Professional Documents
Culture Documents
An Improved Finite-Difference Calculation of Downhole Dynamometer Cards For Sucker-Rod Pumps PDF
An Improved Finite-Difference Calculation of Downhole Dynamometer Cards For Sucker-Rod Pumps PDF
An Improved Finite-Difference Calculation of Downhole Dynamometer Cards For Sucker-Rod Pumps PDF
Calculation of Downhole
Dynamometer Cards for
Sucker-Rod Pumps
T.A. Everitt, * SPE, Chevron Oil Field Research Co., and J.W. Jennings, * * SPE, Texas A&M U.
Summary. This paper presents a finite-difference representation of the wave equation developed for diagnostic analyses of sucker-
rod pumping systems. A consistent method of computing the viscous damping term associated with the damped-wave equation is also
presented.
Introduction
Sucker-rod pumping is the most widely used means of artificial lift. Knapp 3 was the first to present a method for computing down-
About 85% to 90% of all producing wells in the U.S. are rod- hole dynamometer cards using finite differences. His formulation
pumped. Thus, a reliable method of analyzing these pumping sys- . does not account for variable rod diameter or rod material. Knapp's
tems is a necessity. theory was used in the development of the model presented here.
For many years, the surface dynamometer has been used to ana-
lyze sucker-rod systems. Interpretation of actual pump conditions Model Development
from surface dynamometer cards is often difficult, if not impossi- The behavior of the sucker-rod pumping system is complex. This
ble. Results obtained from surface cards are strictly qualitative and study entails modeling a portion of this system, namely the sucker-
are dependent on the analyzer's expertise. rod string from the surface to the pump. The wave equation is ideal
The ideal analysis procedure would be to measure the actual pump for this purpose because the problem at hand involves the propa-
conditions with a downhole dynamometer. However, this situation gation of waves in a continuous medium.
is not economically feasible. Therefore, an accurate method of cal-
culating downhole pump cards from measured surface cards is need- Wave Equation. The ID wave equation is a linear hyperbolic dif-
ed. This paper presents a method for calculating these downhole ferential equation that describes the longitudinal vibrations of a long,
cards that uses a finite-difference representation of the wave equa- slender rod. Using this equation with viscous damping, we can ap-
tion. First, a brief description of previous calculation techniques proximate the motion of the sucker-rod string. In its simplified form,
is given. the wave equation is given by
o uuuuuuuuuuuuuuuuuuuuuuuuuuuuu
Eq. 3 is used to transmit the surface position downhole by cal-
uuuuuuuuuuuuuuuuuuuuuuuuuuuuu culating displacements at each node along the rod string until the
last node just above the pump is reached. The Appendix gives a
2 ?uuuuuuuuuuuuuuuuuuuuuuuuuuu? complete derivation of Eq. 3.
Note that Eq. 3 requires knowing displacements two nodes be-
3 ??uuuuuuuuuuuuuuuuuuuuuuuuu?? hind in space, Uj,j and Ui-l,j relative to the node being calculat-
-
~
III
CD
???uuuuuuuuuuuuuuuuuuuuuuu???
ed, ui+l,j. Therefore, to start the solution, the displacements uO.j
and Ul,j must be known for all timesteps. The displacements at
UO,j are known from the surface dynamometer card. The displace-
~ ????uuuuuuuuuuuuuuuuuuuuu???? ments at Ul,j are calculated with Hooke's law:
Z
?????uuuuuuuuuuuuuuuuuuu????? F=E4(ou/ox). . ................................. (4)
Substituting the polished-rod load, FPR, for F and a first-order-
??????uuuuuuuuuuuuuuuuu?????? correct forward-difference analog for ou/ox yields
???i???uuuuuuuuuuuuuuu??????? FpR,j=(EA/~x)(Ul,rUO,j) ........................ (5a)
or ul,j=(FpR,j ~/EA)+uO,j' ........................ (5b)
m ????????uuuuuuuuuuuuu????????
Thus, the displacements needed to start Eq. 3 are obtained. Note
u - Displacements That Can Be Calculated
that FPR is the dynamic polished-rod load (the surface-recorded
? - Displacements That Cannot Be Calculated
load minus the buoyed weight of the rods).
Fig. 1-Dlsplacements not calculated. Another interesting aspect of Eq. 3 is that the determination of
ui+l,j requires knowing the displacements ui.j-l and Ui,j+l. At
Finite-Difference Analogs. Finite differences are used in the model the lower end of the timestep scale (whenj=O), Ui,j-l does not
development to obtain the numerical solution of the ID, damped- exist; at the upper end of the timestep scale (whenj=n), Ui,j+l
does not exist. Therefore, the two endpoints at each node cannot
wave equation. Taylor series approximations are used to generate
be calculated. Fig. 1 illustrates this phenomenon.
finite-difference analogs for the derivatives of displacement that ap- If the calculations are started from the surface' node (i =0) with
pear in the wave equation. Eq. 3 is the result of substituting these only one cycle ofloads and positions, the solution at the pump (i =m)
Taylor series approximations into Eq. 2. will not represent a complete cycle because the endpoints at Nodes
2 through m cannot be calculated. To solve this problem, enough
Ui+l,j= ([a(l +C~t)]Ui,j+l -[a(2+c~t)-(E4/~x)+ surface points (loads and positions) must be repeated before the
-(E4/~x)-]ui,j+aui,j_l-(E4/~x)-Ui_l,j}/(E4/~x)+ , calculations begin so that when points are lost as the calculations
progress downhole, the pump card will still represent a complete
..................................... (3) cycle. The number of points that will be lost can be determined
Rod Data
Diameter, in. 0.75 Material Steel
Length, ft 2,000 Elasticity, psi 30.5x10 6
Downhole Data
Plunger diameter, in. 2.5 Pump depth, ft 2,000
Pumping speed, strokes/min 15 Fluid level, ft 2,000
Pump condition Full Fluid specific gravity 1.0
Pumping·Unit Data
Unit designation C·114·119·54 Stroke length, in. 54
Manufacturer Lufkin Rotation Clockwise
Rod Data
Diameter Length Elasticity
(in.) ~ Material (psi)
1.250 3,000 Fiberglass 8.03 x 10 6
0.875 3,000 Steel 30.5 x 10 6
Downhole Data
Plunger diameter, in. 1.5 Pump depth, ft 6,000
Pumping speed, strokes/min 10 Fluid level, ft 6,000
Pump condition Fluid pound Fluid specific gravity 1.0
Pumping·Unit Data
Unit designation C·228·213·100 Stroke length, in. 100
Manufacturer American Rotation Clockwise
-
iii
.a -
iii
.a
::!. ::!.
'0 '0
II II
0 0
...I ...I
o 20 40 60
Position (In.) Position (In.)
Fig. 2-Surface dynamometer card for Case 1. Fig. 3-Surface dynamometer card for Case 2.
from the number of nodes (number of lost points=number of noted the stability criterion of the finite-difference predictive model
nodesx2-2). For example, if the rod string has 20 nodes, 38 of as
the original surface points will not be transmitted to the pump. In .:1xlv.:1t~ 1. ....................................... (8)
this case, 38 surface points must be repeated before the calcula-
tions begin (Le., Point n + 1= Point 1, Point n + 2 = Point 2 ... Point Remember that the predictive and diagnostic models solve two
n+38=Point 38) to obtain a complete pump cycle. different problems. The predictive model calculates (predicts) sur-
As mentioned before, Eq. 3 transmits displacements downhole face dynamometer cards, whereas the diagnostic model calculates
to the node just above the pump. To obtain pump displacement, downhole cards from known surface conditions. The diagnostic
a different equation must be used because (EA) + and [pAl model solves for displacements ahead in space; the predictive model,
(144g c )] + do not exist at the pump. The equation used is the sim- on the other hand, solves for displacements ahead in time. Thus,
plified form of Eq. 3 for constant rod diameter and rod material: the stability criterion for the diagnostic model is not the same as
Eq. 8. Everitt 4 derived the stability criterion for the diagnostic
upump,j=(1 +c.:1t)U m _l,j+l -c.:1tu m - 1,j+Um -l,j-l -Um -2,j' model. For brevity, only the final result is shown here:
...................................... (6) .:1 xlv.:1t:ii 1. ...................................... (9)
Now that pump displacement is known, pump load may be cal- This criterion is the converse of Eq. 8, the stability criterion of
culated. Hooke's law (Eq. 4) is used, substituting a second-order- the predictive model.
correct backward difference for (juNk
Model Verification
Fpump,j = (EAI2.:1 x)(3u m ,j -4U m -l,j +U m -2,j)' ......... (7)
Using the finite-difference equations presented earlier, we devel-
Thus, equations have been generated for calculating pump dy- oped algorithms and programmed them in FORTRAN on a personal
namometer cards. In the model development, higher-order Taylor computer. Standards used to test the model were the finite-difference
series approximations could have been chosen for the derivatives. predictive model and the Fourier series (analytical) diagnostic
It is expected, however, that the additional accuracy would be small model. Schafer and Jennings 5 programmed these two models and
and would not compensate for the increased computational speed studied the parameters involved. These programs were used to verify
and supplementary storage requirements. the numerical diagnostic model developed here.
Stability Criterion. Because the model developed here is an ex- Predictive-Model Comparisons. The predictive model calculates
plicit finite-difference method, the stability of the solution is of sig- surface load from known surface position and pump load. The proce-
nificant importance and must be considered at this point. Gibbs 8 dure for comparison is as follows.
6000 6000
- - - Diagnostic Model
- - - - Predictive Model
4000
-iii
.a
::!. 2000
'0
S
0
- - Diagnostic Model
---- Predictive Model
-2OOO~----~------L-----~----~------...1 -2000
o 10 20 30 40 50 0 20 40 60 80 100
Position (In.) Position (In.)
Fig. 4-Comparl80n of numerical diagnostic and predictive Fig. 5-Comparl80n of numerical diagnostic and predictive
models for Case 1. models for Case 2.
Rod Data
Diameter length Elasticity
(in.) ~ Material (psi)
0.875 1,500 Steel 30.5x10 6
0.750 2,400 Steel 30.5x10 6
Downhole Data
Plunger diameter, in. 2.25 Pump depth, ft 3,900
Pumping speed, strokes/min 8 Fluid level, ft 2,850
Fluid specific gravity 0.92
Pumping-Unit Data
Unit designation M-228-256-120 Stroke length, in. 121
Manufacturer lufkin Rotation Counterclockwise
Rod Data
Diameter, in. 0.75 Material Steel
length, ft 3, 179 Elasticity, psi 30.5x 10 6
Downhole Data
Plunger diameter, in. 1.5 Pump depth, ft 3,234
Pumping speed, strokes/min 10 Fluid level, ft 3,234
Fluid specific gravity 0.8
Pumping-Unit Data
Unit designation C-114-143-64 Stroke length, in. 64
Manufacturer Lufkin Rotation Counterclockwise
1. Generate surface cards from known downhole conditions with diagnostic models. Figs. 6 and 7 show the recorded surface dy-
the predictive model. namometer cards for these wells, and Figs. 8 and 9 compare the
2. Calculate pump cards with the numerical diagnostic model by pump cards calculated with the two diagnostic models. These cases
use of the surface cards generated in Step 1. illustrate good agreement between the two models.
3. Compare pump cards calculated with the numerical diagnos-
tic model to actual pump cards in the predictive model. Numerical vs. Analytical. After good agreement between the nu-
Tables 1 and 2 are examples of artificial data sets used in these merical and analytical diagnostic models is shown, the question
comparisons. Figs. 2 and 3 show the surface cards generated with arises as to which model is more accurate. A description of the
the predictive model for these data sets. Figs. 4 and 5 compare procedure used to answer this question follows.
the predictive-model and diagnostic-model pump cards. These ex- 1. Generate surface dynamometer cards with the predictive model.
amples demonstrate the diagnostic model's ability to reproduce the 2. Calculate pump dynamometer cards from these surface cards
actual pump cards from surface information. with the numerical diagnostic model.
3. Calculate pump cards from the surface cards with the analyti-
Analytical-Model Comparisons. The Fourier series (analytical) cal diagnostic model.
model is probably the most well-known means of calculating down- 4. Plot the pump cards calculated with both diagnostic models
hole dynamometer cards at this time. Tables 3 and 4 give exam- on the same graph with the actual pump card from the predictive
ples of actual wells used to compare the numerical and analytical model.
15000
_ 10000 _ 6000
£ !
~ 5000
i
-I 4000
Fig. 6-Surface dynamometer card for Case 3. Fig. 7-Surface dynamometer card for Case 4.
2000
- 5000
-
~
iii
.1:1
do
3 2500
"C
.9
1000
-1000
-25OO0~----2~5----~50~--~7~5~---1~OO~--~125 0 20 40 80
Fig. 8-Comparlson of numerical and analytical diagnostic Fig. 9-Comparlson of numerical and analytical diagnostic
models for Case 3. models for Case 4.
5. The calculated pump card that matches the pump card from 7. If the difference between the pump horsepower and the hy-
the predictive model more closely is the more accurate model. draulic horsepower is within an acceptable tolerance, the proce-
Data sets described in Tables 1 and 2 are examples used in per- dure is complete. Otherwise, adjust the damping coefficient
forming this analysis. The generated surface cards are shown in accordingly and repeat Steps 5 and 6.
Figs. 2 and 3. Figs. 10 and 11 illustrate the results of the compari- Fig. 12 is a flow chart illustrating the damping calculation proce-
sons. In each case, lO to 15 Fourier coefficients were used in the dure. Running many cases has shown that the damping coefficient
analytical model, as suggested by Schafer and Jennings. 5 The an- converges within at most four to five iterations.
alytical model tends to round the comers of the pump card. This To verify the damping calculation procedure, the predictive model
is explained by the difficulty in approximating a square wave Ooad was used to generate surface cards with a damping coefficient of
at the pump) with a Fourier series. unity. Pump dynamometer cards were calculated, allowing the nu-
merical diagnostic model to compute its own damping coefficient.
Damping Calculation Pump cards from both models were then compared to verify the
damping calculation. Fig. 13 is an example of this comparison using
Past methods of sucker-rod analysis lacked a consistent means of
data listed in Table 2. For this example, the damping coefficient
simulating the damping forces inherent in sucker-rod pumping sys-
calculated in three iterations is within 2 % of actual.
tems. Nicol 6 presented a workable method of calculating damp-
ing, but it requires knowledge of fluid viscosity.
Conclusions
The intent of this section is to define a viscous damping coeffi-
cient, c, that may be used in Eq. 2 to model these damping forces. 1. The finite-difference diagnostic technique is an excellent means
The damping coefficient used is similar to that presented by of analyzing the behavior of sucker-rod systems.
Gibbs 9 : 2. The equations presented are valid for tapered-rod strings, rod
strings with sinker bars, and steel and fiberglass rods.
(550)(144g c ) (HpR -HH)7 2 3. The stability criterion for the fmite-difference diagnostic model
c , ................... (10) is the converse of the stability criterion for the finite-difference
.../21C' (Ep j A j L j )S2
predictive model.
where HH=7.36xlO- 6 q-yZ. 4. The viscous damping term associated with the wave equation
may' be computed with knowledge of the fluid level.
Eq. lO is valid for any rod material, whereas Gibbs' damping 5. The finite-difference solution better approximates square-wave
coefficient is valid only for steel rods. Ref. 4 gives a complete deri- pump loads than the truncated Fourier series solution (10 to 15 Fou-
vation of Eq. lO. Calculation of hydraulic horsepower, HH, re-
rier coefficients).
quires knowledge of the fluid level, which mayor may not be
known, and of pump production rate, which is dependent on the
Nomenclature
net pump stroke. This net pump stroke may be obtained from a
pump dynamometer card; however, the damping coefficient must A =. rod cross-sectional area, in. 2
be known before the pump card can be calculated accurately. This c = damping coefficient, seconds- 1
predicament leads to an iterative procedure for calculating the cor- E = Young's modulus of elasticity, psi
rect damping coefficient. F PR = polished-rod load, lbf
1. Calculate the damping coefficient with a fraction of the Fpump = pump load, lbf
polished-rod stroke as a first guess for the net pump stroke. gc = units conversion factor, 32.20bm-ft)/Obf-sec 2 )
2. Calculate the pump dynamometer card and determine the net HH = hydraulic horsepower, hp
pump stroke from this pump card. HPR = polished-rod horsepower, hp
3. Recalculate the damping coefficient, calculate a new pump L j = length of individual rod section, ft
card, and again determine the net pump stroke. q = pump production rate, BID
Experience has shown that the net pump stroke converges with-
S = polished-rod stroke, ft
in an acceptable tolerance in two iterations. The next task is iterat-
ing until the damping coefficient converges. Sn = net pump stroke, in.
4. Recalculate the damping coefficient with the now-fixed net t = time, seconds
pump stroke. u = rod deformation (displacement), ft
5. Calculate the pump dynamometer card with the new damping v = velocity of force propagation in rods, ftlsec
coefficient. x = axial distance along the rod string, ft
6. Determine the pump horsepower by computing the area of the Z = fluid level or net lift, ft
pump card. -y = specific gravity of fluid, fraction
SPE Production Engineering, February 1992 125
- - Numerical Diagnostic - - Numerical Diagnostic
6000 ........ Analytical Diagnoatlc
...•.... Analytical Diagnostic 6000
- - - Predictive - - - Predictive
...... .. '"
4000
4000
';i
.Q
do ';i
.Q
'1:J 2000 do
3
~ .
'1:J
~
0
2000
0
0 ............. .... .
'
-2000
0 10 20 30 40 50 -2000
0 20 40 60 80 100
Position (In.)
Position (In.)
Fig. 10-Accuracy of numerical and analytical diagnostic Fig. 11-Accuracy of numerical and analytical diagnostic
models for Case 1. models for Case 2.
4000
';i
.Q
do
'1:J 2000
C
HPPump )
= c· - - ~
( HPhyd
0
No
-2000
0 20 40 60 80 100
Position (In.)
aUI
2 (aulax)ii+ 'h,j- (aulax)ii_ Y2,j
............ (A-3)
ax 2 I,j
.. Ilx 1 (EA)+ 1 (EA)-
2pA
~_(EA)+ -(EA)-JUi'j'
in rod diameter and rod material. However, Eq. A-6 is not the fi-
nal form of the difference analog to be used for the space deriva- -[(2+ellt) Ilx
tive. Another preferred aspect of the finite-difference model is the Ilt 2 I44gc Ilx Ilx
ability to allow for a variable Ilx.
With tapered-rod strings, the solutions at the rod tapers are often
desirable so that maximum rod stresses may be analyzed. A constant-
+( Ilx ~)u.
Ilt2 I44g
c
'_I_(EA)
I,j Ilx
-u'_1 '.......... (A-lO)
1 ,j
Ilx scheme would require interpolation at the rod tapers, whereas
a variable-Ilx method would permit a different Ilx for each rod Dividing through by (EAlllx) + and taking an average for
size, providing exact solutions at the rod tapers. Thus, rewriting (pA/144g c) gives the final form of the main working equation:
Eq. A-6 to account for a variable Ilx yields
ui+l,j = ([a(I +ellt)]ui,j+1 -[a(2+ellt)-(EAlllx) +
__
2
a U I I
= ___ u'+1
1 ,J,-u,I,j. ____ I u·I,j,-u'_1
1 ,j.
-(EAlllx) - ]Ui,j+aui,j_1 -(EAlllx) - Ui-I,j }/(EAlllx) + ,
ax 2 I,j
.. (Ilx) + Ilx (Ilx) - Ilx ' ................................... (A-II)
................................... (A-7)
where 1lx=[(llx) + +(llx)-]/2.
Ilx [(PA/144g c) + +(pA/144g c) -
where a = - -
J.
Eq. A-7 is the final form of the finite-difference analog for the Ilt 2 2
seconq derivative with respect to distance. Eq. A-II is identical to Eq. 3, the equation used to transmit the
The next step is to substitute Eqs. A-I, A-2, and A-7 into Eq. 2, surface position downhole by calculating displacements at each node
along the rod string.
=Ilx1 (EA)
-Ilx + ui+l,j= (PA + ePA) Ui,j+1
144gcllt2 144gcllt
July 5, 1990. Paper accepted for publication March 15, 1991. Paper (SPE 18189) first
presented at the 1988 SPE Annual Technical Conference and Exhibition held in Houston,
Del. 2-5.