Sie sind auf Seite 1von 8

1.

4 System Accuracy*
B. G. LIPTÁK (1982, 1995, 2003)

DEFINITIONS OF TERMS Reproducibility (ISA). In process instrumentation, the


closeness of agreement among repeated measure-
Accuracy (Webster). Freedom from error or the absence of ments of the output for the same value of input
error. Syn. precision, correctness, exactness. (In this made under the same operating conditions over a
sense, the term is a qualitative, not quantitative, concept.) period of time, approaching from both directions.
Accuracy (ISA). In process instrumentation, degree of con- Reproducibility (NIST). Closeness of agreement between
formity of an indicated value to a recognized accepted the results of measurements of the same measurand
standard value, or ideal value. carried out under changed conditions of measurement.
Accuracy, measured (ISA). The maximum positive and Uncertainty (Webster). A feeling of unsureness about
negative deviation observed in testing a device under something.
specified conditions and by a specified procedure. Uncertainty (IEH Section 1.5). Measurement uncertainty is
Accuracy of measurement (NIST). Closeness of the expressed to a confidence level of 95%, and it is the
agreement between the result of a measurement and limit to which an error may extend.
the value of the measurand…. Because accuracy is
a quantitative concept, one should not use it quan- Language, Terminology, and Reality
titatively or associate numbers with it. (NIST also
The guide titled International Vocabulary of Basic and Gen-
advises that neither precision nor inaccuracy should
eral Terms in Metrology (commonly referred to as VIM ) was
be used in place of accuracy.) published by ISO in the name of seven organizations and
Error (ISA). In process instrumentation, the algebraic dif- contains the VIM definitions of 24 terms relevant to measure-
ference between the indication and the ideal value ment and accuracy. So, from a theoretical point of view, we
of the measured signal. It is the quantity that, alge- do have standards and internationally agreed upon definitions.
braically subtracted from the indication, gives the But the reality in the average industrial plant is different,
ideal value. and this Instrument Engineers’ Handbook is written for the
Range (ISA). The region between the limits within which average instrumentation and control (I&C) engineer in those
a quantity is measured, received, or transmitted, plants. Therefore, when we quantify an error herein, which
expressed by stating the lower and upper range values. one should expect when making a measurement with a par-
Rangeability (recommended by IEH). Rangeability of a ticular instrument, we will not (yet) use terms such as uncer-
sensor is the measurement range over which the error tainty but will try to stay on familiar grounds. On the other
statement, in the units of a percentage of actual read- hand, we will try to take a step in the right direction by
ing, is guaranteed. improving the clarity of our language.
Repeatability (ISA). The closeness of agreement among a When an instrument is specified to have ±1% accuracy,
number of consecutive measurements of the output people do not expect it to have 99% error! The intended mean-
for the same value of the input under the same oper- ing of that statement is ±1% inaccuracy or a ±1% error relative
ating conditions, approaching from the same direc- to some reference standard. It is important to emphasize the
tion, for full-range traverses. role of a reference standard in all measurements, as we humans
Repeatability (NIST). Closeness of agreement between are incapable of measuring anything in the absolute. All we
the results of successive measurements of the same can do is compare an unknown quantity to a known one and
measurand carried out under the same conditions of determine which is larger or smaller and by how much. The
measurement .... Repeatability may be expressed presence of a reference also means that a measurement can be
quantitatively in terms of the dispersion character- in error not only because the sensor is inaccurate but also
istics of the results. because the reference has drifted or was inaccurate to start with.

* Used with permission of the Instrumentation, Systems and Automation Society.

78
© 2003 by Béla Lipták
1.4 System Accuracy 79

CLARIFYING THE “ACCURACY” STATEMENT ±


Bias

In a volume dealing with process measurement, no subject


is more deserving of in-depth evaluation than the error that 50% of Area ±0.67δ True Value

is inherent in all measurement. Good control is possible only


68.3% of Area ±1δ
if the controlled variable is precisely measured. Yet the term
accuracy (or, more precisely, inaccuracy or uncertainty)
itself is poorly defined, frequently misunderstood, and often
used as a sales gimmick. Consequently, use of this term cries
Precision 95% of Area ±2δ
out for international standardization and, as was noted above,
ISO has already prepared such standards. The need for clar- −3 −2 −1 0 1 2 3
ity of language and standardization exists for the following −0.67 0.67
Total Uncertainty
reasons:
FIG. 1.4a
1. When the error or inaccuracy of an instrument is stated In any measurement, the total uncertainty (total error) is the sum of
to be ±1%, one would assume that this statement refers the sensor’s random error (precision) and its systematic error (bias).
to the actual measurement—the actual reading. One
would assume that, if this particular instrument hap-
4. Yet another source of confusion is the fact that, when
pens to read 100, the true value of that measurement
the error of 100 sensors is tested, the results fall onto
must fall between 99 and 101, but this frequently is
a “bell curve” (Figure 1.4a). It would be desirable to
not the case. Some manufacturers express their error
reach international agreement so that all error state-
statements (inaccuracy percentages) on the basis of
ments would always be based on the performance of at
“percent of actual span,” while others might base it on
least 95% of the units tested. In addition, an error state-
“percent of full scale,” “percent of range,” or “percent
ment should always state if it is based on self-evaluation
of upper range value,” and so on. This inconsistency
performed by the supplier or on an evaluation by an
is undesirable, because it is confusing. It would be
independent testing laboratory and, in the latter case, if
better if all measurement error statements always
the test report is available for review.
referred to the actual measurement.
2. To make error statements expressed as percentages of
If the above four recommendations were universally
the actual measurement truly meaningful, the state-
accepted, the subject of sensor error and inaccuracy would
ment should also specify the measurement range over
be much less confusing. While this is not likely to occur soon,
which the statement holds true. This would be a simple
a better understanding of the factors that cause the present
matter if all manufacturers agreed to define rangeabil-
state of confusion should be helpful, because it can speed
ity as the measurement range over which their error
the development of universal standards for sensor error and
statement (as a percentage of actual reading units) is
performance.
guaranteed. This approach would allow all sensor inac-
curacies to be stated on the same basis and therefore
would eliminate the confusion. If all detector inaccura-
cies were stated as “x% of actual reading throughout TERMINOLOGY OF INACCURACY AND REPEATABILITY
the range of y,” users could be “comparing apples with
apples” when comparing bids, and the room for “cre- The purpose of all measurement is to obtain the true value
ative specmanship” would at least be reduced. of the quantity being measured, and error is thought of as
3. Further confusion occurs because different manufactur- the difference between the measured and the true quantity.
ers include different factors in their error statements. Because it is impossible to measure a value without some
Most suppliers include only linearity, rangeability, and uncertainty, it is equally impossible to know the exact size
hysteresis errors in their total error statement; they list of the error. What is possible is to state the limits within
the error contributions caused by drift, temperature which the true value of a measurement will fall.
effects, overrange, power supply, humidity, RFI, and The accuracy-related terminology used in the process
vibration separately. Actually, some manufacturers control industry can be illustrated by an example of target
claim an apparent increase in accuracy not by improving shooting (Figure 1.4b). The spread of the nine shots fired into
precision but by considering fewer and fewer effects in the upper right-hand corner of the target in a tight pattern
the total error statement. Naturally, to reverse this trend, represents the random error of the shooter. Looking at the
international agreement is needed with regard to the penetration of the bullets, one can say that his shooting is
amount of variation (in ambient temperature, power repeatable and precise, but precision alone does not guarantee
supply, and others) that the manufacturer’s error state- accuracy; it is only the measure of the ability of the shooter,
ments must include. which is called random error.

© 2003 by Béla Lipták


80 General Considerations

THE ACCURACY STATEMENT


Repeatability
Random Error
The accuracy of a product category (sensors, transmitters,
(Precision)
and so on) is established on the basis of testing large numbers
Systematic of these products. For the more important sensors, the accu-
Error 1
racy statements should also include information on testing.
(Bias)
An example of such a statement is quoted below from a
National Bureau of Standards Calibration Certificate for a
turbine flowmeter:

The results given are the arithmetic mean of ten separate


observations, taken in groups of five successive runs on
two different days. The reported values have an estimated
overall uncertainty of ±0.13%, based on a standard error
of ±0.01% and an allowance of ±0.1% for possible sys-
tematic error.
Total
Illegitimate Error Figure 1.4a illustrates the results of such a test. In that test,
Error (Inaccuracy) the precision (half of repeatability) of 68% of the devices tested
has been found to be ±1% of the true value, while, for 95%
FIG. 1.4b
Accuracy terminology. of the devices, it fell within ±2%, and, for all 100% of the
devices, it amounted to ± 3%. The total error (total uncertainty)
is the sum of precision plus bias, which is the systematic error
of the bell curve itself. Because the bias can be reduced by
The distance between the mean impact of the nine bullets calibration and rezeroing, but the precision (or repeatability)
and the bull’s-eye of the target is the systematic error. This cannot, it would be desirable if manufacturers identified both
error (caused by the wind or by the faulty adjustments of the of these values. Manufacturers should also state if the basis of
sights) is repeatable and can be eliminated (in case of sensors, this data is 68, 95, or 100% of the devices tested. To allow the
by calibration or by rezeroing; in case of the shooter, by I&C profession to mature, manufacturers should eliminate
waiting until the wind stops or by readjusting the sights). This “specmanship” from their sales literature, so that users can
error is not related to the shooter’s inability to duplicate his “compare apples with apples” when making a selection.
shots. Systematic error is also referred to as bias, which is the
systematic displacement of the measured value from the true
one. It can be reduced by recalibrating the sensor against a
reference standard, such as a calibrated (standard) thermal FLOW MEASUREMENT EXAMPLE
element, a known composition analytical sample, or dead
weights. Figure 1.4c illustrates three flow sensors installed in series
The shot in the lower left-hand corner of the target is an in the same process pipe, with each measurement signal being
illegitimate error, which is caused by blunders and can be totalized. All three flow sensors are sized for the same full
totally eliminated. range of 100 GPM (380 l/min) flow rate. The goal of this
The total error in a measurement can thus be defined as example is to illustrate how the total system error is determined
the sum of the random error and the systematic error or bias. at the flow rates of 20 GPM (76 l/min) and 80 GPM (304
If the purpose of an installation is to maintain the process l/min) in two different cases. In Case 1, the basic assumption
conditions at previously experienced levels, and there is no is that the component errors are additive. In Case 2, the assump-
interest in their true values, then the goal is to reduce the tion is that the total system error will be the error of the least
random error, without paying much attention to the remaining precise component in the system. Errors introduced by counter-
bias. In many industrial installations, such a repeatable (but totalizers, which is usually one count, will be neglected.
inaccurate) measurement is sufficient. For the purposes of the example of Figure 1.4c, the mag-
Conversely, if the interest is in determining the true value netic flowmeter, transmitters, and integrators will all have
of the measurement, because the installation serves such ±0.5% full-scale (FS) error. The orifice plate will be assumed
absolute purposes as accounting or quality control, the repeat- to have an inaccuracy of ±0.5% of rate and the error of the
able measurement is insufficient, and attention must be con- turbine flowmeter will be assumed to have an inaccuracy of
centrated on absolute (total) accuracy. This can be obtained ±0.25%. (The orientation table in Chapter 2 [Table 2.1b]
only through the reduction of both the random and the sys- provides complete data for all flow sensors, including their
tematic errors, which is usually achieved by recalibration. performance characteristics.)

© 2003 by Béla Lipták


1.4 System Accuracy 81

Scaler Flow Sensor


FQI FQI FQI
and
The
Output
Totalizer Totalizer (GPM or %)
Inaccuracy
of These
FQ FQ Devices is 100
± I Count
The
Inaccuracy Integrators
of These FY 80
Inaccuracy is
Expressed
Εxpresssed as Actual
As ± 0.5% of FT
± % of Actual
Full Scale 60
FT Measurement
Range

Ideal
40
Flow
FEM
(0−100 GPM)

Linear Nonlinear Linear 20


Analog Analog Digital
System System System Actual Flow
(Magnetic (Orifice (Turbine 0
0 20 40 60 80 100 (GPM or %)
Flowmeter) Plate) Flowmeter)

FIG. 1.4c FIG. 1.4d


Errors of the components of three different flow totalization loops. Performance of a linear analog flow sensor, such as a magnetic
flowmeter.

In the more detailed discussion that follows, it will be shown Flow Sensor Error
that the overall system error can be much greater than the com- Based on Full Scale
ponent errors. It will be shown that, at 20% of full-scale flow, (GPM or %)
the error of a turbine flowmeter will be around 0.25% of actual
flow, the error of a magnetic flowmeter might range from 3% 1.5
+
to 9% of actual flow, and the measurement error of an orifice-
based measurement error will range from 5 to 12% of actual rate. 1.0
Limit of Error
The ±0.5% maximum inaccuracy (some based on actual
readings, others on full-scale readings) was selected to reflect 0.5
the typical installations in the existing plants. Today, when
Actual
“smart” transmitters and improved sensors are available, one 0
can select more accurate system components, some with max-
imum errors of ±0.1% of actual span. Here, we will assume Ideal
0.5
that the maximum error of any of the system components is
Limit of Error
±0.5% and, based on that assumption, we will determine the
1.0
resulting total system error. The performance of analog and dig-
ital, linear and nonlinear devices will be discussed separately. −
1.5 Actual Flow
20 40 60 80 100 (GPM or %)

Analog and Linear Devices—Traditional FIG. 1.4e


Magnetic Flowmeters Error plot for a percentage of full-scale flow sensor.
The performance of a linear analog flow sensor, such as a
magnetic flowmeter, is shown in Figure 1.4d. The line It should be noted that the performance described here
marked “actual” represents the relationship between the is representative of the older designs of magnetic flowmeters,
actual flow and the output signal generated by the flow sensor. which continuously maintained their magnetic fields. In the
Figure 1.4e illustrates this error as a percentage of full scale newer designs, the field is cycled on and off, and the sensor
(FS), with the error limits being ±0.5% FS. can be automatically rezeroed, so the measurement error can
In Figure 1.4f, the same ±0.5% FS sensor performance be reduced. Therefore, the inaccuracy of these newer mag-
is illustrated, but against a vertical coordinate that is a per- netic flowmeters can approach ±0.5% of actual flow.
centage of actual flow units (instead of full scale). The spe-
cific detector performance is likely to be better at most points Analog, Nonlinear—Orifice Plates
of its range than what these error limits would imply. The
main message is that, for sensor with percent-FS perfor- The orifice plate itself is rather accurate and is a percent-of-
mance, the measurement error increases as the flow rate actual-flow sensor, having an error limit of ±0.5% of actual
drops, as shown in Figure 1.4g. flow rate, as shown in Figure 1.4h.

© 2003 by Béla Lipták


82 General Considerations

Flow Sensor Error Inaccuracy Based on


Based on % of Actual Actual Reading
Flow Rate Reading (%) (± %)

1.5 5
+ Limit of Error

1.0 4

0.5 3
Actual

0 2

Ideal
0.5 1
± 0.50 % of Measurement

0 Actual Flow
1.0 (GPM or %)
0 20 40 60 80 100
− Limit of Error
1.5 Actual Flow
(GPM or %) FIG. 1.4h
20 40 60 80 100
The error contribution of the orifice plate alone.
FIG. 1.4f
The error of a linear flow sensor shown in units of percentage of
full-scale flow.
Desired Measurement,
Flow (%)
Inaccuracy Based on Actual
Flow Rate Reading 100
(± %)
Actual
80
5
Ideal
60
4

40
3

20 Actual
2 Measurement,
Orifice
0
0 20 40 60 80 100 Pressure
1
Drop (%)
Actual Flow
0 FIG. 1.4i
0 20 40 60 80 100 (GPM or %)
Performance of an orifice type nonlinear analog flow sensor.

FIG. 1.4g
The error of a linear flow sensor shown in units of percentage of
actual flow. the error contribution (the gain effect) of this extraction
must also be recognized. Figure 1.4j illustrates that this
extraction of the square root improves the accuracy at the
The pressure drop through an orifice relates to the square higher flow rates but degrades it as the flow rate is reduced.
of the flowing velocity or volumetric flow rate through the
orifice plate. Figure 1.4i illustrates both, i.e., this ideal non-
Digital Linear—Turbine Flowmeter
linear (square root) relationship and the actual performance
of a specific differential-pressure (d/p) cell used in an orifice The calibration of a turbine meter in terms of the K factor,
type, nonlinear flow sensor. given in units of pulses per gallon, is rather similar to the
To the error contribution of the orifice plate shown in calibration curve of an orifice plate (Figure 1.4k). The inaccu-
Figure 1.4h (±0.5% of actual flow rate), one must add the error racy of a turbine meter is also in units of percentage of the
of the differential-pressure transmitter shown in Figure 1.4i actual flow and is rather constant over a fairly wide range of
(±0.5% FS). In addition, when the square root must be flows. Turbine flowmeter inaccuracy can be improved by
extracted before the signal can be integrated (Figure 1.4c), reducing the rangeability requirement of the unit (Figure 1.4l).

© 2003 by Béla Lipták


1.4 System Accuracy 83

Inaccuracy Based on Inaccuracy Based


Actual Reading on Actual Reading
(± %) (± %)

5 2.5

4 2.0

3 1.5
Nonlinear

2 1.0

Linear ± 0.5% at 30:1 Rangeability


1 0.5
± 0.25% at 10:1 Rangeability
Actual Flow
0 Actual Flow 0
0 20 40 60 80 100 (GPM or %)
0 20 40 60 80 100 (GPM or %)

FIG. 1.4j FIG. 1.4l


Comparing the inaccuracies of linear and a nonlinear flowmeter. Turbine flowmeter inaccuracy as a function of rangeability.

Turbine Meter TABLE 1.4m


K Factor System Inaccuracy (Total Loop Error) in Units of Percentage
(Pulses/Gallon) of Actual Readings

Assumption Used to
Basis 1 Basis 2
Estimate Accumulated
System Inaccuracy Operating Flow Rate (GPM)
Limit of Error Type of flow
detection loop 20 80 20 80

Mean K Analog, linear ±9.0% ±1.5% ±3.0% ±0.5%


(magnetic flowmeter)
Analog, nonlinear ±12.0% ±2.0% ±5.0% ±0.5%
(orifice flowmeter)
Limit of Error
Actual Digital, linear ±0.25% ±0.25% ±0.25% ±0.25%
Actual Flow (turbine flowmeter)
5 10 20 50 100 (GPM or %)

FIG. 1.4k the error is reduced by the square root of the number of sensors
Turbine flowmeter calibration curve. in parallel. So, if two sensor outputs (each having a 1% error)
are averaged, the error will be reduced to 1/ 2 = 0.7% (and
with three outputs, to 0.58%, with four outputs, to 0.5%, and
Combined System Accuracy
so on).
Having reviewed the inaccuracies of the three flow sensors Without actual system calibration, the evaluation of the
and the various loop components shown in Figure 1.4c, the overall loop accuracy must be based on some assumptions.
next step is to evaluate the resulting total loop errors. There Table 1.4m summarizes the system inaccuracies for the three
is no proven basis for determining the accumulative effect of loops in Figure 1.4c at 20 and 80% of flow rate and by evaluat-
component inaccuracies, and only an actual system calibra- ing the accumulated effect of component inaccuracies on the
tion can reliably establish the total loop inaccuracy. basis of one of two assumptions:
Still, we have learned the following from experience. We
know that, the fewer the number of components in an analog Basis 1 — Here, it is assumed that the inaccuracy of each
measurement loop, the better the loop’s performance. In dig- component is additive, and therefore the total loop inaccuracy
ital systems, no additional error seems to be introduced by is the sum of component inaccuracies (a very conservative
the addition of functional modules. basis).
It has also been reported that the averaging of the outputs
of several sensors that are detecting the same process variable Basis 2 — Here, the assumption is that the system inaccuracy
will reduce the measurement error. These reports suggest that is the same as the inaccuracy of the least accurate component

© 2003 by Béla Lipták


84 General Considerations

and therefore other inaccuracies can be neglected (a very TEMPERATURE AND PRESSURE EFFECTS
optimistic assumption).
If Basis 1 is accepted for evaluating the total system error, If a sensor such as a d/p cell has been tested at temperatures
an orifice-type installation operating at 20% of full-scale flow and pressures that differ from the operating temperature and
will have an error of ±12% of the reading, although the inac- pressure, this will affect the total error. The total error includes
curacy of any component in the loop does not exceed ±0.5% FS. the d/p cell error (E), which is determined under atmospheric
The data in Table 1.4m is based on the performance of ambient conditions. Therefore, E reflects the linearity, repeat-
conventional d/p transmitters and on conventional magnetic ability, and hysteresis errors of the sensor.
flowmeters. With the newer, pulsed DC magnetic flowmeters, For the purposes of this example, it is assumed that E =
the error can be reduced to ±0.5% of actual flow over a 10:1 ±0.2% of actual span. Other factors that affect the total error
range. Similarly, if the intelligent, multiple-range d/p cells include the zero (Tz) and span shifts (Ts) that might occur as
are used, orifice measurement error can be reduced to ±1% a result of temperature variations. For a temperature variation
2
of actual flow over a 10:1 range. To achieve this level of of 100°F (55°C), Tz is assumed to be ±0.5% of maximum
performance, it is necessary to automatically switch the d/p range, while Ts is assumed to be ±0.5% of actual reading.
cell span from its “high” to its “low” setting, based on the The effect of changes in static pressure on the zero and
actual flow measurement. span are noted by Pz and Ps. They are evaluated as the con-
If the conventional magnetic flowmeters and d/p cells are sequence of the physical distortion caused by 2000 psig (138
considered, and if they are evaluated on a basis that is slightly bars) of operating pressure. For the purposes of this example,
more conservative than Basis 2 but less conservative than it will be assumed that Pz = ±0.25% of maximum range, and
Basis 1, the resulting loop errors are as shown in Figure 1.4n. Ps = ±0.5% of actual reading.
From the data in Table 1.4m and Figure 1.4n, it can be For the purposes of this example, assume a d/p cell with a
concluded that neither error nor inaccuracy is by any means maximum range of 0 to 750 in (0 to 19 m) H2O and an actual
a clearly defined single number and that the required range- span of 0 to 100 in (0 to 2.54 m) H2O. It is further assumed that
ability of the measurement has a substantial impact on perfor- the actual operating temperature of the d/p cell is within 50°F
mance. Therefore, a meaningful accuracy statement should (18°C) of the temperature at which the unit was calibrated and
answer the following questions: (1) What portion of the total that the actual operating pressure is 1000 psig (69 bars). When
error is the precision (random error) of the sensor? (2) Is the the process measurement is 100 in. (2.54 m) H2O, the above
sensor error based on full scale (FS) or on actual reading assumptions will result in the following error components:
(AR)? (3) Over what range of measurement values is the
error statement applicable? E =
±0.2%
Tz =
0.5 (750 in./100 in.) (50°F/100°F) = ±1.875%
Ts =
0.5 (50°F/100°F) = ±0.25%
Pz =
0.25 (1000 psig/2000 psig) (750 in./100 in.) =
System Inaccuracy Based ±0.9375%
on Actual Reading Ps = 0.5 (1000 psig/2000 psig) = ±0.25%
(± %)

5.0
If we calculate the total error (Et) as being the square
root of the sum of the square of the individual errors, the
result is:
4.0

3.0 Et = (0.22 + 1.8752 + 0.252 + 0.93752 + 0.252 ) = ± 2.13%


Nonlinear
1.4(1)
2.0

Linear From the above example, one might note that the largest
1.0 contributions to the total error are the zero shifts caused by
Digital
the pressure and temperature differences between the cali-
Actual Flow
bration and the operating conditions. These errors can be
(GPM or %)
0 20 40 60 80 100 reduced by selecting a d/p cell with a maximum range that
is closer to the actual reading. One might also note that the
FIG. 1.4n
total error (Et) would have been even higher if the actual
Total loop inaccuracies as a function of sensor type and flow rate,
calculated on the basis of equation 1.4(1), where the total loop error measurement did not correspond to 100% of the actual span
is obtained by taking the square root of the sum of the component (100 in H2O), but only some fraction of it.
errors squared. (Accuracy in simple flow measurement. TI-1-30a. It should also be noted that the above Et value is not
The Foxboro Company.) the total measurement error of the loop but only the error

© 2003 by Béla Lipták


1.4 System Accuracy 85

contribution of the d/p cell. Finally, one should note that more accurate than the system being calibrated. (c) Peri-
one advantage of the “smart” transmitters is their ability to odic recalibration is a prerequisite to good control.
reduce the pressure and temperature effects on the span and 5. Instrumentation worth installing should also be worth
zero. Therefore, if E is ±0.1% in an intelligent transmitter, keeping in good condition. The performance of all sen-
the total error Et can be kept within about ±0.3%. sors is affected by corrosion, plugging, coating, and
process property variations. Therefore, scheduled main-
tenance is required to guarantee reliable operation.
REPEATABILITY VS. TOTAL ERROR In summary, (a) inaccuracy should be stated as a function
of rangeability, (b) multicomponent systems require system
Based on the information presented above, the following calibration, and (c) maintaining good performance requires
qualitative conclusions can be drawn: periodic recalibration and scheduled maintenance.
1. Inaccuracy is likely to be improved by reducing the
number of components in a measurement loop. References
2. Inaccuracy statements are meaningful only when given
in combination with rangeability. The wider the range- 1. Kemp, R. E., Accuracy for Engineers, Instrumentation Technology,
ability required (expected load variations), the more Inc., Painesville, Ohio.
inaccurate the measurement is likely to be. Furthermore, 2. Rudbäck, S., Optimization of orifice plates, venturies and nozzles,
the rangeability effect on digital systems is the least; it Meas. Control, June 1991.
increases when linear analog system are used, and it is
the highest in case of nonlinear analog systems.
Bibliography
3. On nonaccounting systems, the interest is focused on
repeatability (random error) and not on total inaccu- Applicable standards: DIN/IEC Standard #770 and ASME PTC19.1.
racy. The repeatability of most measurement loops is Englund, D. R., Loading Effects in Measurement Systems, Instrument and
several-fold better than their total error. Control Syst., February 1970, 63–68.
4. Instrumentation worth installing is usually also worth Shinskey, F. G., Estimating System Accuracy, Foxboro Publication #413–5,
calibrating. In this regard, several points should be Invensys Systems, Inc., Foxboro, MA.
Taylor, B. N. and Kuyatt, C. E., Guidelines for Evaluating and Expressing
made: (a) The accuracy of a multicomponent system the Uncertainty of NIST Measurement Results, NIST Technical Note
is unknown unless it is calibrated as a system. (b) The 1297, NIST, Gaithersburg, MD, 1994.
calibration equipment used must be at least three times Vom Berg, H., What is accuracy? Meas. Control, April 1991.

© 2003 by Béla Lipták