Fluid flow measurement symposium

Anchorage, 2009

A NOVEL CONCEPT OF A HIGH ACCURACY CALIBRATION RIG

A. Pfau, H. Hfelfinger, J. Honegger, J. Gfeller Endress + Hauser Flowtec AG Kgenstrasse 7, CH-4153 Reinach, Switzerland Tel. +41-61-715 6111, [email protected] Abstract: Within a project period of one and a half years a substantially improved calibration facility at Endress+Hauser Flowtecs production centre in Reinach, Switzerland, was developed, built and commissioned. The first run with a meter under test happened in March 2007. In August 2008, the facility was accredited according to ISO/IEC 17025 to within 0.015% best uncertainty covering a flow rate of 0.7t/h 432t/h, with gravimetric reference (79kg/h 1.2t/h, 0.021% with volumetric reference). The stated objective was to provide a facility capable of operating to a very high level of accuracy, to be employed to calibrate customers Promass 83/84F Coriolis mass flow meters fully traceable, with improved specification of 0.05%. To achieve this, a number of elements of two existing calibration rigs had to be analyzed and improved. To guarantee this low level of uncertainty of the facility, frequent calibrations-, and if necessary adjustment of the system to minimal deviation as compared to the reference (weighing system) are implemented. For that purpose, certified weights of the OIML class F2 are permanently located at the sides of the weighing tanks. Based on this concept, the calibration can be carried out fully automatically every two weeks. In this presentation the key elements of this facility, the achieved performance and the traceability to the national standard will be outlined.

1. INTRODUCTION Stricter process safety and product quality requirements not only demand highly accurate flow meters with long-term stability, but also highprecision calibration rigs Coriolis mass flow meters Promass as the most successful measuring instruments in key industries like Life sciences, Chemicals, food, Oil & Gas in the following fields of application: quality-specific measuring points in production and accurate measurement of expensive raw materials and active ingredients. Thanks to continuous improvement in technology and design, a new optional accuracy specification for Promass 83/84F to +/- 0.05% could be introduced. Up until now, there have been no traceable production calibration rigs in the world where this high level of accuracy could be proven and verified. Our own rigs ( 0.05%) were also not able to provide this verification end to end (turn down). The higher accuracy specification of those flow meters required calibration facilities, which could be utilized to calibrate them fully traceable, based on the laboratory accreditation standard ISO/IEC 17025, not least to increase credibility by the approval of an independent organization. From a production point of view it makes a big difference whether a Vortex flow meter, an

Ultrasonic meter, a magnetic-inductive, a thermal mass or a Coriolis flow meter is calibrated. These are flow meters with measurement principles, which are offered by Endress + Hauser Flowtec. With the exception of thermal mass flow meters, all other principles are calibrated with water as calibration fluid. Therefore, Endress + Hauser Flowtecs water calibration rigs are multi functional, i.e. capable of calibrating all types of flow meters at the shortest production time possible and at the rig uncertainty level needed. The offered types of meters cover different levels of specifications. Some being volumetric flow meters, some being mass flow meters. Therefore, E+H Flowtecs calibration rigs comprise basically three modes of operation: one using the gravimetric method with flying start stop technique (scale, diverter) the other using master meter method (Coriolis master meters) or the volumetric calibration method (pipe prover). In addition, the calibration rigs are highly automated in terms of process control and data acquisition in order to keep the throughput of devices high. If low uncertainty is needed, i.e. a Coriolis meter is calibrated, the scale is always used. If uncertainty is less critical, i.e. devices of all other measurement principles are calibrated, production time can be optimized by using master meters. At very low flows, i.e. below 0.1kg/s pipe provers are utilized.

Fluid flow measurement symposium

Anchorage, 2009

This rather expensive rig set up stems from the global strategy of being able to produce and calibrate close to the global markets. This double equipped calibration rig design has several key advantages, which contribute to the ability of achieving a measuring uncertainty as low as 0.015%. 2. THE CALIBRATION RIG A simplified installation layout is given in Fig.1. The water loop starts with a tank of 600m3 capacity. Speed controlled pumps deliver a flow range of 0.2 to 120 kg/s. The size of the tank ensures stable temperature conditions during a calibration cycle. The water flows through the master meters of four different line sizes before reaching the measurement section. Each master line size is constructed with two Coriolis meters in series. The line sizes of the DUTs (Device Under Test) range between DN40 and DN250. The pressure in the measurement section is controlled. Downstream of it the flow control valve and the scales are situated. The rig is equipped with two scales of 400kg and 4t. The switch point between the two scales is 10kg/s. Supplier of the scale measurement equipment was HBM. The gravimetric method is realized in a flying start stop mode, i.e. a diverter channels the water into the scales container during the measurement time. The diverters are activated by a stepping motor in case of the 4t scale and by a pneumatic cylinder for the 400kg scale. Both actuators ensure a fast and repeatable switching time.

uncertainty elements with higher influences, instead of trying to improve values close to zero. The uncertainty factors with potential for improvement were for instance, the weighing system in general, the scale adjustment, as well as the flow diversion part.

Diverter

Scale Adjustment Scale

30 Uncertainty contribution [%] 25 20 15 10 5 0

Fig 2 Relative contribution of different uncertainty elements 4. RIG IMPROVEMENTS The first point to mention is not an improvement in itself, but a definition of measurement quantity: the calibration process is optimised for mass flow measurement. Volume effects, e.g. compressibility or thermal expansion, of the calibrating fluid have no influence. The measurement uncertainty is evaluated for mass flow as measured variable only. This restriction reduces the uncertainty of certain contributors, for instance the inter space pipe between DUT and the scale (see 5.2). 4.1 Calibration weights The original calibration weights were made out of cast steel fulfilling the OIML uncertainty class M1. The new weights are made out of precision machined and polished stainless steel, and comply with the OIML class F2. The improvement in uncertainty is given in Table 1. The weights are protected against dust and dirt by cover hoods made out of Plexiglas. The weights come with a calibration certificate and are traceable to the national standard.
S tandard Weight [kg] OIML Classes Tolerance 20 M1 1g 50 500 M1 25 g PremiumCal OIML Classes Tolerance F2 F2 0.8 g 8g

DUT
DN80 DN50 DN25

A A

4t scale

400kg scale
500 kg 500 kg 50 kg 50 kg

Master meters

Tank

Fig. 1 Simplified installation plan: FCP7.1.5

3.

Uncertainty Analysis

For the improvement of the measuring uncertainty of the facility, we intended to concentrate on the

DN150

DN150

Un ce Co Res rta rr olu int ec tio y o t io n f o n D Cou ne ive n Di rte ter ve r E Re Drif r ter rro Li so t D sw r ne lu ar tio ive itch isa tio Sca n S rter l c n de e ad ale er ro v. r d j fro ustm ispl m e ay Re ad nt E pr jus r r od t. or u Te Cal cib Poin m ibr ilit ts pe at yS ra io n c Li ture in ale n t Te ear dri er fa m iy ft l ce pe Sc oa ra al d c tu e re am ell dr pl ift ifie Sc am r Te al p m ea pe d Sc lifie ra jus ale r tu tm d re en rift ch an t Er ge ror De in t ns ub Dd en De ity e sit ns ste ya ity el ir du D w at r in en er s g ad ity a jus ir tm en t

Table 1 Improved calibration weights

Fluid flow measurement symposium

Anchorage, 2009

4.2 4t scale The scale is equipped with nine calibration weights 500kg each. The vessel is positioned on three RTN load cells of the accuracy class C5. In the cases of weight load and water load, the centre of gravity of the vessel always is situated in the middle of the three load cells. The calibration weights are actuated with hydraulic cylinders and perform an up and down movement. Putting weights down on the vessel resulted in a hit on the load cells. The characteristic of a load cell is sensitive to mechanical hits, which in turn deteriorates the reproducibility of the scales calibration. Therefore, the load increase while putting the weight down was damped by introducing springs into the weight mounting claw. Instead of a hit the load cell now sees a linear increase of force during a period of several seconds. The reproducibility of the scale improved from 18g to 10g as given in Table 2. Load cells are susceptible to temperature changes of the ambient. Each load cell has an individual sensitivity to temperature changes even in the highest level of load cell classes. This is due to the applied temperature compensation inside the load cell. Since the scale is equipped with three cells and since positive and negative temperature sensitivity of the load cells occurs, the three load cells can be combined such, that the combined temperature effect of all three load cells is minimal. The technical specification of HBMs RTN C5 load cells gives 0.0062%/10K as temperature effect. Combining three load cells accordingly reduces this influence to 0.0001%/10K. 4.3 400kg scale The standard 400kg scale is characterised with - Single point mounting of vessel - Single load cell (HBM Z6FC4) - Calibration weight handling by hand: 20 weights of 20kg each. - No spring damping of the weight load Introducing automated weight handling in this case meant a veritable redesign of the scale. Consequently, a new mounting system of the vessel was designed, too. The over all changes list to - Three point mounting system of the vessel - Three load cells of better OIML class (HBM Z6FC6) - Automated calibration weight handling: 7 weights of 50kg each actuated by pneumatic cylinders - Spring dampened weight deposition All improvements contribute to a better reproducibility of the scale. It dropped from a value of nearly 4g to 1g as it is documented in Table 2.

400 kg scale 4 t scale

Standard scale R eproducibility [g] 3.9 18

PremiumCal scale R eproducibility [g] 1.0 10

Table 2 Improvements of reproducibility due to changes of the scale design The calibration weights allow for 350kg of maximum load. Therefore, talking about a 400kg scale is not exactly true. In case of a calibration of optimized uncertainty budget, the scale is never filled to a higher level than 350kg. This is ensured by the control software and will be discussed in the following section. However, if the rig operates under standard conditions fillings up to 400kg may occur. In this case the scales characteristics can be extrapolated to 400kg. The additional uncertainty contribution is taken into account in the standard uncertainty budget, which is not subject of this paper. In terms of temperature effects the choice of load cells of the C6 class reduced this effect from 0.007%/10K to 0.004%/10K according to the technical specifications.

4.4 Rig operation A large contributor to the uncertainty of the scale measurement is the characteristic of the load cell, i.e. its non-linearity. The 4t and the 400kg scales are adjusted at 9 respectively at 7 weight loads. Between the adjustment points the amplifier linear interpolates the characteristic. One way of reducing this error is to fill the vessel to the equivalent load of the adjustment points. This means, that the measurement time depends on the flow rate and the next adjustment load of the scale. The measurement time in addition must be greater than 30s as stated in [1]. The resulting rig operation map is depicted in Fig. 3.
1000

Through flow [kg/s]

100

3500kg

10

2500kg 1500kg

350kg 250kg 150kg 50kg

0.1 10 100 1000

Measurement time [s]

Fig. 3 Operation map: load, flow, measurement time

Fluid flow measurement symposium

Anchorage, 2009

Considering the 400kg scale at the lowest flow and the minimum load of 50kg the measurement time becomes as high as 250s. In case of the 4t scale the minimum load was chosen to be 1500kg, which results in a maximum measurement time of 150s. A side effect of increased minimum loads and increased measurement time is that the relative uncertainty of the scale measurement is decreased in comparison to the standard operation conditions. Standard operation conditions are minimum loads of 20kg for the 400kg scale and 300kg for the 4t scale and a minimum measurement time of 30s. The relative reduction due to the increased minimum load quantifies to the order of 0.01% The filling of the vessel into the adjustment point was found to be very reproducible for both scales. This results in a reduction of uncertainty due to nonlinearity of the load cells from 50g to 2g for the 4t scale and 5.5g to 0.2g for the 400kg scale. 4.5 Diverter One of the most important contributors to the overall uncertainty budget is the diverter unit. In [1] a procedure is described to determine the average diverter error. Two sets of experiments are performed. In the first the scale is compared against another reference, e.g. a flow meter. In the second the scale is filled again, but the diverter is switching n times in addition to the start and stop switching operation, with n>10. The difference of scale reading between the two experiments divided by the number of switching is the average diverter error. In literature the error is often expressed in terms of measurement time since it can be interpreted as a gate error. The gate signal links the water flow diversion into the vessel to the counting time of the DUTs pulses. The time error is linked to the equivalent in mass via the flow rate. Since it is more intuitive working with a scale as reference, the diverter error in this paper is expressed as mass error. In general, the diverter error depends on the flow rate. An easy way of reducing the relative contribution to the uncertainty is to increase the load of the scale. For low flows this has to be balanced with the increasing measurement time, i.e. production time. The calibration results can easily be corrected for the diverter error in the evaluation steps of the data acquisition software. The characteristics of the diverters being subject of this paper are determined in seven points each of their individual flow range. The diverter error is measured at least three times per flow rate as described above. The average of these three results is taken as the actual diverter correction value.

The diverters of both scales were designed to run within an uncertainty of 0.02% with a mechanically nicely adjusted gate and without any software correction. Using the diverter error correction the relative diverter error isnt exceeding 0.002% under the existing operation conditions. 5. MEASUREMENT UNCERTAINTY BUDGET

5.1 General In order to perform a calibration of a DUT all system components of the measurement chain must be under control and running correctly. To ensure this, several levels of periodical checks and adjustments of the system components must be set up. Components which are critical to the overall uncertainty budget are adjusted on a regular basis. The period of adjustment depends on the components long time behaviour, i.e. drift of characteristics. Less critical components are simply checked regularly but only adjusted if outside certain tolerances. The difference between periodical adjustment and periodical checks is given in Fig. 4. Having set up a system of periodical checks and adjustments the necessary data becomes available to describe system components in terms of uncertainty. A system component can be characterized by three parameters: - reproducibility - systematic error, e.g. non-linearity - drift
+ Tolerance

Error Periodical check

Time Periodical adjustment

- Tolerance

P1

P2

P3

P4

P5

P6

Fig. 4 Periodical check vs. periodical adjustment The traceability chain of a calibration consists of several procedures of comparison measurements. In Fig. 5 these steps of comparison measurements are framed with ellipses and highlighted with light blue. The chain starts way back with the national standard of weight whose accuracy is transferred to the calibration weights according to a certain class of uncertainty using comparison measurements. The calibration weights are used in a second measurement procedure (level of traceability) to determine the characteristics of the load cells of the scale. The last element of the traceability chain is the comparison of the DUT against the scale, i.e. the

Fluid flow measurement symposium

Anchorage, 2009

calibration of the DUT. In the case of a master meter calibration an additional level of traceability is implemented: that of comparing the master meter against the scale.

the correction between the periodical adjustments (drift). Both are evaluated taking the standard deviation of the corresponding parameters: deviation of the individual samples of the diverter error to the average and the differences of the correction curve from one periodical adjustment to the other. Both parameters depend on the through flow. The upper limits of each are given in Table 3.
Diverter correction Drift 400kg scale <3g <6g 4t scale <20g <40g

Table 3 Uncertainty of deviation and drift Uncertainty of a single diverter operation The uncertainty, i.e. the reproducibility, of a single switching operation of the diverter is not directly accessible. In literature two approaches are presented as how to find a number for this uncertainty contributor. In [2] a statistical approach is presented. The authors extrapolate the uncertainty of a single switching operation from data taken with different numbers of switching operations (n>5) to the single switching operation. In [3] a model based approach is proposed. Here great experimental effort is undertaken to gain additional information about the water jet, which is cut by the diverter, and about the reproducibility of the diverter movement. From a practical, i.e. production point of view, the latter is not affordable. In this paper an alternative statistical approach is presented. A data set used to determine the average diverter error also inherently contains information about the uncertainty of one switching operation. The heart of this approach is a flow meter of very good reproducibility and stability of its calibration factor. Such flow meters are readily available in the calibration rig due to the presence of the master Coriolis meters. It is reasonable to assume that the pulse value k [kg/pulse] of the master meter is constant during the experiment described in 4.5. The uncertainty of the diverter switching operation can be expressed in an average value and a varying parameter of expected value zero as discussed in [4].

Fig. 5 Periodical check vs. periodical adjustment Beside these comparison measurements a lot of other measurements are necessary to provide knowledge about the systems, which can be used to correct the measurement results for systematic influences. A good example of this is the buoyancy correction of the scale. Also, other process conditions of a calibration must be measured such as the line pressure or the actual through flow. This kind of knowledge is highlighted in light red in Fig.5. It does not directly link into the traceability chain. However, it does contribute to the overall uncertainty via the uncertainty of a correction. The model equations and the uncertainty budget were setup and calculated using the software GUM Workbench 2.3. The equations were structured according to the above described traceability chain. The uncertainty of a DUT calibration depends on two rig operation parameters: Measurement time and through flow. Therefore, the uncertainty budget was calculated in several points of the actual rig operation map.

m div = m div + m div

5.2 Uncertainty Contributors In this section some uncertainty contributors are discussed in greater detail Correction of the average diverter error Correcting the average diverter error in the software has its own uncertainty. The uncertainty can be separated in two parts: the uncertainty of the correction curve (non-linearity) and the change of

(1)

The average value and the varying value can be expressed in terms of the pulse values k1, kn, of the sum of pulses I, of the number of multiple switch operations n and of the corresponding variances (see equation (2) and (3)). In order to get a statistical meaningful data set both experiments, single switch and multiple switch operation, must be performed several times.

Fluid flow measurement symposium

Anchorage, 2009

mdiv =

(k n k1 ) I
n 1 ( 2 (k n ) 2 (k1 ))

1.4

(2) (3)

1.2 1 0.8

n 1 The assumptions here are: a) The sum of pulses of the single switching case I1 and of the multiple switching case In are the same. b) The varying parameters of k1 and kn are independent. c) The varying parameter of the diverter has a constant variance. Equation (3) shows, that the variance of the multiple switching case must be larger than the single switching case to get a meaningful value. Knowing the variance of the single switching experiments the statistically significantly differing variance of the multiple switching cases can be determined using an F-test. Assuming 15 measurements of single switching operation and 15 measurements of 15 additional switching operations lead to an F-factor of 2.5, retaining a confidence level of 95% with a degree of freedom of 14.

signal [g]

(mdiv ) = I

0.6 0.4 0.2 0 0 -0.2 -0.4 20 40 60 80 100 120 140

time [s]

Fig. 6 The resolution of the 400kg scale reading: load 10s, unload 20s with a 1g weight Buoyancy effects The scales reading is corrected for the air buoyancy effect. This is done with the ambient conditions present during the measurement time. Also the ambient conditions are stored which were present when the scale was adjusted. For this correction also the water density is needed.

2 (k n ) > F (14,14) = 2.5 2 ( k1 )

(4)

The uncertainty of a single switching operation is taken into account within the uncertainty budget by equation (3) if 2 (k n ) > 2.5 2 (k1 ) or equation (5) if

2 (k n ) 2.5 2 (k1 ) :

1 .5 2 ( k 1 ) (5) n 1 This uncertainty contributor is flow depending. Therefore it is determined in several through flows. Worst case uncertainties of one switching operation are 3.5g for the 400kg scale and 50g for the 4t scale.

(m div ) = I

Resolution of the DUT count reading The operation conditions of the rig and the measurement chain is modelled in the uncertainty budget. This also includes the sum of pulses which are emitted from the DUT during the measurement time. The worst case here is a minimum sum of 60000 pulses with a maximum relative uncertainty of 0.0016%.

Scales Concerning the reproducibility and the non-linearity of the scales values are given in the section 4.2 and 4.3. The resolution of the scale depends on the amplifier. The DMP40 offers 2x106 steps across the measurement range of a load cell. Depending on the filter settings this may be reduced to lower values. The filter setting is a trade off between the stabilisation time of the signal after filling the water into the vessel and the resolution. The resolution was experimentally verified to be 10g respectively 1g. The results of the 400kg scale are shown in Fig. 6.

Inter space pipe The inter space pipe connects the measurement section with the reference, i.e. the scale. Mass defects may occur if the pipe shell changes its volume during measurement time. Then the pipe acts as a source or a sink depending on the temperature change. This effect is estimated using the thermal expansion coefficient of the pipe material. It amounts to 0.0015% if the temperature changes by 0.3C during a measurement time. Experiments were performed which show, that the change of temperature is usually much smaller than 0.3C. The large tank from which the water is taken ensures stable temperature conditions. 5.3 Rig uncertainty budget In Fig. 7 the expanded uncertainty of mass flow measurement of the calibration rig is depicted as a function of flow. The diagram covers the flow range

Fluid flow measurement symposium

Anchorage, 2009

of both scales. The maximum uncertainty of both scales appears at their low flow end, i.e. 1kg/s and 11kg/s respectively. This is where the compromise between short measurement time and low uncertainty gets visible. By choosing the next higher filling point of the scale these values could even be decreased. At high flows the uncertainty drops to values of 0.012% and lower. The degree of filling of the scale is high and absolute uncertainties, such like the diverter correction, do contribute relatively less. This diagram shows that an overall level of uncertainty of 0.015% is realistic and may be maintained with increased but affordable effort. The latter point will be discussed in section 6.
0.02

6.

MAINTAINING THE UNCERTAINTY LEVEL

0.015

0.01

400kg scale
0.005

4t scale

Apart from the organisation of an accredited flow calibration laboratory including trained staff, periodical checks and the proof of traceability, additional measures were developed in order to maintain the uncertainty level described in the above sections. Depending on the drift of a system component the sampling period must be short enough to keep its uncertainty contribution within the limits. For that reason history files of all relevant components were set up. The continuously sampled data are stored here which makes this data accessible for statistical evaluations. Also programmed limits help the quality engineer to decide whether the component is still running within the uncertainty level or not, whether a component should be replaced or not. The full cycle of periodical adjustment is semi automated. While the data acquisition is performed fully automated during the night shift, the adjustment of components, such as the scale or the diverter, need to be done manually by an experienced quality engineer. This ensures that a plausibility check based on experienced values is performed. 7. SUMMARY

Expanded uncertainty %

0 0.1 1 10 100

flow [kg/s]

Fig. 7 Expanded uncertainty (k=2) 5.4 Uncertainty statement of calibration On a calibration certificate of an accredited flow laboratory the over all uncertainty of measurement must be given according to GUM. This comprises the rig uncertainty and the reproducibility of the DUT. Both are combined to
2 2 MU = MU Rig + reprDUT .

(6)

Here it is important to distinguish between a single measurement and a set of multiple measurements. A customer will most likely see a single measurement point on his certificate. Therefore, the reproducibility of the DUT must be fully taken into account. In the case of a Promass 83 F the uncertainty of measurement stated on the accredited calibration certificate becomes 0.025% for a single point measurement. The value of the reproducibility of the Promass 83/84 F is a typical value for this device family obtained by measuring several devices out of the running production. If a set of n measurements under the same conditions are taken, the reproducibility of the DUT is calculated according ISO 11631 [5].

This paper presented the Endress + Hauser Flowtec approach of minimizing the measurement uncertainty of a calibration rig under production conditions. A whole basket of measures were proposed and realized including: - filling in scales adjustment points - diverter error correction - optimized scale design - stainless steel calibration weights - automated weight movement The uncertainty budget includes around 30 contributors. A special statistical approach was developed in order to quantify the uncertainty of a single switching operation of the diverter. The efforts succeeded in the accreditation of the expanded uncertainty of 0.015% for mass flow. First experience shows that production calibration time and time spent for maintenance roughly doubled with respect to the standard calibration procedure. ACKNOWLEDGEMENTS The authors are grateful for the permission of publishing this paper. Special thanks go to R. Kessel, who contributed to the statistical approach of determining the uncertainty of one diverter switch.

Fluid flow measurement symposium

Anchorage, 2009

REFERENCES [1] EN 24185: Durchflussmessung von Flssigkeiten in geschlossenen Leitungen Wgeverfahren Mathies, N., Lederer, T., Technisches Messen 74, p333-341, 2007: Neue Methode zur Bestimmung des Zeitfehlers fr Umschalteinrichtungen von gravimetrischen Messanlagen Engel, R., Baade, H.-J., FLOMEKO 18.-21. September 2007, Johannesburg, South Africa: Model-based fluid diverter analysis for improved uncertainty determination in liquid flow calibration facilities, exemplified with PTBs Hydrodynamic Test Field Kessel, R., 2007: unpublished technical report International Standard ISO 11631: Measurement of fluid flow-Methods of specifying flowmeter performance

[2]

[3]

[4] [5]