Schusterweg 1 * D-85375 Neufahrn * Germany

REPEATABILITY REPRODUCIBILITY LINEARITY Tel: +49-8165-64 72 0 * Fax: +49-8165-64 72 29

Flowmeter Repeatability, Reproducibility, Linearity

by: A. Trigas
TrigasFI GmbH, Dec. 2008

Repeatability is defined as the ability of a flowmeter to reproduce a measurement while being tested
under an unchanging set of conditions. For example, testing would be performed by fixing the flowrate
and stabilizing ambient conditions and then taking point after point without changing the flow rate setting.
It is not to be implied that the indicated flow is correct, but rather that the indication is the same every time.

Reproducibility is in every respect the same as repeatability with one significant exception. During
reproducibility testing, the flow rate is changed and then reset between repeats. Reproducibility is
therefore a more realistic indication of a flowmeter’s ability to reproduce a measurement whenever a
predefined set of conditions is recreated.

The terms Reproducibility and Repeatability are some times used interchangeably, but since the former is
a more rigorous form of testing, it is some times avoided by marketers determined to present their product
in the best light possible.

The results from either form of testing must be analyzed using valid statistical methods. The method of
choice in the flow measurement field is Student’s t analysis with 95% C.L. (Confidence Level).
Description is attached in the appendix.
In order to generate a useful indication of the true repeatability/reproducibility characteristics of a
flowmeter, a large number of repeats must be taken for each set of test conditions. It is recommended
that 30 repeats are taken in order to reduce the value of the t coefficient (uncertainty level).

Obviously 30 repeats at a large number of different sets of conditions would not be practical to perform
routinely. Fortunately however, flowmeter performance is usually predicable, consistent and linear. This
holds especially true for Turbine flowmeters. As a result, it can easily be shown that repeatability
characteristics throughout the entire range of a flowmeter can be evaluated by testing at a limited number
of test conditions and even at a single flowrate.

It is recommended that testing is performed at the flowrate of approximately 25-35% of the maximum
flowrate of a given flowmeter. In the attached sample results from testing of an ANC 12 flowmeter which
has a range of 3.8 to 114 lpm, we have chosen 35 lpm. This selection allows testing in the linear area of
the meter and at the same time reasonably close to the bottom end of the flow range where repeatability
problems would be more easily detectable.

The attached sample repeatability test results for an ANC 12 demonstrate the above concepts. As can be
seen, Student’s t analysis shows that determination of repeatability based on 4 repeats is unreliable and
can vary by a factor of 3 (0.027 to 0.113% in the example). The same test results used as a group of 30
yielded a reliable and believable measure of Repeatability of 0.051% with 95% C.L.

Tests for the same flowmeter yielded a value for Reproducibility of 0.055% again with 95% C.L. As
anticipated, it was slightly worse than the repeatability results. One would expect that the difference is the
result of the greater variability of the test conditions which is inherent to testing where conditions are
intentionally varied and then reset.
But is this really the case?

Linearity is another important characteristic of flowmeter performance. It is a measure of how close the
relationship between flow and flowmeter output approximates a linear function. For turbine flowmeters, a
constant K-factor (pulses per unit volume) is indication of linearity and the reason why turbine meter
performance is plotted as K-factor vs flowrate. In this representation, a horizontal line represents ideal
linearity.

Page 1 of 4
 Schusterweg 1 * D-85375 Neufahrn * Germany
REPEATABILITY REPRODUCIBILITY LINEARITY Tel: +49-8165-64 72 0 * Fax: +49-8165-64 72 29

By analyzing the reproducibility test results graphically, it can be clearly seen that the K-factor values
increase as flowrate increases. Consequently, some of the 0.055% reproducibility error is due to the non
linear nature of the flowmeter performance.

In order to separate the systematic linearity error from the random reproducibility error the following
method has been used as demonstrated in the attached graph of the ANC 12 reproducibility data. A least
square line was created with the following formula:

y=ax+b or K=aQ+b Where: K - K-factor, pul/lit

Q - Flowrate, lit/min

The constants were determined to be: a = 1.421 (slope)

b = 719.875 (intercept)

The deviation of the individual reproducibility points from this line were then calculated and it was then
again estimated that the pure Reproducibility error is 0.043% with 95% C.L.

This means that the original random reproducibility error calculation actually contained a systematic
Linearity error of 0.012% with 95% C.L., or more that a fifth of the total. If not separated by the methods
described above, the contribution of linearity to the error calculation is much more significant and in fact
becomes predominant when testing is performed in the non-linear region of a flowmeter’s range.

This is an important contributing factor to the often mistaken belief that flowmeters exhibit deteriorating
repeatability/reproducibility as flowrate decreases.

ANC 12 Reproducibility Tests

770.0

769.9

769.8

769.7
K-Factor (pul/lit)

769.6

769.5

769.4

769.3

769.2

769.1

769.0
34.75 34.80 34.85 34.90 34.95 35.00 35.05 35.10 35.15

Flowrate (lit/min)

Page 2 of 4
 Schusterweg 1 * D-85375 Neufahrn * Germany
REPEATABILITY REPRODUCIBILITY LINEARITY Tel: +49-8165-64 72 0 * Fax: +49-8165-64 72 29

ANC 12 Reproducibility test Revised data: Deviations calculated from best straight line
fit
(flowrate reset between
repeats)
Estimated Error Estim. Delta Estimated Error
Freq. Rate K-factor Std. 95% C.L. 99% C.L. Straight K-Factor K- Std. 95% C.L. 99% C.L.
Dev. Factor Dev.
Hz lpm p/lit 4 consecutive samples running Line Fit p/lit 4 consecutive samples running

449.130 35.020 769.500 Slope: 769.649 -0.149

447.568 34.902 769.425 1.421 769.481 -0.056
450.536 35.120 769.704 Intercept 769.791 -0.087
449.150 35.027 769.381 0.124 0.051% 0.094% 719.875 769.659 -0.277 0.085 0.035% 0.065%
450.684 35.126 769.835 0.190 0.078% 0.144% 769.799 0.035 0.114 0.047% 0.086%
449.236 35.028 769.513 0.174 0.072% 0.132% 769.660 -0.146 0.113 0.047% 0.085%
448.277 34.959 769.386 0.184 0.076% 0.140% 769.562 -0.176 0.113 0.047% 0.086%
450.369 35.103 769.789 0.187 0.077% 0.142% 769.767 0.021 0.095 0.039% 0.072%
448.095 34.938 769.524 0.146 0.061% 0.111% 769.533 -0.008 0.085 0.035% 0.064%
446.947 34.866 769.142 0.234 0.097% 0.178% 769.430 -0.288 0.126 0.052% 0.096%
450.196 35.087 769.851 0.279 0.115% 0.212% 769.744 0.106 0.148 0.061% 0.112%
448.640 34.977 769.595 0.254 0.105% 0.193% 769.588 0.007 0.147 0.061% 0.111%
447.166 34.871 769.416 0.259 0.107% 0.196% 769.437 -0.021 0.146 0.060% 0.111%
449.997 35.073 769.819 0.177 0.073% 0.134% 769.724 0.095 0.055 0.023% 0.042%
449.212 35.020 769.634 0.143 0.059% 0.109% 769.649 -0.015 0.047 0.019% 0.035%
447.427 34.884 769.564 0.145 0.060% 0.110% 769.456 0.108 0.060 0.025% 0.045%
450.397 35.105 769.792 0.107 0.044% 0.081% 769.770 0.022 0.051 0.021% 0.039%
448.366 34.955 769.621 0.085 0.035% 0.064% 769.556 0.064 0.046 0.019% 0.035%
448.045 34.935 769.505 0.107 0.044% 0.081% 769.528 -0.024 0.049 0.020% 0.037%
449.377 35.026 769.799 0.123 0.051% 0.094% 769.657 0.142 0.061 0.025% 0.046%
447.908 34.916 769.696 0.107 0.044% 0.082% 769.501 0.195 0.083 0.034% 0.063%
447.287 34.888 769.232 0.216 0.089% 0.164% 769.462 -0.230 0.166 0.069% 0.126%
449.346 35.021 769.850 0.245 0.101% 0.186% 769.650 0.200 0.179 0.074% 0.136%
448.192 34.945 769.538 0.229 0.095% 0.174% 769.542 -0.005 0.177 0.073% 0.134%
446.876 34.864 769.058 0.302 0.125% 0.229% 769.427 -0.370 0.217 0.090% 0.165%
449.510 35.043 769.633 0.290 0.120% 0.220% 769.682 -0.049 0.204 0.084% 0.155%
446.651 34.823 769.571 0.229 0.095% 0.174% 769.370 0.201 0.205 0.085% 0.155%
449.005 34.993 769.871 0.296 0.123% 0.225% 769.611 0.260 0.248 0.103% 0.188%
448.156 34.936 769.684 0.112 0.046% 0.085% 769.529 0.155 0.116 0.048% 0.088%
446.338 34.797 769.621 0.114 0.047% 0.086% 769.332 0.289 0.052 0.021% 0.039%
Mean K 30 samples results Mean K 30 samples results
769.585 0.206 0.055% 0.074% 769.585 0.162 0.043% 0.058%

20-Nov-01 55162c-3192

Page 3 of 4
 Schusterweg 1 * D-85375 Neufahrn * Germany
REPEATABILITY REPRODUCIBILITY LINEARITY Tel: +49-8165-64 72 0 * Fax: +49-8165-64 72 29

Error Calculation Using Student’s t

t = (X-µ)/σ = Student’s t

Where X = Sample Value

µ = Mean of Population
σ = Standard Deviation of Population

s = [Σ(Xi-Xm)2/ν]1/2 = Standard Deviation of Sample Population

Where Xi = Individual Sample Value

Xm = Mean of Sample Population
ν = n-1 = Degrees of Freedom
n = Number in Sample

% error = +/- t*s

Sample t values are given in the table below:

Samples ν=n-1 t with 95% C.L. t with 99% C.L.

2 1 12.71 63.66
4 3 3.18 5.84
30 29 2.04 2.76
  1.96 2.58

Example:

From attached ANC 12 Reproducibility test data:

Sample population n = 30
ν = 30-1 = 29
Consequently t = 2.04 for 95% C.L.

Population Mean Xm = 769.6

Sample Std. Dev. s = 0.206

Error = t*s = 2.04*0.206 = +/- 0.420

% Error = t*s*100/Xm = 2.04*0.206*100/769.6 = +/- 0.055% with 95% C.L.

Repeatability Reproducibility Linearity

Page 4 of 4