Commercial Scale Experiments That Provide Insight on Packed Tower

May 24, 1987 - reflux ratio at q = 1 are available (Glinos and Malone,. 1984). For binary mixtures, this approximation gives al- most exact results in...
0 downloads 0 Views 1MB Size
Ind. Eng, Chem. Res. 1987,26, 1845-1850 1.5 and is useful for binary, ternary, and quaternary mixtures where simple approximate expressions for the reflux ratio a t q = 1 are available (Glinos and Malone, 1984). For binary mixtures, this approximation gives almost exact results in comparison to the analytical solution (Treybal, 1980),which is a second-order polynomial in feed quality, feed composition, and distillate composition. In Figure 5, we present a comparison of the approximate and exact reflux ratios for ternary A/BC and AB/C splits. For the range of feed quality examined, the error never exceeded 7%. Literature Cited

1845

Glinos, K.; Malone, M. F. Ind. Eng. Chem. Process Des. Deu. 1984, 23, 764. Glinos, K.; Malone, M. F. Znd. Eng. Chem. Process Des. Deu. 1985, 24, 822. Levy, S.; Doherty, M. F. Znd. Eng. Chem. Fundam. 1986, 25, 269. Petlyuk, F. B.; Platonov, U. M.; Slavinskii, D. M. Znt. Chem. Eng. 1965, 5(3), 555. Simulation Science Inc., PROCESS, Revision 4, Sept 1985. Treybal, R. E. Mass Transfer Operations, 3rd ed.; McGraw-Hill: New York, 1980. Underwood, A. J. V. J.Znst. Pet. 1946, 32, 614. Underwood, A. J. V. Chem. Eng. Prog. 1948, 44(8), 603. Van Winkle, M. Distillation, 2nd ed.; McGraw-Hill: New York, 1981. Yawns, C. L.; Ku-Yen, Li; Fang, C. S. Chem. Eng. 1981,88(10), 153.

American Institute of Chemical Engineers Student Contest Problems 1950-1958; Wiley: New York, 1958; p 221. Elaahi, A.; Luyben, W. L. Znd. Eng. Chem. Process Des. Deu. 1983, 22, 80.

Received f o r review January 9, 1986 Revised manuscript received April 21, 1987 Accepted May 24, 1987

Commercial Scale Experiments That Provide Insight on Packed Tower Distributors John

G. Kunesh,* L a w r e n c e Lahm, and Takashi Yanagi

Fractionation Research, Znc.,South Pasadena, California 91030

Recent research has shown that liquid distribution is even more important to the apparent efficiency of packed columns than was previously believed. A patented “Adjustable Liquid Distributor” which allows distribution t o be changed with the column in operation has been designed and was used to gain new insight into packed column operation utilizing 1-in. Pall rings and the cyclohexaneln-heptane system a t atmospheric pressure. Preliminary conclusions are t h a t discontinuities or step changes in the flow from various zones of the distributor have the most severe impact on efficiency, while modest amounts of tilting and sagging are tolerable. The effect of liquid distribution on packed tower performance has been the subject of both experimental and theoretical work for over 50 years. One of the earliest significant studies, and one which is still being cited today, was that of Baker et al. (1935) which established 8 as the ratio of tower diameter to packing size below which significant wall flow developed. A great deal of the work since then, such as Billet and Mackowiak (1982), has concentrated not only on distribution but also on how it affects scaling up laboratory column results to a commercial size. Maldistribution, which becomes more and more of a concern as column diameters become larger, is often cited as the reason for the extensive scatter in published efficiency data (Hoek, 1983) and is blamed for the failure of commercial columns to produce the results predicted from laboratory tests (Billet and Mackowiak, 1982). Most authors who have attempted to model the flow in random dumped packed beds start with a fine scale statistical approach. Cihla and Schmidt (1957) employed a Gaussian distribution function for flow through a differential element of a packed bed and found that the resulting expression had the form of the equations used to describe molecular diffusional processes. They then obtained series solutions for several particular boundary conditions. Hoek (1983) used numerical solutions of the same differential equation to model his experimental results in which an imposed “large scale” maldistribution (such as 20% nonirrigation at the top of the bed) slowly evolved into “small scale maldistribution” or channeling which he concluded is an inherent and stable property of the packing. Albright (1984) used a random number technique to model flow

through a packed bed and obtained similar results to Hoek’s, except that he chose to utilize the term “natural distribution” to describe the fine scale flow pattern which he also considered to be a property of the packing. He concluded that an initial distribution that is better than the natural distribution will degrade to it quite quickly. Conversely, an initial distribution that is poorer than natural distribution will ultimately achieve it but sometimes at a very slow rate. For example, he computed that initial regions of nonhomogeneous flows such as a nonlevel drip pan could double the distance required to reach “natural flow” compared to a homogeneous initial distribution. This could seriously impact the anticipated overall column efficiency. The issue thus facing the distributor designer/column operator is not only how to design a good distributor but, perhaps more importantly, what to set as the fabrication and installation tolerances. Ideal Distribution An example of the effect of initial liquid distribution on packed column performance comes from FRI’s data bank. Early experimental work (Silvey and Keller, 1966) employed a notched trough distributor (Figure 1) almost exclusively. A t the suggestion of Zuiderweg (1983), a Tubed Drip Pan Distributor similar to that described by Billet (1979) was fabricated (Figure 2). However, Billet’s 600 pour points per square meter were reduced to approximately 100 corresponding to the inherent channel density or “natural distribution” found by Hoek. The dramatic results are presented in Figure 3. Using this distributor, it is now possible to determine the inherent

0888-5885/87/2626-1845$01.50/0 0 1987 American Chemical Society

1846 Ind. Eng. Chem. Res., Vol. 26, No. 9, 1987 LOCKNUT

&PACKED

I 1

=

Y

Y

'

Y

Y

Y

Y

Y

Figure 1. Notched trough distributor.

1/2' STEEL PLATE

1

"f-

STOP

I

-

I A-

48'

SI CONVERSION: mm

.

ln z 15.4

Figure 4. Adjustable liquid distributor.

Figure 2. Tubed drip pan in service.

EFFICIENCY 1 INCH PALL RINGS C&, 24 PSlA 2.0

1 7

0

5

1.5

Q

LiI

\

TUBED DRIP PAN

SI CONV€RSION: LPa =pel m 0.89

0

26

60

76

100

PCT OF USEABLE CAPACITY

Figure 3. Efficiency comparison, notched trough vs. tubed drip pan.

efficiency of any packing under ideal distribution conditions. However, the Tubed Drip Pan Distributor as designed, fabricated, and assembled for this study is commercially impractical. For example, the problem of warpage during welding was solved by employing 1/2-in.-thick (12.7-mm-thick) steel plate for the floor of the pan. The major deviation from a commercial distributor, however, lies in the unique construction of the commercial size experimental column. The entire 4-ft-diameter (1.22-m-diameter) top head is removable. This means that the distributor does not have to be fabricated in sections which must be passed through a manway for assembly inside the column. The pan was fabricated in one piece for calibration and testing on the ground and then lifted intact and placed in the column where final leveling took place. Therefore, it was decided to initiate a program to quantify the penalty for deviation from ideal distribution and thereby provide guidelines for the design, fabrication, and

Figure S. Adjustable liquid distributor.

installation of commercial distributors.

The Adjustable Liquid Distributor From the moment that a controlled maldistribution program was first contemplated, it was recognized that the necessity of shutting down and opening up a 4-ft-diameter column for each distribution change would be prohibitively expensive and inordinately time consuming. Therefore, it would be necessary to change the distribution in a controlled manner while the column was in operation. The design finally chosen consisted of supplying liquid to each pour point by means of a slotted cylinder and piston arrangement operated by a rod extending through the top head of the column. This was made possible by equipping the column with a temporary flat head and sealing each rod by passing it through a "Swagelok" fitting with a Teflon ferrule. Operation was restricted to atmospheric pressure. Figure 4 is a drawing of the device, and Figures 5 and 6 are photographs. The cylinders were installed with the bottom of each slot 1/2 in. (12.7 mm) above the pan floor to avoid plugging by scale or dirt. The slotted cylinder piston pairs were then adjusted so that the bottom of each slot and the bottom of each piston were the same distance from the floor of the pan. Piston adjustment (slot opening control) was accomplished by means of a wing nut and brace arrangement. The piston was bottomed against an internal stop nut with the wing nut in an elevated position. The wing nut was then screwed down until it met resistance from the brace, and the piston was raised by the

h d . Eng. Chem. Res., Vol. 26, No. 9, 1987 1847 DEVIATIONS FROM AVERAGE FLOW

-

.. -

. -

-... --

..

1 2 1 TUBES,

: ’!--

.-

---_---

0 . 7 5 In. SLOT OPENINGS

3 In. depth, 5 2 gpm

I

Figure 6. Adjustable liquid distrihutor mean: 28.1 ml/8

I-

utd. dev.: 1.8 ml/r

. -- - -----.-J Figure 9. Calibration results.

EFFICIENCY 1 INCH PALL RINGS c,-c,.TDP vs ALD

0 + a I

Figure 7. \Vater calibration.

3 A

1.5

0. kW

1

--

’ SSFRPU I *2.0 lF. (

0

TOP. 24 PSIA

25

50

15

100

PCT OF USEABLE CAPACITY

Figure 10. Efficiency comparison, tubed drip pan vs. ALD.

Figure 8. Setting pistons at top of column during operation.

appropriate number of turns of the wing nut. Since the threading on the rod is 20 revolutions/in., it is believed that this arrangement allowed certainty of piston elevation to about 0.5 mm (0.02 in.). The device was calibrated with water a t grade by measuring the flow through each drip point and making adjustments until satisfactory uniformitv of flows was achieved. Figure 7 is a photograph of the calibration, and Figure 8 shows the adjusting rods a t the top of the column. Figure 9 presents the results of the final calibration prior to installation in the column. The small triangles point in the direction of deviation from the mean, and their height is proportional to the magnitude. The standard deviation is 6.4% of the mean which was felt to be as low as could be practically achieved. This was tested by replicating the previous Tubed Drip Pan runs a t the outset of the experimental program. Figure 10 presents a comparison of the ALD results with the previous results

Figure 11. Liquid sampling device.

The excellent agreement provided the necessary confidence to embark on the full experimental program. There were two intermediate shut downs for minor adjustments during the program, and the base-line run was repeated after each one. The average of these runs is referred to hereafter as ..base”.

Experimental Program As mentioned previously, the necessity of sealing all of the adjustment rods mandated running a t atmospheric pressure. The cyclohexaneln-heptane system which has been an FRI standard for 30 years was utilized with a 12-ft (3.658-m) bed of I-in. (25.4-mm) stainless steel Pall rings. Liquid samplers (Figure 11) which have been described previouslv (Silvey and Keller, 1966) were installed a t 2-ft

1848 Ind. Eng. Chem. Res., Vol. 26, No. 9, 1987 INSTALLATION DIAGRAM .R

ADJUSTABLE LIQUID DISTRIBUTOR

6' LIQUID INLET

-

4

f

1/2' STEELPLATE

.

c

80'

WINDOWS-

.-

L .

-

S SAMPLE T =TEMPERATURE

1 BUBBLE CAP (L 3 SIEVE TRAYS, 6' SPACING

a

1 ---{h 1

BE0 DEPTH

Figure 13. Typical profile; good initial distribution

x

14' VAPOR INLET -

-.

s\CawEnsion

m m -Jn 1 z 5 4

' R

Figure 12. Installation diagram.

/ A

i

intervals in the 4-ft column. Figure 12 is an installation drawing. The experimental program was designed to furnish design basis guidelines and insight into allowable tolerances. In all, a total of 104 runs were made, all at total reflux. Three boil-up rates were used for each distribution pattern to test for loading effects. The target values were 30%, 6070, and 90% of the maximum usuable capacity. Higher values were avoided because no holddown device was used, and lower values were not desired because very small liquid rates magnify small differences in piston settings. Rates are presented as percent of usable capacity following the approach of Strigle (1985). Our observations confirm his: modern, high open area packings remain hydraulically operable at rates well past the point where the efficiency has totally deteriorated; thus, percent of flood is irrelevant to efficiency comparisons. Usable capacity is the loading at which separating efficiency begins to rapidly deteriorate. Interpretation of Results The results were analyzed in two ways. The apparent overall HETP was determined, and individual composition profiles were analyzed. The apparent overall HETP is, of course, the only thing the operator of a commercial column would observe, but as Silvey and Keller (1966) pointed out, it is very sensitive to the handling of end effects and, in this case, to changing distribution. Analysis of bed profiles was done as follows: At total reflux, making the usual assumptions, the Fenske equation is NTheo

= In

01

In

[(

1-x

and bed depth HETP If CY and HETP are constant, a plot of In ( X / ( l - X))vs. bed depth should yield a straight line with slope In a / HETP. In a reasonably close boiling, almost ideal system, such as cyclohexaneln-heptane, CY may be taken as constant and NTheo

=

i \

X

I ~

SIMULATED SAG TOWARD CENTER

BED DEPTH

Figure 14. Typical profile; mild maldistribution with recovery.

it is generally accepted that the HETP should be almost constant. Therefore, a curving line implies that the effective L / V is changing. Consequently, if a packed bed achieves natural flow almost immediately, a plot of In (X/(1 - X))vs. bed depth will be a straight line. If an initial maldistribution is being corrected, the slope of the In (X/(1 - X))vs. bed depth plot will increase from its initial value a t the top of the bed until natural flow is achieved and then it will remain constant (unless wall flow develops which will flatten it out again). The data were examined in this manner and exhibited the anticipated behavior. Figure 13 shows the results from a base-line run where every effort was made to achieve good initial distribution. As may be seen, the profile is a straight line. (Note that reflux and bottoms compositions are indicated for information only but not used in any of the computations.) Figure 14 is a case of maldistribution with recovery. The profile in the lower portion of the bed is a straight line with a slope numerically equal to that for the good initial distribution case. The profile at the top of the bed shows the effect of moderate maldistribution, in this case, simulation of a uniformly sagging distributor, Le., simulation of an insufficiently stiffened distributor which sags under load from a high point at the wall to a low point in the center. Flow varies linearly with radius from a value of 1.0 at the wall to 1.25 at the center. Figure 15 presents a case where the maldistribution is so severe that there is no indication of recovery in 1 2 f t (3.658 m) of packing. This was an extreme case in which

Ind. Eng. Chem. Res., Vol. 26, No. 9, 1987 1849

2.0 0 BASE A 16% CENTRAL BLANK 11% CHORDAL BLANK

-

1.6

!i 1.0

\

:

1

I

0

1

BED DEPTH

Figure 15. Typical profile; severe maldistribution with no recovery. SIMULATED SEAM LEAK MED

LOW

26

50

75

100

PCT OF USEABLE CAPACITY

Figure 18. Uniform pattern tolerance.

EFFICIENCY

*

1 INCH PALL RINGS C&,1 ATMOS

1

0 BASE

BED DEPTH

Figure 16. Loading effect.

liquid flow to one-half of the bed was shut off. However, it dramatically illustrates the fact that recovery can be very slow. Initially, it had been hoped that a simple correlation such as “type and severity of maldistribution” vs. “distance to reach natural flow” could be developed. Unfortunately, this is not the case. There is a complex loading effect as may be seen in Figure 16 which presents the results of simulating leakage of a seam (all of the slots at a minimum opening with one row in the center wide open). For this situation there is complete recovery after about 6 f t (1.8 m) at the high and medium rate and no recovery at all after 12 f t (3.658 m) at the low rate. As noted above, it had been anticipated that different types of maldistribution would have greater or lesser effects. The types were chosen based on commercial design and installation problems such as out of levelness, sagging under load, and constrictions in piping network branches. Another question concerned the necessity of stripping old hardware (such as a seal pan) when converting a trayed column to packing. The seal pan was simulated by shutting off flow or blanking a chordal segment of the distributor. Figure 17 compares blanking a chordal segment containing 11% of the pour points to blanking an essentially circular central segment containing 16% of the pour points (simulation of a malfunctioning ring distributor). Results are presented in terms of the ratio of the apparent overall HETP to the base value. As can be seen, the 11% chlordal blank has a more harmful effect than blanking off 16% of the area in the center. Note also that chordal blanking was the only type of maldistribution in which the apparent overall HETP deteriorated with in-

I

0

26

50

76

100

PCT OF USEABLE CAPACITY

Figure 19. Effect of discontinuity.

creasing loading. For all others, it either was essentially constant or improved as the rates were increased.

Implications for the Designer/Installer/Operator Analysis of the detailed profiles from a fundamental point of view is proceeding. Additional experimental work is planned to extend the work to other packing sizes and to explore certain anomalies in the data. However, review of the apparent overall HETP results leads to some general observations of importance to the designer/installer/operator. First, comfort may be derived from the fact that there appears to be a reasonable tolerance for a uniform maldistribution pattern. Figure 18 shows that there is virtually no penalty if a distributor uniformly tilts or uniformly sags (center to wall or vice versa) such that the ratio of highest to lowest flow is 25%. However, if a discontinuity occurs, the results could be disastrous. A tilting distributor will ultimately begin to lose flow in a manner identical with the chordal blank discussed above. Figure 19 compares the 25 percent tilt, which is virtually identical with the base, to an 11% chordal blank of an otherwise level distributor. Another form of discontinuity is step changes in the flow rates from different zones of the distributor or zonal flow. This could be caused, for example, by an obstruction in

1850 Ind. Eng. Chem. Res., Vol. 26, No. 9, 1987 EFFICIENCY IINCH PALL RINGS C&,1 ATMOS

.

2 0

A

BASE 1 2 5 TILT

v 1 5 TILT 4 2 ZONE 1 2 5

..

I

I .

--_

1 0

0

26

*

75

60

100

PCT OF USEABLE CAPACITY

Figure 20. Effect of zone flow.

EFFICIENCY 1 INCH PALL RINGS C,-C;,l ATMOS 2.0

I

2

BASE RANDOM RANDOM ZONE

1.5

F -

~

-_

EzzZl 25

PCT

50

?6

100

OF USEABLE CAPACITY

Figure 21. Effect of zone flow.

a major branch point of a pipe grid type of distributor. Figure 20 compares tilts with a ratio of maximum to minimum flow of 1.25 and 1.5 to a situation where one-half of the distributor is passing 25% more liquid than the other half. As may be seen, the loss of efficiency for the zonal flow case is twice as great as for a “uniform” maldistribution with twice as much variation from maximum to minimum. Two runs which should be of particular interest to installers of distributors were made a t the suggestion of a member company. They reported that inspection of a large distributor following installation showed that the variations followed a Gaussian pattern. To investigate this situation, a random number generator was used to generate a Gaussian flow distribution (maxjmum/mean = 2 ) which was then randomly assigned. This was extended to sim-

ulate warping by dividing the distributor into six zones, three designated high and three low. All randomly generated flows greater than the mean were randomly assigned to a high zone and all less than the mean to a low zone. Figure 21 presents the results which show practically no effect of the purely random variation but the zonal flow had a 20% decrease in efficiency. Runs were also made in which pour point density was decreased. It was found that a significant reduction was possible before the effects could be observed. However, it must be remembered that this was achieved with a distributor that had been adjusted and leveled much more painstakingly than is usually practical or even possible commercially. Future work will test the combination of density reduction and nonideal distribution. Summary Studies of controlled maldistribution of reflux to a packed bed being used for distillation have shown that recovery to natural flow and constant HETP is an extremely complex function of many variables. However, a general conclusion which provides guidance to designers/installers/operators is that a packed bed has a reasonable tolerance for both a uniform or smooth variation in liquid distribution and for one that is totally random. However, the impact of discontinuities or zonal flow is much more severe. Additional work is in progress to obtain a more complete understanding of these phenomena. Literature Cited Albright, M. A. Hydrocarbon Process. 1984, Sept, 173. Baker, T.; Chilton, T. H.; Vernon, H. C. Trans. Am. Znst. Chen. Eng. 1935, 31, 296-315. Billet, A. Distillation Engineering; Chemical Publishing: New York, 1979; p 103. Billet, R.; Mackowiak, J., EFCE Working Party on Distillation, Absorption and Extraction Meeting in Helsinki, June 2-4, 1982. Cihla, Z.; Schmidt, 0. Collect. Czech. Chem. Commun. 1957,22,896. Hoek, P. Doctoral Thesis, Delft University, 1983. Silvey, F. C.; Keller, G. J. Chem. Eng. Prog. 1966, 62, 68. Strigle, R. F. Chem. Eng. Prog. 1985, 81(3), 67. Zuiderweg, F., Fractionation Research, Inc., Consultant, personal communication, 1983. Received f o r review April 7 , 1986 Revised manuscript received May 8, 1987 Accepted May 24, 1987