Anda di halaman 1dari 10

CALIBRATION OF NONLINEAR LAMINAR FLOW ELEMENTS FOR MULTI-GAS APPLICATIONS

Chiun Wang, Ph.D. Cardinal Health* 22745 Savi Ranch Parkway Yorba Linda, CA 92887 (714)919-3227 ccwang_2000@yahoo.com
Abstract - Laminar flow elements are frequently used in flow measurement devices either as flow splitters or as differential pressure generators. To a certain extent all laminar flow elements suffer from the entrance effect, rendering the pressure versus flow relationship nonlinear. The non-linearity is sensitive to even minor differences in the geometry of the entrance to the tubes or the channels of the laminar flow element, which is highly susceptible to the manufacturing tolerances. Attempts to model the nonlinear laminar-flow elements mathematically for high accuracy applications have seen only limited success. This paper explores similarity theory as the method to extract the multi-gas scaling model from the flow meter calibration data and to establish the pressure-flow relationship by using dimensionless quantities. Without resorting to any explicit functional forms, the flow meters calibrated by this method can accurately predict the flow of any process gas after they are calibrated by using a single inert gas.

INTRODUCTION
Laminar flow elements are frequently used in flow meters either as the flow sensing or as the flow-splitting device. Practical laminar flow elements are usually constructed from honeycombs with either tubular or channel-like structures. When used as a flow-sensing device, the laminar flow element generates the pressure drop to infer the flow rate, and the flow versus pressure drop characteristics is responsible for the accuracy of the flow meter. When used as a flow splitting device, the laminar flow element divides the flow between the sensor and the bypass flow paths, and the pressure drop characteristics is responsible for accurately establishing the flow-splitting ratio. Fully Developed Laminar Flow The pressure drop in fully developed laminar flow in tubes, honeycomb or two-dimensional channels depends linearly on the flow rate and on the streamwise distance. For examples, the pressure drop in round tubes is governed by Hagen-Poiseuilles law:

P =

128 QL D4 12 QL H3

... (1)

; and, the pressure drop in two-dimensional channels is governed by Couettes law:

P =

... (2)

, where is the fluid viscosity, L is the streamwise distance, D is the tube diameter, H is the channel height, and Q is either the volumetric flow-rate for round tubes or the flow-rate per unit width for two-dimensional channels. At first sight, these simple relationships may suggest to a flow meter designer that tubular or channel-like structures offer a perfectly linear relationship between the pressure drop and the flow rate. Unfortunately no

laminar flow elements can deliver purely fully developed flow. In fact any laminar flow element must have an entrance, where the boundary layers grow and the pressure drop is a nonlinear function of the flow rate. Of course, if the laminar flow elements are sufficiently long the contribution from the entrance effect may be relatively small and may even be negligible in comparison to the overall pressure drop. Due to space limitations, however, most practical laminar flow elements are relatively short. As a result the pressure-flow relationship for these devices is often quite nonlinear, and for high accuracy applications the nonlinearity can strongly affect the flow meters performance and calibration. Entrance Effect The nonlinear pressure drop at the entrance of a laminar flow element is associated with the loss of fluid momentum occurring when the flow velocity evolves from flat into the parabolic profile. The nonlinearity is a function of not only the flow but also the detailed shape of the entrance. The development of the boundary layer near the tube entrance, with or without separation, is quite similar to what happens near the VenaContracta of an orifice. The extensive data compiled by ASME [1] for orifice flow meters suffice to illustrate just how sensitive the orifice discharge coefficient is to the geometry at the inlet. In the same manner the entrance geometry significantly affects the nonlinearity of the pressure drop in a laminar flow element. The exact shape at the inlet of the laminar flow element is determined by the manufacturing tolerance. Consider laminar flow elements composed of bundles of round tubes for example. For tubes with 0.020 ID and 0.005 thick wall, the corner radius may be specified as R 0.001 by design. Tubes with actual corner radius less than 0.0001 or equal to 0.001 will both be accepted by the quality inspection process. The corner to radius ratio, however, will vary from less than 1 percent to up to 10 percent, and the corresponding flow characteristics will be quite different. In practice, as illustrated in Fig. 1, unless the design spec is extremely tight, the actual corner to tube radius ratio can vary substantially, and the nonlinearity of the laminar flow elements can be considerably different from unit to unit. Here we assume the corner of the inlet having the shape of a circular arc. In reality the shape of the corner is never an exact circular arc but an undefined curve, for which additional geometrical parameters must be introduced before the characteristics of the flow may be accurately defined. Other Effects Apart from the entrance geometry effect, other upstream disturbances can also affect the laminar flow elements pressure drop characteristics. The sensitivity of flow meter calibration to the geometrical details in the upstream housing, such as elbows, screens or protuberances is well known. The laminar flow elements wall (or web) thickness is also important because it determines the blockage ratio and hence the pressure gradient immediately upstream of the entrance. Due to all the reasons given above, calibrating and modeling the nonlinear laminar flow elements for highaccuracy applications is not as simple a task as it may sound. What is more, for certain industrial applications such as the semiconductor wafer fabrication processes, the flow meter manufacturer is required to guarantee the accuracy of the flow meter for the process gases after it is calibrated only with a single inert calibration gas. For these multi-gas applications an accurate and efficient method is desperately needed for calibrating and scaling the laminar flow elements and the flow meters.

FLOW IN TUBES OF FINITE LENGTH PREVIOUS WORK


Langharrs Theory The laminar flow in round tubes with entrance effect has been studied extensively by various investigators using methods ranging from analytical to experimental or even CFD numerical techniques. Yet the case is not closed. For example see Durst et al. [2] who recently revisited the entry-length for the laminar flow in a round tube. For the purpose of scaling and physical similarity, Langharr [3] conducted a theoretical analysis which related the pressure drop to the downstream distance x in an expression of the form:

p u 2

x 1 f . D Re D

(3)

, where ReD = uD/ is the flow Reynolds number. Langharrs theory addressed only the laminar flow within the tube. No consideration was ever given to either the effect of the corner radius at the inlet or the effect of any other upstream disturbances. As we discussed earlier, these factors can all significantly impact the flow characteristics. Consequently it is doubtful whether one can use Langharrs theoretical solution, or any other closed-form solutions alike, to accurately model the entrance flow in laminar flow elements which are often associated with imprecise inlet geometry, bearing in mind that under most circumstances the precise geometry cannot even be defined. Kreith and Eisenstadt Kreith and Eisenstadt [4] compared the experimental data from various sources against Langharrs theory. As shown by Kreith and Eisenstadt, the data from different authors compare very well at small Langharrs number ReDD/L, i.e. when L/D is large and the contribution from the entrance effect is relatively small. On the contrary when the entrance effect is pronounced, i.e., when L/D is small and Langharrs number is large, the data deviate quite significantly. One interesting observation through Kreith and Eisenstadts work is, although there is significant scatter among the data from different experimenters, there is significantly less scatter among the data from the same experimenter. This is true even when Langharrs number is large and the entrance effect is severe. A tentative explanation is that an individual experimenter very likely would have used exactly the same experimental setup involving a specific piece of tube with exactly the same inlet shape, corner radius, aspect ratio L/D, and even surface roughness, etc. Obviously the data from exactly the same device could never reveal the statistical variation among multiple, different devices. Quest for a Universal Characteristic Curve Several investigators attempted to establish a universal characteristics curve for the laminar flow in tubes with the entrance effect but received only limited success, for example see Todd [5]. Some would fit a curve through the existing data similar to those assembled by Kreith and Eisenstadt, and then use the fitted curve to characterize the devices in hand. Others would conduct extensive experiments using samples for the devices of interest, correlate their data according to Langharrs theory, and then fit a curve to establish the desired correlation. Whichever path taken, the strategy was always to count on the data among different devices correlating so well that one can accurately represent them with a universal function. Unfortunately none of these endeavors could achieve the tight tolerance required for high-accuracy multi-gas flow meters, shedding doubt on the validity of the similarity principle. Prompted by the earlier observation through Kreith and Eisenstaedts work that there was little scattering of data from the same experimenter, we speculate that device-to-device variation due to manufacturing tolerance might be the culprit for the data scattering; and, device-to-device variation may forbid any accurate universal model even for devices of the same design. If this conjecture turns out to be true, our hope is that the similarity relation proposed by Langharr should at least hold, and could be further explored for gas-to-gas scaling, provided one could circumvent the device-to-device variability issue. One possible way to circumvent the device-to-device variability issue is to treat each flow meter as a unique entity during its calibration and scaling. Rather than attempting to develop a universal model, in the present study the method of similarity is exploited for the calibration and scaling of individual multi-gas flow meters.

SIMILARITY RELATIONSHIP
Let us re-examine the dimensionless parameters governing the problem of laminar flow in a circular tube of finite length. Assume the inlet conditions such as the test setup, the filter and screen arrangement, and the entrance corner radius of curvature, etc. are all defined exactly and do not change. The physical quantities

governing the laminar flow problem are: the tube length L, the inside diameter D, the fluid viscosity , the fluid density , the mean flow speed u, and the overall pressure drop P. With these, Buckinghams - theorem suggests that there should be three dimensionless groups governing the problem. These may be chosen as 2 D/L, ReDD/L, and q/P. Here q stands for the dynamic head u ; and, instead of Reynolds number ReD, the quantity ReDD/L is adopted to comply with Langharrs theory. It is interesting to compare these dimensionless parameters against those employed by Wright et al. [6], 2 2 where the three quantities chosen are: the aspect ratio L/D, the viscosity coefficient L P/ , and the flow 3 coefficient L P/Q. One may readily show that both the viscosity and the flow coefficient may be obtained from combining the quantities ReD, L/D, and P/q given in the last paragraph. Indeed the viscosity coefficient 2 2 3 is equivalent to ReD (P/q)(L/D) ; and, the flow coefficient is equivalent to ReD(P/q)(L/D) . In addition, both the present analysis and [6] suggest that there should be a third dimensionless parameter L/D governing the problem. For a specific laminar flow element, however, the aspect ratio L/D is fixed and may therefore be omitted from the remaining discussions. One thing we did differently from Kreith and Eisenstadt was, instead of using P/q and 1/(ReDD/L) as the dependent and the independent variables, we used the inverse of these two quantities, i.e., q/P and ReDD/L to characterize the laminar flow elements. The similarity relation we seek was therefore of the following form:

u 2
P

D f Re D . L

... (4)

Although q/P and ReDD/L are mathematically the same as P/q and 1/(ReDD/L), respectively, the present choice of variables does offer some distinctive advantages for data analysis. Particularly in the limits of very low flows or infinitely long tubes where the entrance effect is negligible, when plotted in the traditional fashion Hagen-Poiseuilles law in Eq.(1) reads P/q = 1/(64 ReD/L). Both the left- and the right-hand side of this expression go to infinity as the flow rate approaches zero. Thus any small error in the low flow data will be greatly exaggerated, and the quality of the model may be compromised. When the variables q/P and ReDD/L are used instead, Hagen-Poiseuilles law reads q/P = 64 ReDD/L. Since both q/P and ReDD/L approach zero at very-low flow rates, the present choice of variables assures that the correlation curve should go exactly through the origin. This enabled us to easily sanitize any zero-offset in the measuring instruments and to effectively anchor the calibration curve to the origin. Validation of the Similarity Relationship Three laminar flow elements were investigated during the present study. LFE-A covered up to 6 slm of N2 flow and consisted of 129 round tubes 1.50 long with 0.020 ID and 0.004 thick wall. LFE-B and LFE-C both consisted of 1.50 long 0.030 ID round tubes with 0.004 thick wall. LFE-B had 67 tubes and covered up to 15 slm of N2 flow; whereas LFC-C had 138 tubes and covered up to 30 slm of N2 flow. (Nitrogen flow rate is used here as a means to compare the range of a multi-gas flow meter.) The tubes were all electro-chemically polished. Typical tolerances were 0.0005 for the wall thickness, and 0.001 for the ID, respectively. The corner radius at the tube inlet was not controlled. The as-received corner radius was typically between 10% and 30% of the wall thickness. The tubes in the laminar flow elements were packed in hexagonal patterns and installed inside the hexagonal housings. To make the tube count exact, the vacant spaces surrounding the tube bundles were filled with round solid rods with the same outer diameter as the tubes. Installed a fraction of an inch upstream of each laminar flow element was an 800-mesh Dutch-weave screen to help redistribute the flow coming from the -inch inlet fittings. Eight samples of each laminar flow element design were individually tested using nine different gases including N2, He, Ar, H2, CH4, CO2, CF4, CHF3, and SF6. During these experiments the laminar flow elements were installed in parallel to a thermal mass flow sensor in a mass flow controller arrangement. The sensors were pre-calibrated with N2 and then scaled to run other gases by using the sensor similarity model described in [7] and [8]. The maximum flow rate allowed through the sensor was 10 sccm for N2.

The total mass flow rate through the mass flow controller was measured with a Bios International moving piston calibrator DryCal ML-500, with a quoted accuracy of 0.5% reading, covering the flow rate from 5 sccm to 50 slm. The pressure drop was not directly measured but was inferred from the mass flow rate indicated by the thermal mass flow sensor. With the 1.5-long and 0.014-ID thermal mass flow sensors, the maximum Langharrs number ReDD/L in the sensor was below 0.35 for N2 and 0.60 for SF6 and argon. The entrance effect on the sensor flow was therefore neglected [2] and the total pressure drop calculated according to Hagen-Peusuilles law. Gas viscosity data available from scientific databooks were used for calculating the pressure drop. The gas viscosity in the sensor was evaluated at a mean sensor temperature of 50 C. The viscosity in the laminar flow elements was evaluated at the room temperature. The nominal tube length and tube diameter were used in calculating the mean flow speed and the mean Reynolds number. It was estimated that 1~2 percent of the total flow passed through the crevices in between the tubes or the rods. This flow was neglected during the modeling process on the ground that the fraction of flow through the crevices did not change significantly from gas to gas, and hence should not significantly affect the desired scaling relationship. Figs. 2, 3, and 4 show the 9-gas test data from one member of each of the laminar flow element LFE-A, B, and C, respectively. Although only the data from one member of each LFE family were shown here, data from the remaining members of the LFE all looked very similar. In particular there was very little random scattering of data at either low- or high-flow for any of the gases studied, suggesting that the similarity principle holds quite well. Closer examinations did reveal, however, that minor deviations still remained among the data between different gases for any one laminar flow element. Taking Fig. 4 (corresponding to LFE-C) as an example, the curves for different gases do not lie exactly on top of each other. There is a persistent 2~3% offset for some of the gases studied, with the worst offset (4%) belonging to CHF3. We do not know exactly what caused these offsets. Possible explanations include: (1) the temperature at which the gas viscosity for the sensor was calculated maybe incorrect; and, (2) the gas viscosity data or its temperature coefficient might be inaccurate. After cross-comparing the data from all the devices, we were nevertheless convinced that the gas specific offsets are at least reproducible among devices of the same design. In particular, if one calculates the ratio of q/P (as a function of ReDD/L) for one process gas to that for N2, the ratio remained very much the same for all devices of the same design. A typical example for the ratio calculated for LEF-C is as shown in Fig.5. As shown in Fig. 5, the calculated ratios for all 9 gases lied within 0.94-1.04 throughout the entire range of ReDD/L of interest. They therefore represented a mild and consistent correction to the similarity relationship. Gas specific bias functions based on the mean of the calculated ratios were accordingly introduced for each of the gases tested. Using these gas specific bias functions as corrections to the dimensionless calibration table, an accuracy of better than 1.5% (3) reading was achieved for the gas-togas scaling relationship for all 9 gases studied, for all set-points between 5% and 100% of full-scale, and for all three laminar flow element designs. Examples for the residual error corresponding to LEF_C after the model was compensated by using the bias functions are as shown in Fig. 6. In order to determine the bias function for a certain gas, ideally the experimental data for the specific gas must be available. When no gas-specific test data other than the gas viscosity was available, however, our experience suggests that a default bias function of unity could be used and should provide a calibration accuracy of better than 5% of reading. Considering the wide range of gas properties covered under this study, the present results are very encouraging. Together with the bias functions the similarity principle worked satisfactorily when the 9-gas data from any individual laminar flow element are mutually compared. When the data from different laminar flow elements (although of the same design) are cross-compared, however, the data spread exceeds the tolerance specification quoted above, even after the bias functions are included. This seems to echo our speculation that manufacturing tolerance may have impaired the similarity relationship. For the simple purpose of ranging a flow meter or sizing a laminar flow element, however, the data from different devices can still be averaged to provide a satisfactory mean characteristic curve, regardless of the data spread. Using this mean characteristic curve with similarity scaling, our experience suggests that flow meters may be sized to within 5% of the desired process-gas flow rate without conducting the actual calibration.

FLOW METER CALIBRATION


To apply the method of similarity to the calibration of multi-gas flow meters, each flow meter or laminar flow element was calibrated and scaled as a unique device such that any distinctive features in the flow meter characteristics are deliberately preserved. To initiate the calibration process, the flow meter was first tested at a list of selected flow rates by using the calibration gas. The total flow through the laminar flow element together with the corresponding pressure drop, inferred from the sensor flow rate as in the present case, were measured and recorded. The test was repeated for as many times as necessary until the averaged data meet the tolerance spec, according to the central-limit theorem for example. The mean dimensionless quantities q/P and ReDD/L at each calibration data point were then calculated, recorded, tabulated and permanently stored for the specific flow meter. The above calibration table essentially contained all the information required for scaling the laminar flow element for any process gas. Gas-to-Gas Scaling To construct the calibration curve for a process gas, one began by multiplying the dimensionless calibration table with the appropriate bias function for the process gas of interest. The compensated calibration table was then scaled back into the physical quantities by using the physical properties of the process gas. [9] It was rather easy to calculate the pressure drop corresponding to a given flow rate from the compensated dimensionless calibration table. For a given mean flow speed u, with the gas viscosity and gas density given, 2 the quantities ReDD/L and q = u were first calculated. The quantity q/P corresponding to the specific ReDD/L was then determined from the corrected calibration table by interpolation. Together with the dynamic head q this allows us to determine the pressure drop P. A tension-spline routine was used extensively throughout this study for carrying out the interpolation. The tension-spline routine produced a piecewise smooth curve going exactly through each of the mean calibration data points. This guaranteed that the fitted curves matched the mean calibration data, which in turn met the tolerance spec according to the central-limit-theorem as we mentioned earlier, at least at the calibration data points themselves. Common least-square regression type of curve-fitting using polynomials, either in the linear or in the log-log scale, can easily introduce systematic error exceeding the uncertainty of the original data. The use of interpolating tension-splines avoided this problem. Some iterative calculations were required for calculating the average flow rate corresponding to a given pressure drop. This was because the flow speed u appears simultaneously in the Reynolds number ReDD/L and the quantity q/P, making the functional relationship implicit. The iteration was, however, uncomplicated and easily programmable into any lab computers. To give an example for the iterative process, let us assume a certain P exists across the laminar flow element. To begin the iteration, choose a certain mean velocity u (and hence ReDD/L for given gas viscosity and gas density) as the first guess. Corresponding to the ReDD/L, one may determine the quantity q/P by interpolation from the compensated dimensionless calibration table using the method described in the last paragraph. This determines the first-guess P. Since in general the first-guess P would not match the specified P, one will then choose a new gas flow rate and repeat the above calculations in a trial-and-error process. The above iteration was carried out by using the routine Newton-Ralphsons method and terminated automatically when the calculated P matches the given P to within a tight tolerance. The entire calibration curve for the process gas was established by repeating the above process to cover the entire range of flow rate of interest.

RESULTS AND DISCUSSIONS


The principle of similarity is a powerful tool for developing compact scaling relationships and physical models for nonlinear phenomena. In this paper an accurate method was developed for calibrating laminar flow elements and flow meters for multi-gas applications. By casting the calibration data in similarity form, and by re-scaling the dimensionless calibration table by using the gas physical properties, the multi-gas flow

meters calibrated with the present method can accurately measure the flow of any process gas after it is calibrated only with a single inert gas. With the use of gas-specific bias functions, the multi-gas flow meters calibrated with the present method achieved an accuracy of better than 1.5% of reading for all 9 gases investigated over the entire flow meter operating range of 5% to 100% of full-scale. When no bias functions are available for a certain process gas, the present method suggests the use of the unity bias function as default, and promises an accuracy of 5% of reading depending on the accuracy of the gas viscosity data available. The flow meters calibrated with this method covered full-scale flow rates up to 50 slm for N2, 40 slm for argon, and 8 slm for SF6, corresponding to a maximum Langharr's number of ReDD/L ~ 15. The mass flow controller is an interesting device in that its laminar flow element follows one similarity law, as we discussed in this paper; its thermal mass flow sensor follows a different similarity law [7] [8]; and, its flow control valve follows yet another scaling law - the compressible, orifice flow relation [6]. In order to properly deal with these totally different nonlinear components, the models for the valve, the thermal mass flow sensor, and the laminar flow element must be broken down and treated separately. The performance of the entire multi-gas mass flow controller may then be predicted after re-combining these models according to physics. Proper understanding of the background physical process is critical for developing the appropriate similarity laws. Proper understanding and appreciation of the similarity laws is critical for the successful implementation of the multi-gas mass flow controllers.

References
[1] American Society of Mechanical Engineers, Flow Meters, Their Theory and Applications 6 ed. 1971. [2] Durst, F. Ray, S. Unsal, B. and Bayoumi, O.A. The Development Lengths of Laminar Pipe and Channel Flows, Trans. ASME, Vol127, 1154-1160, Nov. 2005. [3] Langharr, H.L. Steady Flow in the Transition Length of a Straight Tube, J. Applied Mechanics, Vol. 9, No. 2, A55-58, 1942. [4] Kreith, F. and Einsenstadt, R. Pressure Drop and Flow Characteristics of Short Capillary Tubes at Low Reynolds Numbers, ASME Paper No. 56-SA-15, 1956; also Trans. ASME, pp1070-1074, 1957. [5] Todd, D. A. A Universal Calibration Curve for Laminar Flow meters, Proc. 2 on Flow, Instrument Society of America, St. Louis, MO. Mar 1981.
nd th

International Symposium

[6] Wright, J.D., Kayl, J.P., Johnson, A.N., and Kline, G.M. Gas Flowmeter Calibration with the Working Gas Flow Standard, NIST SP250-80, Jan. 2008. [7] Wang, C. A Similarity Theory for Thermal Mass Flow Sensor and Its Gas Conversion Factors, Measurement Science Conference, Jan. 25, 2007, Long Beach, CA. [8] Wang, C. Thermal Mass Flow Sensor Similarity Theory Comparison with Experiment, Measurement Science Conference, Mar. 13, 2008, Anaheim, CA. [9] Wang, C., Lull, J., and Valentine, B. Multi-gas Flow Devices, United States Patent and Trademark Office Patent Publication No. 20080059084, Mar. 6, 2008. *This paper is based on the authors prior work at Celerity - Unit Instruments.

(Published in: Measurement Science Conference, Anaheim, CA. March 2009)

Fig. 1 Effect of inlet geometry on the development of the laminar boundary layer near the entrance of a tube.

Fig. 2 Laminar Flow Element LFE-A


0.08 H2 He 0.06 CH4 N2 CO2 0.04 CF4 CHF3 Ar 0.02 SF6

q / P

0.00 0 1 2 3 4 5

Re DD/L

Fig. 3 Laminar Flow Element LFE-B


0.20 H2 He 0.15 CH4 N2 CO2

q / P

0.10

CF4 CHF3 Ar SF6

0.05

0.00 0 4 8 12 16

Re DD/L

Fig. 4 Laminar Flow Element LFE-C


0.20 H2 He 0.15 CH4 N2 CO2

q / P

CF4 0.10 CHF3 Ar SF6 0.05

0.00 0 4 8 12 16

Re DD/L

Fig. 5 Bias Functions for LFE-C


1.10 He Ar 1.05 H2

Bias Function

N2 1.00 CO2 CH4 CHF3 0.95 CF4 SF6 0.90 0 4 8 Re DD/L 12 16

Fig 6 Residual Error After Compensation


5.0% 4.0% 3.0% 2.0% 1.0% 0.0% -1.0% -2.0% -3.0% -4.0% -5.0%

He Ar H2 N2 CO2 0.0 2.0 4.0 6.0 8.0 10.0 12.0 14.0 16.0 CH4 CHF3 CF4 SF6

ReD D/L

Anda mungkin juga menyukai