Sei sulla pagina 1di 9

Uncertainty Analysis of Turbine and Ultrasonic Meter Volume Measurements Part 2, Advanced Topics THOMAS KEGEL, Senior Staff

f Engineer, COLORADO ENGINEERING EXPERIMENT STATION, INC., NUNN, CO

ABSTRACT This paper continues from the first paper1 in describing the process of estimating the uncertainty of volume measurements made with turbine and ultrasonic flowmeters. Components that contribute uncertainty include the pressure and temperature transducers, the gas chromatograph, state equation and flow computer as well as the meter itself. In the first paper each component was described and numerical uncertainty values were estimated based on a hypothetical set of measurements. The individual component values were combined to provide the uncertainty in the total volume. The present paper discusses uncertainty issues associated with calibration, it is organized based on two examples. The first example concerns pressure transducer calibration, the second example discusses flowmeter calibration. Topics include short and long term random effects, percent-of-reading effects, and full-scale effects Additional discussion covers proper interpretation of flowmeter calibration results and guidance to replace manufacturer specifications with calibration results. STANDARD UNCERTAINTY AND STANDARD DEVIATION A simplified five-step uncertainty analysis procedure was described in the first paper. One of those steps is to identify and then classify the numerical values of uncertainty. In order to combine the uncertainties in the individual components they have to be defined in a uniform manner. The standard uncertainty2 is the term given to the uniform method of expressing numerical values of uncertainty. In order to estimate a standard uncertainty value, a component uncertainty is classified based on how the numerical value is determined. An estimate is classified as Type A when statistical data are available. The standard uncertainty is defined as the statistically determined standard deviation3. An estimate is classified as Type B when statistical data are not available. The standard uncertainty is defined as u = 0.58U where U defines the limits within which the true value is expected to lie. In some cases the standard uncertainty of a Type B estimate is u = 0.50U if it is known to be based on a statistical analysis. In the first paper most of the standard uncertainties were classified as Type B, in the present paper Type A uncertainty estimates are discussed in greater detail. Given a set of n data points, the mean is defined as:

1 x= n

xi
i =1

[1]

and the standard deviation is defined as:

s=

1 n 1

(xi x )2
i =1

[2]

The n data points are samples of a population, if n is sufficiently large then 95% of the data s points lie within 2s of the population mean. The sample mean, x , lies within of the n 1 population mean.
PRESSURE TRANSDUCER CALIBRATION

A simple example has been formulated to demonstrate some important statistical principles. The subject is a fictitious pressure transducer that is used over a 1001000 psi range. The output is a voltage that varies over the 05 volt range; the nominal sensitivity is therefore 200 psi/volt. The transducer is calibrated every 90 days, a history has been developed based on 16 calibrations made between January 1999 and December 2002.
Long and Short Term Random Effects

Each calibration involves determining the transducer sensitivity based on values calculated from pressure and voltage standards. Sensitivity data are obtained at ten pressure levels equally spaced over the input range, typical calibration results contained in Figure 1. The ten sensitivity values that make up a calibration are used to calculate a mean and standard deviation, values are given in Figures 2 and 3. Each symbol represents one calibration; the solid lines are described later in the paper. Two values of standard deviation, with different interpretation, are calculated from the calibration data to form the control limits. A pooled value of calibration standard deviations, called sw for within, represents the average variation observed in the time required to perform a ten-point calibration. The term pooled refers to a process for combining multiple standard deviation values3. The second standard deviation, called sb for between, represents random effects that are only observed over long periods of time. It is calculated as the standard deviation of the mean values. The duration of long and short periods of time depends on the application. In the present example short represents the time required for a single calibration and long represents four years. The reported standard deviation, sr, accounts for both short term and long term random effects4. The measurement uncertainty will be underestimated both effects are not included. The value for sr is calculated from values of sw and sb combined in quadrature (root sum square):
s r = s b 2 + s w 2 = 0.32 2 + 1.032 = 1.08 [psi/v] [3]

The interpretation of sr is as follows: all of the random effects associated with the measurement process amount to (21.08) = 2.16 [psi/v] with a confidence level of 95%. The 2 term means that two standard deviation values are required to achieve the 95% level of confidence. If the sb term is not accounted for the uncertainty is underestimated, sr = 1.03 instead of 1.08. While this omission results in a minor difference in the current example, it could be more significant in other applications.
Statistical Process Control

In addition to providing estimated uncertainty values, the above analysis is used to develop control charts that monitor process consistency. The application of this well developed technique

to the measurement process is described in Reference 4, a pressure transducer applications in Reference 5. The solid lines in Figs. 2 and 3 represent control limits; the calculation procedure is described in the references. All the data points are between the lines in Fig. 2 and below the single limit of Fig 3. Operation within the control limits indicates that the transducer calibration process is operating in a state of statistical control. This means that the observed variations are due only to the inherent random effects that are part of the process.
Separation of Components

The ability to separate short and long term random effects is important in some applications. For example, laboratory testing of flowmeter installation effects involves one test at baseline conditions and a second test at distorted conditions. The installation effects associated with the distorted conditions are calculated as the difference in flowmeter performance. The uncertainty associated with the difference between two measurements made with the same instrument will include only short term random effects provided the measurements are made during a short time.
Systematic Effects

The reported (sr) value of standard deviation includes all random effects present during calibration. It is typically assumed that the same effects will be present when the transducer is used to make a measurement. The combined uncertainty in pressure measurement includes systematic effects as well as the reported standard deviation. Systematic effects contrast with random effects in that they do not contribute random variability in the final result. There are three systematic effects present in a typical pressure measurement. The first is the calibration standard, usually a deadweight tester. The second is the data acquisition system, usually a flow computer. The third systematic effect accounts for ambient temperature effects present when the transducer is used. All three effects are determined from manufacturer specifications.
FLOWMETER CALIBRATION

The previous example concerned a pressure transducer; this section contains an example based on a flowmeter calibration. The previous discussion concerning long and short term effects are fully applicable to flowmeter calibrations, a recent example is described in Reference 6. There are a number of differences between pressure transducer and flowmeter calibration. First, the flowmeter contributes more uncertainty than the pressure transducer. The uncertainty associated with a good pressure transducer might be 0.075% while that associated with a flow meter can be as high as 1%. Second, the pressure typically shows less variation than the flowrate in a when installed in the field. It is more important to characterize the flowmeter over the operating range. Finally, a flow calibration is more costly than a pressure calibration, especially with a large meter. It is important that the flow calibration provides as much information as possible.
Full Scale and Percent Reading Effects

The performance of any instrument is characterized by both full-scale and percent-of-reading effects. A percent-of-reading effect acts equally over the entire range while a full-scale effect dominates at the low end of the range. The manufacturers specifications in Fig. 1 of the first paper1 contained both effects; they are also present in ultrasonic and turbine flowmeters. Typical data contained in Figure 4, the solid lines represent the 95% confidence interval, which is the statistical interval that contains 95% of the data. The interval width varies with flowrate due to full-scale effects associated with the meter. A properly designed calibration process needs to determine both the full-scale and percent-of-reading effects

A typical flowmeter calibration begins with the selection of several velocity or flowrate points. Multiple samples are obtained under steady conditions (flow, pressure, temperature). The total sample time, the sample rate and the total number of samples are dependent upon cost and uncertainty issues associated with the operation of the lab. These issues are discussed in this section based on an example. Assume that calibration conditions are held steady and 512 samples are obtained. In the present example a set of random numbers simulate the multiple samples. The standard deviation of the 512 samples (random values) is calculated to be 0.128%. The interpretation of standard deviation is as follows: if the process continues to hold steady, 95% of any additional samples will lie within 20.128 = 0.256% of the mean value (of the 512 samples). The example continues by grouping the samples, for example, 16 groups of 32 samples each. A mean value is calculated for each group. A standard deviation is calculated for the 16 group means, the value is sb = 0.061%. The interpretation of this standard deviation is as follows: if the process continues to hold steady, 95% of any additional groups will lie within 20.061% = 0.122% of the mean value (of the 16 groups). Each of the 16 groups of 32 samples is characterized by a standard deviation; they range in value from 0.099% to 0.141%. An average value of sw = 0.116% is calculated for the current example. The grouping of data within the present example results in the same two types of standard deviation as with the pressure transducer example. The standard deviation, sw, accounts for random effects within a group while sb accounts for random effects between groups. The present example continues further by organizing the 512 data points into different sized groups and calculating standard deviation values. The results are summarized in Table 1. The first two columns identify the quantity and size of sample groups. The final three columns contain values for the three standard deviations. The lowest reported values (sr = 0.130%) are observed when the data are not grouped. Slightly larger reported values are observed for the shaded rows (6-16 groups, 32-85 samples per group). If either the group size or the samples per group becomes too small, the reported value increases. This increase is a result of statistical considerations beyond the scope of this paper.
Time Scales

Various time scales can be associated with the 512 samples; the same uncertainty considerations apply regardless of the time frame. If the samples were obtained at a rate of between one and five seconds per sample the total time would be between 500 and 2500 seconds (between 8 and 40 minutes). This range of time scales would represent a typical calibration process. In a calibration the term group is replaced by the term data point. If 512 samples were obtained in ten minutes, eight data points based on 64 samples each would be reported. A sample rate between 5 and 10 samples per second would result in a total time of between 50 and 100 seconds to obtain 512 samples. A group of samples might represent, for example, samples leading to a single measurement made by the voltmeter that is part of the flow computer. The sw value would form part of the uncertainty associated with the voltmeter and sb would represent the variation in the voltage being measured. Finally, the samples could be obtained over a long period of time, perhaps many months. Each group of samples would represent a complete calibration. This scenario is similar to the pressure transducer example previously described. There are significant differences between the calibration of a flowmeter and the use of that flowmeter in the field. Calibration data are obtained for a limited time involving a very small volume of gas. The flowrate is the parameter of interest and it is controlled to a steady value as

data are obtained. Operation in the field is based on total volume measured over a long time. The flowrate is less important a parameter than the total volume, and it can vary depending on the application. The correct interpretation of the calibration sample time depends on the application of the flowmeter. The optimum sample time is a value that represents the shortest time period under which the meter will measure steady flow. In general larger meters exhibit slower changes than smaller meters and custody transfer applications exhibit smaller changes than production applications. When the calibration sample time is appropriate to the meter application, the contribution of sw can be reduced by the term

1 where n is the number of samples that n 1

make up the data point. This reduced value of uncertainty is associated with the mean value of the n samples. This interpretation of sample time is illustrated in Figure 5. The thick line represents flowrate changing over time. The initial slope is steeper than the final slope indicating a larger initial change in flowrate over time. Two boxes are superimposed on the flowrate vs. time curve. The height of each box represents a small range of flowrate that is considered steady. The width of each box is subdivided into equal intervals, each representing one calibration sample time interval. The initial slope is considered steady for four calibration samples while the final slope is considered steady for thirteen calibration samples. The sw value resulting from the calibration of the flowmeter can be reduced by the following values: 1 = 0.577 if the initial slope represents typical flowmeter field operation 4 1 1 = 0.289 if the final slope represents typical operation 13 1
SUMMARY

The paper used two numerical examples to illustrate uncertainty concepts associated with calibration. The mean and standard deviation of a series of samples were defined. Several interpretations of the standard deviation were discussed. The standard deviation of the mean differs from that of a single sample. Long and short term random effects are associated with varying time scales. Full-scale effects vary over the range of an instrument. Percent-of-reading effects are consistent over the range. The paper concluded with a discussion of how to interpret the uncertainty of calibration results.
REFERENCES

1. Kegel, T. M., Uncertainty Analysis of Turbine and Ultrasonic Meter Volume Measurements, AGA Operations Conference, Orlando, FL, May, 2003. 2. ___, Guide to the Expression of Uncertainty in Measurement, International Organization for Standardization, 1994. 3. Wadsworth, H. M., Handbook of Statistical Methods for Engineers and Scientists, McGrawHill, 1990. 4. Croarkin, Carroll, Measurement Assurance Programs, Part 2: Development and Implementation, NBS Special Publication 676-2, 1985.

5. Kegel, T. M., Statistical Control of a Differential Pressure Instrument Calibration Process, 45th International Instrumentation Symposium, Albuquerque, New Mexico, May, 1999. 6. Kegel, T. M., Quality Control Program of the CEESI Ventura Calibration Facility, FLOMEKO, Gronigen, The Netherlands, May 2003.

203

Sensitivity [psi/volt]

201

199

197 0 200 400 600 800 1000 Applied Pressure [psia]

Figure 1: Typical Pressure Transducer Calibration Data

203

Mean [psi/v]

201

199

197 Nov-98

Nov-99

Nov-00 Date

Nov-01

Nov-02

Figure 2: Pressure Transducer Calibration History Mean Values

1.6
Standard Deviation [psi/v]

1.2

0.8

0.4 Nov-98

Nov-99

Nov-00 Date

Nov-01

Nov-02

Figure 3: Pressure Transducer Calibration History Standard Deviation Values

Table 1: Standard Deviations of Grouped Samples


Quantity Groups 512 256 128 64 32 16 10 8 6 5 4 3 1 Samples per Group 1 2 4 8 16 32 51,52 64 85,86 102,103 128 170,171 512 Between Standard Deviation 0.098 0.077 0.066 0.060 0.057 0.057 0.057 0.059 0.063 0.058 0.064 Within Standard Deviation 0.120 0.120 0.119 0.119 0.119 0.119 0.119 0.118 0.118 0.120 0.119 Reported Standard Deviation 0.130 0.155 0.143 0.136 0.133 0.132 0.132 0.132 0.132 0.134 0.133 0.135 0.130

0.4

0.2
Residual [%]

0.0

-0.2

-0.4 0 20 40 60 Velocity [ft/s] 80 100

Figure 4: Typical Ultrasonic Flowmeter Calibration Data

4 CALIBRATION SAMPLES 13 CALIBRATION SAMPLES

FLOWRATE VS TIME CURVE

Figure 5: Interpreting Calibration Sample Rate

Potrebbero piacerti anche