Safety

Select the Best Model for Two-Phase Relief Sizing

A variety of methods exist for sizing valves, but not all give the best predictions for certain conditions.
As shown in Figure 1, there are obvious similarities between an ideal nozzle and an actual relief valve, but there are also signicant differences. Proper sizing of a safety relief valve requires knowledge of the conditions upstream and downstream of the device, the physical and thermal properties of the uid(s) at these conditions, and a model that accurately predicts the mass ux through the valve Gv as a function of the uid properties and ow conditions. Methods for valve sizing for single-phase (gas or liquid) ows are well established, and are based upon simple singlephase isentropic ow models, together with discharge coefficients that are measured by the valve manufacturer and reported in the Red Book (1).

he proper nozzle area for a safety relief valve is determined by the required relief massow rate m, which is determined by the specic overpressure scenario and the mass ux capacity of the valve at these conditions: A= m (1) Gv The valve mass ux Gv is the calculated value for an ideal (isentropic) nozzle Gn from an appropriate model, multiplied by a discharge coefficient Kd, which accounts for any discrepancies between the (ideal) nozzle model and ow in a real nozzle, as well as any differences between the ow in the nozzle and the actual valve: Gv = Kd Gn (2)

Spindle Eductor Tube Guide Disc Holder Disc Adjusting Ring Adjusting Ring Pin Primary Orifice Nozzle Base (a) Actual safety-relief valve Inlet Neck Threads Second Annular Orifice

Figure 1. An actual safety relief valve is quite different from an ideal nozzle.
Pc Pn

Pb

Po

(b) Ideal nozzle

56

www.aiche.org/cep/

May 2001

CEP

However, two-phase ow is frequently encountered in various relief scenarios and there are no data or Red Book coefficients, or even an accepted and veried two-phase ow model that may be used to size valves for such conditions. One reason for this is that two-phase ow is considerably more complex than single-phase, since there is a large number of variables associated with the uid properties, distribution of the uid phases, interaction and transformation of the phases, etc. Consequently, there is a variety of models, each of which is based on a specic set of assumptions that may be valid for certain specic conditions, but may not be accurate for others. For example, the models that best describe the ow behavior of frozen gas-liquid ows (i.e., those having no phase change) may be different from those for ashing ows, and the best model may be different for nonequilibrium ow in short nozzles than that for equilibrium ow in long nozzles.

Nomenclature
= area, ft2 = specic heat of liquid at stagnation conditions, ft-lbf / lbmF or Nm/kgC e = pipe roughness, dimensionless G = mass ux, lbm/ft2s or kg/m2s G* = dimensionless mass ux, Gc/(Poo)1 hLGo = heat of vaporization at stagnation conditions, ft-lbf/lbm or Nm/kg Kd = discharge coefficient, dimensionless Kf = friction loss coefficient, dimensionless kNE = empirical nonequilibrium parameter in Eq. 8, dimensionless ks = empirical slip coefficient in Eq. 7, dimensionless L = nozzle length, ft or m NNE = nonequilibrium parameter dened by Eqs. 11 and 12, dimensionless P = pressure, lbf/ft2 or Pa S = ratio of gas phase velocity to liquid phase velocity, dimensionless T = temperature, R or K V = velocity, ft/s or m/s x = quality mass fraction of gas or vapor in mixture, dimensionless y = mole fraction in vapor Z = vertical elevation, ft or m A Cpo Greek letters o = volume fraction of gas phase at stagnation conditions = specic volume, ft3/lbm or m3/kg = specic volume of gas minus specic volume of liquid at LGo saturation conditions, ft3/lbm or m3/kg o = specic volume of liquid at stagnation conditions, ft3/lbm or m3/kg = density, lbm/ft3 or kg/m3 = parameter dened by Eq. 9, dimensionless Subscripts G = gas or vapor phase L = liquid phase o = stagnation state s = saturated state

Undersizing or oversizing An inaccurate model for ow through a valve/nozzle could result in either undersizing or oversizing the valve. The consequences of undersizing are obvious the valve will have insufficient capacity to prevent overpressuring the vessel. However, the consequences of oversizing can also be severe, although maybe not as obvious. An oversized valve results in a higher than expected ow rate through the relief system, which also yields a greater than expected pressure drop both upstream and downstream of the valve. Either or both of these extreme pressure drops can negatively affect the stable operation of the valve, with possible damage or destruction to it. Unexpectedly high ow due to oversizing also results in undersized discharge piping and the effluent handling system downstream of the valve, also with negative consequences. Many models, methods, and procedures have been proposed for analyzing two-phase ows in relief systems (26). However, none of these models has been systematically veried by comparison with extensive reliable experimental data. Here, we will review the predictions of what appear to be the most useful and appropriate models for two-phase ow in valves, nozzles, and tubes, and discuss the conditions under which each gives reliable results by comparison with the best data that have been found for two-phase ashing ows, in terms of the entering uid quality and geometry of the system. The models The two-phase ow methods that are currently considered to be the most appropriate for relief valve sizing are based on either the homogeneous equilibrium (HEM) or the homogeneous nonequilibrium (HNE) models. There are various forms of each of these models in the literature, which vary primarily depending on the method used to evaluate the properties of the two-phase mixture. Both models assume that the two-phase mixture is homogeneous, i.e., the two phases are sufficiently well mixed that they can be described as a pseudo-single phase uid with properties that are a weighted average of those of each phase. The equilibrium assumption implies that both phases are in thermodynamic and mechanical equilibrium, that is, any phase change (e.g., ashing) occurs under equilibrium conditions at the saturation pressure, and both phases move at the same velocity (no slip). The Omega method (3) is a special case of the HEM model in which the two-phase density is represented as a linear function of pressure and the thermal/physical properties of the uid at the stagnation state. This permits an analytical solution of the isentropic nozzle equation. A version of this model is being recommended by the American Petroleum Institute (API RP 520, Sizing, Selection, and Installation of Pressure-Relieving Devices in Reneries, under revision). The nonequilibrium models may account for either the possibility of delayed ashing after the uid mixture reaches the saturation pressure or the occurrence of slip between the phases.
57

CEP

May 2001

www.aiche.org/cep/

Safety

It is generally assumed that the HEM model is adequate in most cases for two-phase ow in relatively long nozzles/pipes, for both frozen (constant quality) and ashing ows, when the uid properties are properly evaluated. For frozen ows, however, large errors can result from neglecting slip in short nozzles. Slip occurs as a result of expansion of the gas/vapor phase as the pressure drops, producing a corresponding increase in the gas-phase velocity. Hence, the gas accelerates relative to the liquid, resulting in a velocity difference and a corresponding drag force between the two phases. The result is that the local in situ mixture density and mass ux (i.e., the holdup) in the presence of slip are greater than they would be without it. Slip is expected to be most pronounced when the pressure gradient is large, such as in the entrance region of a pipe or nozzle. This can occur for both frozen or ashing ows, and has been shown to be especially important in frozen air/water mixtures in short nozzles (7). The most difficult situation to reproduce accurately is when the uid entering the valve is either saturated liquid, liquid just above the saturation pressure (slightly subcooled), or a twophase saturated mixture with very low quality, for which considerable ashing occurs within or near the exit of the nozzle. For such ows, nonequilibrium effects can result in a mass ux that is many times as great as predicted by the HEM model (8). This is because ashing is actually a rate process, rather than an equilibrium one. Nucleation of vapor bubbles only occurs when the pressure drops below the equilibrium saturation pressure (corresponding to a nite superheat), and the bubbles then grow at a nite rate controlled by the rate of heat and mass transfer from the liquid to the vapor phase. Even though this process can be fast, the high velocities experienced in a typical relief scenario can result in a signicant distance of uid travel before the ashing is complete. At a velocity of several hundred feet per second, the uid will travel several inches in just one millisecond. The HNE model accounts for this delayed ashing by assuming that vaporization is not complete until the uid has traveled at least 10 cm (4 in.) along the nozzle. This nonequilibrium effect is more pronounced as the relief pressure rises, since the higher the pressure the higher the velocity through the nozzle, and the further the uid will travel before ashing is complete. Both slip and nonequilibrium effects result in a higher mass ux and holdup relative to that predicted by the HEM model, since both of these effects result in a local quality (i.e., mass fraction of gas) within the tube/nozzle that is smaller than would occur if the ow were at equilibrium.

G n = n 2

Pn Po

1/2

dP

(3)

where Gn is the mass ux through the nozzle, Po is the (upstream) stagnation pressure, Pn is the pressure at the nozzle exit, is the local (two-phase) uid density within the nozzle, and n is the uid density at the nozzle exit. The two-phase density is related to the densities of the gas and liquid phases by: = o + (1 - )L (4) where is the volume fraction of the gas phase. This, in turn, is related to the mass fraction (quality) of the gas phase x and the slip ratio S = VG/VL by: x (5) = x + S(1 x) G/ L There are many models for slip (or holdup (1 - )) that relate S to the properties of the gas and liquid phases, (6), although there are no clear guidelines for determining the conditions under which these models might be valid for nozzle ow. The TPHEM program includes an option that includes the effect of slip through the following expression:
Pn

G n2 =

Po

2 x/ G + (1 x)/ L dP

(6)
t

[x/( GS) + (1 x)/ L] 2 [xS 2 + 1 x]

The subscript t represents conditions at the throat (exit) of the nozzle. The model also includes the following formula for the slip ratio in terms of an empirical parameter ks, which is input by the user (while there are no guidelines for choosing the appropriate value of ks, Ref. 8 offers some experimental guidelines): ks S = 1 x + x L (7) G The TPHEM program also includes an option to account for nonequilibrium behavior by replacing the local equilibrium quality x by a nonequilibrium quality xNE, where xNE < x:
2 x NE = x A + k NE x 2 x A

(8)

TPHEM model This model is based on work by Simpson (9, 10) and is described in a recent CCPS Guidelines book (4) that includes a CD containing a program for implementing the model. It is based on a numerical integration of the isentropic nozzle equation:

Here, kNE is an empirical parameter (input by the user), and xA is the quality at the upstream state for saturated inlet conditions (or zero for subcooled inlet conditions). According to Simpson (9), a value of kNE = 1 gives results comparable to the HNE model (see below), but there are no further guidelines for the appropriate value to use for kNE. A critical aspect of any model is the method used to evaluate the density of the two-phase mixture that appears in the integral in Eq. 1. The TPHEM program incorporates a choice of a variety of possible (two- or three-parameter) empirical models for this density (10). The user can select any of these models, and must input data for the liquid- and gas-phase densities and the quality at the upstream stagna-

58

www.aiche.org/cep/

May 2001

CEP

tion pressure at one or two downstream pressures, depending upon whether a two- or three-parameter density model is chosen. The program evaluates the model parameters by tting the pressure/density data provided, and uses the resulting model to determine the mass ux by numerical integration of Eq. 3 from the stagnation pressure to the discharge pressure. If the ow is choked, the mass ux will reach a limiting value before the discharge pressure is reached, at a pressure corresponding to the critical choke pressure, which can be determined by varying the discharge pressure. The program includes options for nozzle ow with or without friction (i.e., an entrance loss coefficient) and pipe ow with friction, as well as for viscous or inviscid (zero-viscosity) liquid properties. The one-parameter density model in TPHEM is equivalent to the Omega method (3), which is an analytical solution of the nozzle mass-ux integral. The model parameter is equivalent to the parameter of the original Omega method, which can also be evaluated from the thermal/physical properties of the two-phase mixture at the stagnation conditions:
= o 1 C poT oPo vLGo 2PovLGo + vo h LGo h LGo
2

1.2 1.0 0.8

G*

0.6 0.4 0.2 0 0.001 Data HNE TPHEM S =1.5

0.01 Quality

0.1

Figure 2. Graham Nozzle No. 1 with air/water data vs. predictions. G1 = GERM represents the equilibrium rate model (ERM), which describes the critical (choked) mass ux resulting from the phase change (ashing). NNE is the nonequilibrium parameter, which represents delayed ashing for lengths less than 10 cm. Go and G3 are the liquid components of the ow from Po (stagnation) to Ps (saturation) for subcooled inlet conditions, and for Ps to P2 (discharge). For all-liquid ow (no ashing), Ps is replaced by P2 and Gc = Go. For choked ow, P2 is replaced by the choke pressure Pc. The friction loss coefficient Kf can include entrance and tting losses, as well as pipe friction for which Kf = 4fL/D; f is the Fanning friction factor. This model does not include a provision for accounting for slip. The predictions of these various models were compared with the following data sets from the literature to test their validity.

(9)

Alternatively, this parameter can be determined from known values of the two-phase density at the stagnation pressure and one additional pressure between the stagnation and discharge pressures, as specied by the TPHEM program. This is usually more accurate than using Eq. 9.

HNE model This model has been expressed in various ways (5). For ashing ows, it can be represented by the following set of equations:
Gc = G1 Go G1
2 1/2

+ 1 N NE

1 + Kf

(10)

N NE =

G1 2 L + Le G3 for L L e = 10 cm

(11) (12) (13) (14) (15)

NNE = 1
Go = G1 = G3 =

for L > Le
2 Lo Po Ps

Data: Frozen ow The most consistent set of data for frozen flow appears to be that of Graham (11) for air/water mixtures through short (ASME-type) nozzles. Jamerson and Fisher (7) compared the TPHEM computer predictions with these data, and found that good agreement is obtained using a constant slip value S of 1.51.8. An example of these data is shown in Figure 2, in which the dimensionless mass flux, defined as:
G* = Gc Po o

(16)

h LGo = G ERM vLGo T oC pLo 2 Lo Ps P2

is plotted vs. the inlet quality xo for the Graham Nozzle No. 1 (e in.), along with a comparison of the predictions of the TPHEM homogeneous equilibrium model, the TPHEM model with a constant slip ratio of S = 1.5, and the HNE model. The quality is dened as the mass fraction of the vapor (gas) phase, i.e.:

CEP

May 2001

www.aiche.org/cep/

59

Safety

x = L = L = s G L GL GLs

(17)

where = 1/ is the two-phase specic volume, subscripts L and G refer to the liquid and gas phases, and s refers to saturation conditions for ashing ows (subcooled inlet conditions correspond to a negative quality, since < Ls). It is evident that the HEM model drastically underpredicts the data, whereas including slip in the model with S = 1.5 results in close agreement with the data. The HNE model is much better than the HEM model, and agrees with the data at very low and very high values of quality, but, underpredicts G* by up to 20% for qualities near zero. Jamerson and Fisher (7) also analyzed the data of Toner (12) for air/water mixtures in longer converging/diverging nozzles, and found that slip ratios of 1.11.2 gave a reasonable t of these data.

Data: Flashing ow in nozzles The most extensive data in the literature for ashing twophase ow are for steam/water mixtures, mostly from the nuclear power industry. In an Electric Power Research Institute (EPRI) Report, Ilic et al. (13) compiled over 70 data sets from 20 different sources for steam/water ows, which were culled for inconsistent and incomplete data. Thus, this represents what is claimed to be a comprehensive reliable database for steam/water in various nozzles and pipes. The most extensive and consistent of these data sets are those of Sozzi and Sutherland (14) for ows in various lengths of straight and converging/diverging nozzles and tubes. Of particular interest are the data for their Nozzles 2 and 3. Nozzle 2 has a well-rounded entrance that converges from 1.75 in. to 0.5 in. over a distance of 1.75 in., with ten lengths of 0.5 in. straight tubing from 0 to 70 in. long attached. Steam/water ow rates were measured at inlet pressures from 8001,000 psig and inlet quality conditions from -0.004 to 0.007. Nozzle 3 was a square-entrance, 0.5 in. dia. tube, with lengths of 0.18525 in., and inlet conditions similar to those of Nozzle 2. The database includes mass ux, inlet and exit pressures, quality, and enthalpy for each data set for a given pipe/nozzle length. The uid densities and quality of the mixture downstream of the entrance were determined from steam tables assuming isenthalpic ow. The Sozzi and Sutherland data for Nozzle 2 and Nozzle 3 were compared with the HNE and the various TPHEM options by Darby (8) and Darby et al. (15). The Nozzle 2 data were compared with the predictions of the TPHEM equilibrium model, TPHEM with the slip option using the value of the parameter ks that gave the best average t of the data set (to the closest 0.1 unit of ks), the TPHEM nonequilibrium option with the kNE parameter adjusted to give agreement with each data point (to within 1%), the TPHEM nonequilibrium option with the one value of kNE representative of the average value of kNE for the data set, and the HNE model.
60
www.aiche.org/cep/ May 2001 CEP

For the HNE calculation, the value of the nozzle choke pressure (as given in the database or determined by the TPHEM program) was used in place of P2 for the exit pressure, since choked conditions prevailed for all runs. The three-parameter Model F for two-phase density was used for each case. Examples of the results for L = 0 and 1.5 in. are shown in Figures 3 and 4. The results demonstrated that the value of kNE required to t the Nozzle 2 data increases with the nozzle length, from kNE = 1 for L = 0 to kNE > 20 for L = 9 in. The best t value of kNE depends upon both the nozzle length L and the stagnation quality xo, increasing with both the nozzle length and inlet quality for superheated entrance conditions. It also increases with increasing subcooling, with a minimum value at or near saturated inlet conditions. However, an average value of kNE for all data points for each data set (i.e., nozzle length) also gave a reasonable t. An empirical equation was developed to represent this dependence (8), although it is not recommended that this expression be extrapolated beyond the range of these data. The TPHEM equilibrium model was consistently low for the shorter nozzle lengths, but gave good results for the longer nozzles/pipes (L = 7.5 in. and longer), and improved as the inlet quality increased. It was not possible to get agreement with the data by adjusting the slip parameter ks in the TPHEM model for the shorter nozzles or low quality, but a reasonable t was possible as the quality increased. For the longer nozzles (L > 2.5 in.), a good t was obtained for the higher quality ows with ks = 1.5. The goodness of fit was measured by the average and standard deviation of the ratio Gobs/Gcalc for each model and for each data set. A value of Gobs/Gcalc = 1 indicates a perfect match between the model and the data, and the lower the standard deviation the more consistent the prediction. Except for the nonequilibrium TPHEM model option, which was a forced-fit of the data, the HNE model with no arbitrary adjustable parameters gave the most consistent results overall. The agreement of this model was consistently good for all inlet quality conditions for all nozzles from 1.59 in. The values of Gobs/Gcalc varied from about 0.921.04, with a standard deviation as low or lower than the equilibrium models, except for L = 0. For the L = 0 nozzle, the fit was excellent near saturation, and within about 15% for higher and lower quality (see Figure 3). For the square entrance Nozzle 3, the only data set that corresponded to nozzle conditions (i.e., L < 4 in.) was L = 0.185 in. (Figure 5). The nonequilibrium and slip options in TPHEM did not give consistent results for the Nozzle 3 data, so only the TPHEM nozzle option with friction, the TPHEM pipe option, and the HNE models were compared with these data. For this geometry, the TPHEM nozzle and pipe models agreed with the data for high subcooling (liquid ow), but underpredicted the data by 200250% for

1.4 1.2 1.0 Data HNE TPHEM ks = 3 kNE = 1 kNE Fit

0.9 0.8 0.7 Data HNE TPHEM ks = 2 kNE = 5 kNE Fit

G*

G*

0.8 0.6

0.6 0.5 0.4 0.3

0.4 0.2 -0.002

0.002

0.004

0.006

0.008

-0.003

-0.002

-0.001

0.001

0.002

0.003 0.004

Xo

Xo

Figure 3. Sozzi and Sutherland Nozzle No. 2 data; L = 0. two-phase low-quality ow, while the HNE model overpredicted this data set consistently by about 1525% over the whole range of quality.

Figure 4. Sozzi and Sutherland Nozzle No. 2 data; L = 1.5 in.

1.2 1.0

Data: Flashing ow in tubes Nozzles 2 and 3 with straight lengths from 7.570 in. were considered to be pipes, with an appropriate wall roughness and entrance loss coefficient. These data were t by the TPHEM nozzle ow option with a loss coefficient to account for both the entrance loss and the pipe friction loss, the pipe ow option, and the HNE model. The pipe ow TPHEM option uses the following equation for pipe ow, which is the equivalent of Eq. 3 for nozzles:
P2 2 dP + g Z o Z 2 avg

Data TPHEM Pipe, e = 0 HNE, Kf = 4fL/D TPHEM, Kf = 4fL/D

0.8

G*

0.6 0.4 0.2 0 -0.006

-0.004

-0.002

0.000

0.002

0.004

0.006

0.008

G =

Po

Xo

ln 2/ o + K f + 1 / 2
max

(18)

Figure 5. Sozzi and Sutherland Nozzle No. 3 data, L = 0.185 in. HNE model was excellent for saturated and subcooled conditions, but, for higher inlet qualities, the model overpredicts the data by about 10%. Sample results are shown in Figures 6 and 7 for Nozzle 2 tubes 12.5 and 70 in. long, and in Figure 8 for a Nozzle 3 tube 20.2 in. long. All of the models (TPHEM nozzle, TPHEM pipe, and HNE) gave reasonable agreement with the pipe data when the friction loss was included in the model. The HNE model gave as good or better agreement than the other models, with values of Gobs/Gcalc of 0.981.08 for the Nozzle No. 2 tubes, and 0.860.97 for the Nozzle No. 3 tubes, with a standard deviation as low or lower than the TPHEM models. A summary of the various TPHEM program options that were run, along with the corresponding values of the various program parameters, is shown in Table 1.

For the TPHEM nozzle option, the pipe friction loss coefficient is given by 4fL/D. The Fanning friction factor f was taken to be that for fully turbulent flow in a 0.5 in. dia. tube with a roughness of 0.0004 in., typical of that for stainless steel (i.e., f = 0.00465). The TPHEM pipe flow option requires an input of the pipe roughness (which was also taken to be 0.0004 in. for these data); the friction factor and loss coefficient are computed by the program. The same value of Kf = 4fL/D was also used in the HNE model for pipe flow. A square entrance loss coefficient of 0.4 was included with the Nozzle 3 computations (the entrance loss coefficient is zero for Nozzle 2). For the L = 9 in. No. 2 nozzle, the agreement of the

CEP

May 2001

www.aiche.org/cep/

61

Safety

3. Crosby JLT/JBS valve Five data points were reported over a range of 200500 psig with (subcooled) inlet qualiTPHEM Option Parameters* ties from -0.0099 to -0.123. The nozzle tapered from a dia. of 11 in. over a distance Case Description IU IC IPTS IV INES X of 3r in. to a uniform straight section with a dia. of 1.065 in. and a length of TPHEM Nozzle w/ friction 3 1 3 1 --0.875 in. The dimensions of this straight 3 1 3 1 --TPHEM Nozzle w/o friction section were taken to be the controlling nozzle dimensions for the HNE model. TPHEM Nozzle w/ slip 3 3 3 1 2 1.5 4. Data were reported by Bolle et al. TPHEM Nozzle KNE fit 3 3 3 1 11 Various (16) for a Crosby JLT/JOS-15A valve, with a nozzle that converged to a straight TPHEM Pipe w/ friction 3 1 3 -3 --section 0.409 in. long and 0.409 in. dia. The entering fluid was subcooled water HNE w/ friction ------and 16 data points were reported (six of HNE w/o friction ------which were all-liquid discharge), over a pressure range of 26 bar. * TPHEM Parameters: IU Units: 3 = Metric 5. Two-phase data were reported by IC Case: 1 = gives flow rate output; 3 = gives flow rate Lenzing et al. (17) for ow in ve differoutput and also activates INES, advanced options ent valves: Albert Richter (ARI) IPTS Two-phase density model and number of data states: DN25/40, Babcock Sempell SC 01, Bopp 3 = Three-Parameter Model F (3 input data states) IV Input Options: 1 = simple nonviscous nozzle input; & Reuther Si 63 DN25/40, Crosby -3 = pipe input without viscosity correction JLT/JBS, and Leser 441 DN25/40, INES Advanced Options: 2 = user inputs slip ratio, S; DN40/65. The only data reported for 11 = user inputs kNE nonequilibrium parameter. ashing ow were for water in the Leser X Value of Advanced Option parameter (S or kNE ) DN25/40 valve, at a pressure of 10.6 bar, These cases gave poor results, and were eliminated from the final comparison. over a range of inlet quality of The k NE parameter was varied to determine the value that fit each data point, and 0.0010.05. The nozzle minimum dia. varied from 0 to 75. was given as 23 mm, but no nozzle geometry or other dimensions were reported. Data: Flashing ow in valves The calculated mass ux through each valve nozzle was There is a paucity of suitable data in the literature for compared with the reported measured values, and the retwo-phase ow in actual relief valves. A comparison of the sults were expressed as an equivalent discharge coefficient: above models with limited data for ashing steam/water ows in valves was given by Darby et al. (15). A summary Kd = Gexp/Gcalc (19) of these results is given below, with a description of the valve/nozzle used in each study. The models compared with these data were the equilib1. Data were provided by Anderson Greenwood Crosby, rium TPHEM nozzle model with the three-parameter F Inc. for ashing ow in a modied Crosby Series 900 density model, (Kd)F; the one-parameter homogeneous valve, with two different straight uniform-bore nozzles TPHEM nozzle model (comparable to the Omega method), with a smoothly contoured entrance. Both nozzles had a (Kd)omega; the nonequilibrium TPHEM nozzle option bore of 0.4 in. dia., with the shorter nozzle being 0.5 in. with an empirically adjusted value of kNE, (Kd)kne; and long and the longer 5.0 in. Two data points were taken with the HNE model, (Kd)HNE. The HNE model has no adeach nozzle, one at a low temperature (all liquid ow) and justable parameters, and is the only model that is sensitive one at high temperature (saturated water at 365F). to the length of the nozzle. Data were also provided by Anderson Greenwood CrosThe results of tting the various models to each data by, Inc. from tests on the following two valves: point for each of the ve data sets were reported by Darby 2. Crosby 9511 valve The nozzle had a e in. dia. et al. (15). Comparison of the various model predictions inlet straight section that tapered to a 2 in. straight section, with the valve capacity data showed mixed results, i.e., the 0.956 in. long. The diameter and length of this straight secmodel that gave the best t of the data varied from valve to tion were taken to be the governing nozzle dimensions. valve. This could be attributed to the limited range and Four data points were recorded with this valve, two of scope of the data, undened uncertainties in these data, or which were all-liquid discharge and two were ashing twothe deviation of the valve nozzle geometry from the ideal phase discharge at 330 psig. straight bore rounded entrance nozzle.
Table 1 - TPHEM Options Run and Program Parameters

62

www.aiche.org/cep/

May 2001

CEP

Safety

0.8 0.7 0.6 Data TPHEM, Kf = 0 TPHEM Pipe, e = 0 HNE, Kf = 4fL/D TPHEM, Kf = 4fL/D TPHEM, kNE Fit

0.8 0.7 0.6

Data HNE, Kf = 4fL/D TPHEM Pipe e = 0 TPHEM, Kf = 4fL/D

Gc *

Gc *

0.5 0.4 0.3 0.2 -0.002 0

0.5 0.4 0.3 0.2

0.002

0.004

-0.004

-0.002

0.000

0.002

0.004

Xo

Xo

Figure 6. Sozzi and Sutherland Nozzle No. 2, pipe data; L = 12.5 in. Thus, any conclusions based upon these data should be considered to be tentative, at best. This being said, most of the models gave reasonable results for most of the valve data (all Kd values were between 0.71.3), with the exception of the Bolle data, for which all models consistently underpredicted the values by an average of about 30% (which would seem to indicate some systematic error or offset in these data). The equilibrium TPHEM model with the 3-parameter (F) density model gave consistently somewhat better results than the one-parameter (Omega) density model, although for the majority of the data points, these values agreed to within about 10%. Overall, however, the HNE model consistently gave the best results, except for the Lenzing data at the highest quality, for which it was about 1520% low (for an assumed nozzle length of 4 in.). In general, the homogeneous TPHEM model with the F density model gave results almost as good as the HNE model, except for the valves with the shortest nozzles. As an example of the sensitivity of the data to entering conditions, the uid entering the Crosby 900 valve was reported to be saturated at 365F and 150 psig. However, the saturation temperature at this pressure is actually 365.87F. This difference of less than 1F translates to a variation of about 10% in the predicted mass ux and the corresponding values of Kd (i.e., a small error in T can result in a large error in pressure). It was possible to get a good t of most of the valve data using the TPHEM nonequilibrium option by adjusting the parameter kNE. However, the values of kNE needed to t the data varied from about 2.516, with no apparent correlation with the uid properties or conditions. For some data points, the best t was not very sensitive to the actual value of kNE as this value increased.

Figure 7. Sozzi and Sutherland Nozzle No. 2 data, L = 70 in.

0.8 0.7 0.6 Data HNE, Kf = 4fL/D TPHEM Pipe e = 0.0004 TPHEM, Kf = 4fL/D

Gc *

0.5 0.4 0.3 0.2 -0.004

-0.002

0.000

0.002

0.004

Xo

Figure 8. Sozzi and Sutherland Nozzle No. 3 data, L = 20.2 in.

Recommendations The Sozzi and Sutherland database for ashing ow of water in nozzles and tubes at approximately 1,000 psia is much more extensive and consistent than the available data for ashing ow in valves. Based on these data, the HNE model gives reasonably accurate values over the widest range of inlet quality and nozzle length conditions, when the effect of friction loss is included for the longer nozzles and tubes. The equilibrium TPHEM models signicantly underpredict the mass ux for subcooled and saturated inlet conditions in short nozzles, but are in good agreement with the data for higher inlet quality conditions and longer tubes when friction loss is included. It is possible to get good agreement with the data for the short Nozzle No. 2 (rounded entrance) data with the nonequilibrium option of
63

CEP

May 2001

www.aiche.org/cep/

Safety

TPHEM by adjusting the kNE parameter, but this does not work very well for the Nozzle No. 3 (square-entrance) nozzle and pipe data. Although it has been shown that slip can be important in frozen ows, it should not be signicant in ashing ows relative to the effects of ashing. The available data for valve ow are much more limited, and are complicated by the variable geometry of the valve nozzle. Because the ashing pressure for these data is considerably below that of the Sozzi and Sutherland data (typically about 100 psia), nonequilibrium effects are much less pronounced. Considering the limited extent and the uncertainty in these data, the tentative conclusion with re-

Literature Cited
1. Pressure Relief Device Certications, National Board of Boiler and Pressure Vessel Inspectors, Columbus, OH (updated annually). 2. Emergency Relief System Design Using DIERS Technology, Design Institute for Emergency Relief Systems (DIERS), AIChE, New York (1992). 3. Leung, J. C., Easily Size Relief Devices and Piping for Two-Phase Flow, Chem. Eng. Progress, 92 (12), pp. 28-50 (Dec. 1996). 4. Guidelines for Pressure Relief and Effluent Handling Systems, Center for Chemical Process Safety (CCPS), AIChE, New York (1998). 5. Fauske, H. K., Determine Two-Phase Flows During Releases, Chem. Eng. Progress, 95 (2), pp. 5558 (Feb. 1999). 6. Darby, R., Viscous Two-Phase Flow in Relief Valves, Phase I Report to Design Institute for Emergency Relief Systems (DIERS), AIChE, New York (Nov. 1997). 7. Jamerson, S. C., and H. G. Fisher, Using Constant Slip Ratios to Model Non-Flashing (Frozen) Two-Phase Flow Through Nozzles, Process Safety Progress, 18 (2), pp. 8998 (Summer 1999). 8. Darby, R., Evaluation of Two-Phase Flow Models for Flashing Flow in Nozzles, Process Safety Progress, 19 (1), pp. 3239 (Spring 2000). 9. Simpson, L. L., Estimate Two-Phase Flow in Safety Devices, Chem. Eng., 98 (8), pp. 98102 (Aug. 1991). 10. Simpson, L. L., Navigating the Two-Phase Maze in International Symposium on Runaway Reactions and Pressure Relief Design, G. A. Melham and H. G. Fisher, eds., Design Institute for Emergency Relief Systems (DIERS), AIChE, New York, meeting held in Boston, MA, pp. 394417 (Aug. 24 1995). 11. Graham, E. J., The Flow of Air-Water Mixtures Through Nozzles, Ministry of Technology NEL Report No. 308, U.K. (Aug. 1967). 12. Toner, S. J., Two-Phase Flow Research, Phase I, Two-Phase Nozzle Research: U.S. Department of Energy Report, DOE/ER/10687-T1 (July 1981). 13. Ilic, V, et al., A Qualied Database for the Critical Flow of Water, Electric Power Research Institute (EPRI) Report NP-4556, EPRI, Palo Alto, CA (May 1986). 14. Sozzi, G. L., and W. A. Sutherland, Critical Flow of Saturated and Subcooled Water at High Pressure, Report NEDO-13418, General Electric Company, San Jose, CA (July 1975). 15. Darby, R., et al., Relief Sizing for Two-Phase Flow, paper presented at 34th Loss Prevention Symposium, AIChE, Atlanta (Mar. 59, 2000). 16. Bolle, L., et al., Experimental and Theoretical Analysis of Flashing Water Flow Through a Safety Valve, J. Hazardous Materials, 46, pp. 105116 (1996). 17. Lenzing, T., et al.,Prediction of the Maximum Full Lift Safety Valve Two-Phase Flow Capacity, J. Loss Prevention in the Proc. Industries, 11 (5), pp. 307321 (1998).

gard to the suitability of the models is roughly the same as for the nozzle data. Within the uncertainty of the data, the HNE model gave the most consistent predictions for nozzles, pipes, and valves, as well as for rounded and square entrance congurations. It is also the most consistent over the range of inlet quality from subcooled to two-phase inlet and for short nozzles to long pipes and valves. For ashing ow in valves, a discharge coefficient Kd of 1.0 is tentatively recommended for use with the HNE model based on these results. This is reasonable, since ashing ows are inevitably choked, so that the valve capacity is determined by ow in the nozzle only and is independent of the ow conditions downstream of the nozzle in the valve body, etc. Values of Kd that are signicantly lower than 1.0 normally result for gases under subcritical conditions and liquid ows, for which the ow in the valve body has a signicant effect on the valve capacity. It should be emphasized that to obtain reliable results from the models, it is necessary to have a complete understanding of the thermodynamic and physical state of the uid entering and leaving the nozzle, which implies accurate values of temperature and pressure, as well as uid transport and thermodynamic properties at these conditions. For example, for an entering uid near saturation, a variation of 1F in the temperature can have an effect as large as 20% on the calculated mass ux through the valve.

Acknowledgments The authors would like to recognize the DIERS Users Group of AIChE and ALCOA, Inc. for partial support of this project, and to Anderson Greenwood Crosby, Inc. for CEP providing the data for the Crosby valves.

< Discuss This Article! >

R. DARBY is a professor of chemical engineering at Texas A&M University, College Station, TX (Phone: (979) 845-3301; Fax: (979) 845-6446; E-mail: r-darby@tamu.edu). He has been active in research on non-Newtoninan and viscoelastic fluids, two-phase solid-liquid and gas-liquid flows, and has numerous publications in these and other topics. He has authored Viscoelastic Fluids, and the recently revised version of Chemical Engineering Fluid Mechanics, published by Marcel Dekker. Darby holds BS and PhD degrees in chemical engineering from Rice University. He is a Fellow of AIChE, and has received numerous publication awards from AIChE, a Former Students Award from Texas A&M for excellence in teaching, and was designated Minnie Stevens Piper Professor in 1981. P. R. MEILLER is a senior chemical engineering student at Texas A&M. He served an internship last summer with BASF in Ludwigshafen, Germany, and will work for BASF in Freeport, TX, after graduation. J. R. STOCKTON is a product engineer with Ruska Instrument Corp., Houston (Phone: (713) 975-0547; Fax: (713) 975-6338). He received his BS in engineering from Trinity University in San Antonio and his ME in chemical engineering from Texas A&M. He is a member of AIChE.

64

www.aiche.org/cep/

May 2001

CEP