value.
Note: Adjusting the analyser to agree exactly with the assigned reference when
not outside normal operational variation limits (site precision) makes the analyser
system less precise (more noisy).
If the analyser results comparison falls outside of its performance limits, action is
taken to identify the cause and correct it. After the completion of repairs or the
identification and solution of problems, a calibration adjustment is performed to
re-establish the analyser result within its performance limits.
17
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
For optimum precision, over the long term, adjustments should only be made to
the analyser results if the routine validation result is beyond its performance action
limits.
For results that are just below performance action limits it is statistically likely that
the next result will move away from, rather than go beyond the limit. A more
detailed study of the routine validation results for previous routine validations
should be made to determine if there are any underlying problems that need to be
investigated. Some examples of this, using control charts (run rules), are shown
in Appendix 1.
The choice of the validation intervals is important to ensure that process analyser
condition information is available on a regular basis to support the chosen
maintenance and operational strategy. The most important/critical analysers will
have a shorter interval between routine validations.
The frequency of these activities will depend upon many factors specific to each
application. These include:
For a new installation or new analyser type data may initially need to be collected
at shorter intervals, for example, on a daily or weekly basis.
Statistical tools should be used to determine if the initial routine validation interval
is appropriate. See Chapter 8 for more details.
18
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
outside of the normal operating envelope. Of more concern is the small but
significant shift only detected by specific routine validation. Many analysers have
a self-diagnostic alarm feature, which is also useful to detect failures between
validations.
The number of results or tests conducted, on the same sample, effects how close
the average value of these results is to the true value. As the number of results
increases their average will become increasingly closer to the true value of the
property or composition being measured.
Agreement with the true value, which is the average or mean of a much larger
number of results, improves in proportion to the square root of the number of
results.
For example, if a single analyser result has an overall uncertainty of plus or minus
6 wppm (watt parts per million) at a confidence level of 95%, any single result can
be up to 6 wppm above or below (plus or minus) the true value with a 95%
confidence level. Increasing the number of samples from 1 to 9 will bring the
average of the 9 results to within plus or minus 2 wppm of the true value with a
95% confidence level. This is summarised for other multiples of results in the
following table. For a single result there is an uncertainty U1 = 6 wppm at a 95 %
confidence level.
The result values used for comparing the analyser results and the reference
method results can be the average of a number of results to improve the
uncertainty value.
Process analyser precision can be better than that of the reference method/
standard test. In certain applications, for example, where the reference sample is
collected over a period of time, the value of a property measured with a process
analyser is the average of a large number of result readings (perhaps over a whole
production batch). In these circumstances the precision of the estimated average
19
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
value of the analyser is much better than its repeatability. In this case the
precision of calibration (and validation) depends largely upon the precision of the
results derived from reference method.
In practice, reference method testing is often expensive in terms of the time and
facilities required, so the number of tests requested for routine validation and
calibration adjustment purposes needs to be limited. Stored analyser reference
samples may be limited in quantity or plant stability uncertainty. Also, access to
technician resources may constrain the time available to collect larger numbers of
analyser results.
If the routine validation comparison between the analyser result and the reference
method results exceeds the control chart action limits detailed checks on the
analyser are usually required. Any faults should be rectified in order to restore the
analyser diagnostic responses to the required levels. For example, using raw
detector responses to single component reference or blank/zero level materials.
Due to the complex calibration relationships that may be present, particularly with
predictive results from one analyser that are constructed from calibrations
involving reference values obtained from more than one reference method, any
calibration adjustment is likely to be carried out by the owner of the chemometric
model and could involve significant additional processes and reference samples.
This may involve the equipment vendor, a laboratory development or research
department.
20
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
Calibration, correlation, and validation can be done using either or both of the
following methods:
The reference sample method can be used with confidence on analysers that have
shown no significant effect from sample type, sensitivity or contaminants. The
principal attraction of this method is that no reference sample testing is required
at the time of the routine validation, the sample property is known precisely in
advance and the whole routine validation procedure can be automated if required.
Subject to a demonstration of sample stability, sufficient reference sample (or
several different samples) can be stored for routine validations over several
months, reducing the total reference testing effort. If reference sample injection is
automated, the routine validation facility could be made available on demand and
could if necessary be operated at the start and end of a batch transfer, for
example.
Each method has certain advantages and limitations and the choice of which to
use, and when, depends upon several factors. For factory/works testing and for
site testing before a plant is running, the reference sample method should be
used. When the process stream is available both methods can be used to their
best advantage. Each method is dealt with more fully below.
This method uses reference samples, which have been accurately tested by the
reference method, are injected into the analyser and the corresponding analyser
result value(s) obtained.
21
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
7.2.1.1 Advantages
It is the only method applicable when the normal process stream is not
available.
When suitably stable, the reference sample may be a stored process sample
that has been analysed and certified using the reference method.
A reference sample may be more easily handled and/or its reference value
may be more accurately measurable than the process stream.
With a sufficient number of reference samples, the whole range of the
analyser can be calibrated or tested at the user's convenience without the
need to wait for suitable values to occur in the process stream.
The same reference samples can be used for initial calibration and for
subsequent routine validation and calibration adjustment. It can also be
used for a cross comparison between different analysers.
It gives improved precision as it is related to the repeatability rather than
the reproducibility of the standard test method.
A sample can be stored in a suitable container as part of the analyser
sample conditioning system, in the field, to facilitate manual or automatic
validation.
The analyser calibration is directly linked to the reference method as used
in the site laboratory if a plant sample is used with an assigned reference
value determined by the site laboratory.
7.2.1.2 Limitations
The choice and quantity of reference sample material will be dependent on the
type of analyser. However the following points should be considered:
22
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
For plant samples, unless otherwise specified by the reference test method, the
accepted reference value for use with the reference sample method is the average
of a minimum of four valid results taken over several shifts/days.
This method uses samples of the normal process stream, which is also being
continuously sampled by the process analyser. These are drawn off and then
tested in a laboratory using the standard reference method. The corresponding
analyser results for the time when the samples were taken and reference analysis
values are then compared and used for validation or calibration.
7.3.1.1 Advantages
23
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
7.3.1.2 Limitations
The cross check precision is a function of the analyser and laboratory test
precision. This must be reflected in the control chart limits that are used
(based on actual cross check precision established over a period of time);
not the analyser or laboratory method precision.
24
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
Special Cases. There are certain types of analysers, for example those
measuring water in gases, for which the standard method involves
measurement or special sample collection techniques over a long period.
This can be many hours and generally much longer than one analyser time
constant. The property being measured is very likely to change over this
period. In this special case the analyser value should be the average over
the whole period of the standard test and only one standard test result is
practicable for each routine validation. The procedure should be carried out
a number of times to achieve sufficient points for confident calibration and
routine validation. The methods described in this Publication do not cover
such special cases, and users, if in doubt, are advised to discuss the details
with the equipment supplier.
Sampling should be conducted when the process plant operation and the process
analyser are performing normally. In other words, the operating parameters and
values (sample and service temperatures, pressures and flows) are substantially
those which were used for the initial calibration where this is appropriate.
A time should be chosen when the analyser result readings and other evidence
indicate that the plant is reasonably steady and is expected to remain so for the
duration of the sampling and data collection period.
The time constant/cycle time for process analyser systems is generally between 1
minute and 30 minutes. In practice the analyser readings before and after the one
reading taken as the analyser result value for validation will serve to show the
steadiness of the process and the process analyser during the test period. This
supporting evidence improves confidence in the analyser results obtained.
The reference sample container should be carefully labelled with analyser identity,
date and time of sampling and any other relevant data, including the plant and
analyser operating conditions. If being used for calibration adjustment, the
accepted value for the paired sample method should be taken from a minimum of
four results over different days. The manual sample point should be close to the
analyser sample take off point or close to the sample transport system.
25
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
26
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
The guidance in this Chapter applies to both the reference sample method and the
paired-sample method. The chosen evaluation tools should be further adapted to
suit the user's particular needs and circumstances.
Numerical statistical tools are available for the analysis of analyser validation
results or to determine if the analyser is still in control and/or the calibration has
drifted. These techniques are applied directly to the data values obtained during
routine validation.
This Publication presents one simplified control chart technique using the principles
of these tools to track process analyser performance.
Other more rigorous techniques can be applied if required. Examples of which are
detailed in standards such as ASTM D6299 - Applying statistical quality assurance
and control charting techniques to evaluate analytical measurement system
performance(8).
The purpose of the control chart is to monitor or track minor differences due to
normal expected deviations, which do not require recalibration around the centre
line of the control chart to occur without adjusting the calibration, e.g. span and
zero settings of the analyser.
The causes of the differences, e.g. reference variation, sample handling variations,
temperature, and pressure may cycle naturally and may not drive the difference
outside of the control limits. The control chart limits include these natural
27
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
It is usual to set two control chart limits against which different actions can be
taken. Guidance on some of the methods used for setting these limits is covered
in Section 8.4.
The limit closest to the aim line is the ‘warning limit’. The upper warning limit is
the positive side of the aim line and the lower warning limit is to the negative side.
In general when a difference is outside this limit it is a prompt that the analyser
may need calibration adjustment or adjustment in the future after further routine
validations. A shorter routine validation interval may be put in place until the
reason for any change in the analyser performance is understood.
The limit furthest from the aim line is the ‘control limit’. The upper control limit is
the positive side of the aim line and the lower control limit is to the negative side.
In general, when the difference exceeds the control limit it is an indication of
possible system malfunction. It may be required after an investigation, to
establish the reason for the difference and the repair of any analyser equipment
fault.
Only over a number of routine validations, which may be several weeks or months,
can a pattern of results be obtained that will indicate any bias or drift from within
the normal scatter of results. This scatter is inherent within the uncertainties of
the analyser results and the reference method results.
Difference
- Analyser +ve (upper) control Limit
Result
minus +ve (upper) warning Limit
reference
(Lab)
result
Where bias is encountered, the offset of the difference should remain constant
unless drift is also present.
For bias correction, the control chart aim line can then be re-defined or the
analyser results given an appropriate offset to account for the bias. No further
action is needed if the differences remain within the new control chart control
limits.
28
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
CUSUM is defined as the cumulative sum of the differences between the reference
method/laboratory and the analyser results. The difference is then added to a
running total to form an updated running total.
Where
n = current time paired value
n-1 = previous time paired value
L(n) = current reference laboratory result or identified reference sample certified
value
A(n) = current analyser result
CUSUM(n) = cumulative sum (current value)
CUSUM(n-1) = cumulative sum (previous value)
difference
The horizontal axis is related to time, i.e. the results are plotted in chronological
order. However, the interval between the samples is plotted as a constant
irrespective of the actual time interval.
29
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
The slope of the plot is the average bias between the analyser result and the
reference method/laboratory result, or reference material certified value.
Action limits using CUSUM would include the maximum change in slope (bias)
permissible and what to do if the limits are exceeded.
Control chart warning and control limits should be based on the actual
performance data of the analyser under plant operating conditions (site precision
conditions) for the reference sample method or the actual cross check performance
data with the reference method/standard test/laboratory test when the paired
sample method is used.
Using other data, e.g. short term repeatability data from the vendor or the
laboratory method, reproducibility may result in control limits that are either too
tight or too wide. It is important that the on-going validations are assessed
against the actual long term performance of the analyser as installed at site.
However the other data may be used to initially set control chart limits when the
analyser is first installed before sufficient operational data is obtained to calculate
the actual true in-plant operational performance limits, or if the analyser
incentives do not justify the work to develop the true operational performance
limits.
The control chart warning limits are normally set, in statistical terms, equal to two
standard deviations of the mean of the difference between the analyser results and
the accepted reference value. This is equivalent to uncertainties quoted at 95%
confidence limits, i.e. only one difference in twenty is expected to fall outside the
warning limit under normal circumstances. Warning limits can be calculated as the
control limit multiplied by 0.75.
The control chart control limits (CL) are normally set, in statistical terms, equal to
three standard deviations, as defined above, which is equivalent to uncertainties
with 99% confidence limits, i.e. only one difference in 100 is expected to fall
outside the control limit under normal circumstances. These limits can be set from
the ‘warning limit’ values multiplied by 1.5.
30
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
Once the initial settings of the control limits have been determined, verification
routine validations will be required to establish whether the analyser is within
limits (or in control). The frequency of on-going planned maintenance routine
validations should then be decided by interpretation of the results.
The standard deviation and hence chart limits should be determined in operation
by initially calibrating the analyser and then performing a minimum of 25-30
routine validations with the reference sample or paired-sample cross checks with
the laboratory without further calibration adjustments. The validations should be
taken over a period of a few weeks at different times of the day and with different
technicians if possible to ensure they capture the normal analyser variations in
plant operating conditions.
When 25-30 routine validations have been completed the calculation of the
estimated standard deviation can be performed. The data should be checked by
initially plotting in time sequence to ensure there are no anomalies or significant
drift during the period the data was collected. The data should then also be
statistically checked for outliers which should be removed from the data set and
for a normal distribution (see ASTM D6299) to confirm the variations are truly
random. The standard deviation can then be calculated using the root mean
square technique and the warning and control limits can then be derived.
The control chart limits should be regularly reviewed and adjusted after every 25-
30 data points are obtained. Twenty five to thirty results are the minimum
considered to be statistically relevant.
31
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
32
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
CL
CL
CL
33
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
CL
34
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
For example, the control chart limits for a distillation analyser measuring 90%
point would be determined as follows:
The estimated standard deviation can be used to set the control chart limits
The sum of differences in the table is zero which indicates the analyser is not
displaying any particular bias therefore the ‘aim line’ of the control chart will
represent zero deviation between analyser and reference and no offset adjustment
is needed.
It must be noted that these control chart limits will differ from the limits calculated
from standard reference test uncertainties and analyser system uncertainties as
35
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
the measured data used includes an element of sample handling and the fact that
the process itself is not steady therefore any non-linearity in the analyser will also
be included. This usually results in wider limits being set than strictly necessary.
The above distillation analyser example would produce a control chart as shown
below.
Note: this example is based on only eight data points but normally the standard
deviation should be calculated based on a minimum of 25-30 data points.
X
WL
+ 6.1 Deg C
(2 x ESDev)
X
Aim line Time
X 0 Deg C
X X
X X X
WL +
- 6.1 Deg C
(-2 x ESDev)
CL
- 9.3 Deg C
(-3 x ESDev)
36
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
7. NPL G11, NPL measurement good practice guide 11 (Issue 2): a beginner’s
guide to uncertainty of measurement, National Physical Laboratory, 2001.
9. EEMUA Publication 138 Design and installation of on-line analyser systems, 2nd
Edition, Engineering and Equipment Materials Users' Association, 2010, ISBN
978 0 85931 174 8.
10. API RP555 - Process analyzers, Third Edition, American Petroleum Institute,
2013.
11. ASTM D6708 - Standard practice for statistical assessment and improvement of
expected agreement between two test methods that purport to measure the
same property of a material, American Society for Testing and Materials, 2008.
Bibliography
37
EEMUA Publication 175 – Calibration and validation of process analysers © EEMUA
38
EEMUA Publication: feedback form
An electronic version of this form is available at www.eemua.org/publications.htm
When you have completed this form with all your comments, please save and e-
mail to info@eemua.org or post a paper copy to: Publications Department,
EEMUA, 63 Mark Lane, London EC3R 7NQ.
EEMUA Publications Catalogue
All EEMUA publications can be purchased on-line. To order a publication, please
contact EEMUA at www.eemua.org, telephone: +44 (0)20 7488 0801, fax: +44
(0)20 7488 3499 or e-mail: sales@eemua.org.
206 Risk Based Inspection: A Guide to Effective Use of the RBI Process
181 A Guide to Risk Based Assessments of In-situ Large Ex 'e' and Ex 'N' Machines
161 Guide to the Selection and Assessment of Silencers and Acoustic Enclosures
194 Guidelines for Materials Selection and Corrosion Control for Subsea Oil and
Gas Production Equipment
182 Specification for Integral Block and Bleed Valve Manifolds for Direct
Connection to Pipework
179 A Working Guide for Carbon Steel Equipment in Wet H2S Service
158 Construction Specification for Fixed Offshore Structures in the North Sea
146 90/10 Copper Nickel Alloy Piping for Offshore Applications - Specification:
Fittings
145 90/10 Copper Nickel Alloy Piping for Offshore Applications - Specification:
Flanges Composite and Solid
144 90/10 Copper Nickel Alloy Piping for Offshore Applications - Specification:
Tubes Seamless and Welded
143 Recommendations for Tube End Welding: Tubular Heat Transfer Equipment,
Part 1 - Ferrous Materials
Pressure equipment
204 Piping and the European Pressure Equipment Directive: Guidance for Plant
Owners/Operators
215 Industrial Cooling Tower Fans and Fin Fans – Guide for Design,
Maintenance and Operation
205 Guide to the Production Testing of Valves for the Energy, Process, Oil and
Gas Industries
204 Piping and the European Pressure Equipment Directive: Guidance for Plant
Owners/Operators
192 Guide for the Procurement of Valves for Low Temperature (non-cryogenic) Service
182 Specification for Integral Block and Bleed Valve Manifolds for Direct
Connection to Pipework
217 Safe and effective operation of storage tanks for oil movements
213 Emission reduction from oil storage tanks and loading operations
190 Guide for the Design, Construction and Use of Mounded Horizontal
Cylindrical Bulk Storage Vessels for Pressurised LPG at Ambient
Temperatures
183 Guide for the Prevention of Bottom Leakage from Vertical Cylindrical Steel
Storage Tanks
180 Frangible Roof Joints for Fixed Roof Storage Tanks: Guide for Designers and
Users