6th IFAC Conference on Management and Control of Production

and Logistics
The International Federation of Automatic Control
September 11-13, 2013. Fortaleza, Brazil

Sucker-Rod Pumping System of Oil Wells:

Modelling, Identification and Process
Control
Luiz H. S. Torres ∗ Leizer Schntman ∗∗
∗
ÁREA1 - Faculdade de Ciência e Tecnologia, Av. Paralela, 3172,
Salvador, Bahia, Brazil (Tel: +55-71-2106-3911; e-mail:
[email protected]).
∗∗
Universidade Federal da Bahia, Centro de Capacitação Tecnológica,
em Automação Industrial (CTAI), Rua Aristides Novis, n 02, Escola
Politécnica, segundo andar, 40.210-630, Salvador, Bahia, Brazil (Tel:
+55-71-3283-9755; e-mail: [email protected])

Abstract: Several studies have shown that the satisfactory oil well operations with sucker-rod
pumps is attributed to the techniques and methods which are able to control the performance
of the well. However, the presence, for example, of uncertainties in the sucker-rod dynamical
models or unmodelled dynamics, related to fluid characteristics in the well or associated with the
mechanical assembly, may jeopardize the desired performance of the control system and the rod
pump system productivity. The purpose of this paper is related to two main objectives. First,
a mathematical dynamic model of a real sucker-rod pump is developed based on identification
techniques by using input-output measured data. Second, an robust adaptive controller is used to
deal with some uncertainties in the system model, unmodelled dynamics, parameter variations,
and the presence of perturbations in the automatic level control system of the annular well.
The results obtained have shown that the developed model is representative and the adaptive
controller is able to deal satisfactorily with uncertainties and variations, such as for instance,
in the pumped fluid composition (water, oil and/or gas), a very common situation in real
production fields. Experimental tests were conducted in a real plant to check the validate and
robustness of the used controller algorithm. The results are also compared with a conventional
PID controller.

1. INTRODUCTION increases the possibilities of a successful implementation

of a control system, for instance. However, such simula-
The sucker-rod pump system is the artificial lift method tors, generally speaking, are limited to theoretical models
most used in the current on-shore petroleum industry (mostly phenomenological) and therefore lacking of evi-
due to the simplicity of its equipments and facilities Lake dences in situations close to the real production fields.
[2006]. This method is also considered as the first technique On the other hand, in terms of process control of the
used to lift oil up from wells. Studies have shown that pump unit, the presence of uncertainties in parameters,
its popularity is related to low cost of investments and parameter variations, unmodelled dynamics, and pertur-
maintenance, deep and outflow flexibility, good energy bations are challenges to the control system and they can
efficiency and the possibility for operating with different jeopardize the good performance of a conventional PID
fluid compositions and viscosities in a wide range of controller.
temperatures Takács [2002]. This paper is organized as follow: section 2 presents a
Although this lift method is already well-known and widely typical sucker-rod pump system and some operational
used, there are still some circumstances in which improve- aspects of this artificial oil lift method. In section 3
ments of operational conditions are still possible, especially presents the application of the modelling and identification
when dealing with process control strategies of the pump tools over a real sucker-rod pump system. The section 4
unit for increasing the system productivity. The develop- is devoted to a brief description about the algorithm of
ment of low cost sensors turned possible the measurement the adaptive controller chosen. The results obtained with
of bottomhole variables that assists the monitoring of the the adaptive controller and its performance are shown in
production, application of new control strategies, and to section 5 and section 6 presents some conclusions.
enhance of the process automation Moisés et al. [2008],
Smith et al. [2008], Bezerra et al. [2009]. 2. SUCKER-ROD PUMP SYSTEM

Simulation softwares have assisted the studies about the In this artificial lift method a rotary movement of prime
sucker-rod pump system in modelling and design of control mover (either an electric or a combustion motor) localized
systems, especially into controlling the dynamic fluid level on the surface of the pump unit is converted in alternative
of the annular well. The validation of these softwares movement of the rod string. This same column transmits

978-3-902823-50-2/2013 © IFAC 260 10.3182/20130911-3-BR-3021.00052

IFAC MCPL 2013
September 11-13, 2013. Fortaleza, Brazil

the an alternative movement to the pump components that possible annular level (minimum bottomhole pressure) the
are located at the bottom of the well, that are responsible reservoir oil outflow is maximized Ordoñez et al. [2009].
to elevate the fluid from reservoir up to the surface. The In therms of a control system design to increase the oil
sucker-rod pump system could be divided in downhole and production, a dynamic model of a sucker-rod pump system
surface elements (see Fig.(1)). may relate the pumping speed of the unit and the fluid
level in the annular well. This dynamic model may also
reveal the relationship of its parameters with the real
process. According to the literature Barreto Filho [2001],
Ordoñez et al. [2009], these parameters are normally
related to fluid characteristics in the well, environmental
properties at the bottom of the hole, and mechanical
assembly, for example. However, there are negative effects
over the pump unit controller caused by uncertainties in
parameters, parameter variations, unmodelled dynamics,
and perturbations. They are normally related to fluid
characteristics in the well, environmental properties at the
bottom of the hole, electrical components and mechanical
assembly, for example.

3. MODELLING AND IDENTIFICATION OF A

SUCKER-ROD PUMP SYSTEM

In systems that are assembled with sucker-rod pumps,

Fig. 1. Components of a sucker-rod pump system. often it is desired that an operation range is very close to
the pump inlet level. This operation range is characterized
The rod string is the link between the pump unit localized by the complete pump filling with the least bottomhole
on the surface and the bottomhole pump. The bottomhole pressure possible. That provides the minimum back pres-
pump is a kind of alternative pump of positive displace- sure on the production zone of the reservoir and, in turn,
ment of simple effect, in other words, the fluid is displaced it increases the oil production Ordoñez et al. [2009]. In
in a one way direction of the alternative movement. The the Laboratório de Elevação Artificial - LEA (in english
function of the bottomhole pump is providing energy to the Artificial Lift Lab) at the Universidade Federal da Bahia
fluid from reservoir Lake [2006]. In Fig.(2) the bottomhole (UFBA), there is a real plant of a sucker-rod pump with
scheme is presented. The annular well and pump inlet level an artificial well of 32m of height, fully instrumented,
are also shown. with full access and visible bottomhole. All components
of this equipment are industrial and the plant also has
a supervisory system to data acquisition and control. A
scheme of this well is presented in Fig(3).

Fig. 2. Bottomhole scheme with sucker-rod pump system.

The pumping cycle generated by the relative movement of

the valves have repercussions on the bottomhole pressure.
Fig. 3. Well scheme with the sucker-rod pump system at
The oil production is controlled by varying the prime
the UFBA.
mover velocity, which implies in the manipulation of the
pumping speed, measured in cycles per minute (CPM). In modelling and identification procedures it often used
In this control strategy the variable speed drive (VSD) a deviation variables (input and output signals of the
technique is used. That allows to adjust the pumping process) from a reference or a nominal operation point. In
speed through a frequency inverter device Peterson et al. this work, a reference is chosen by considering the desired
[2006], Ordoñez et al. [2008]. It is important to remark operation range and the pump inlet level. A scheme of this
that the production performance is associated with the well with the reference chosen and this operation range
annular fluid level, and the operation with the minimum is also shown in Fig(3). It is important to remark that

261
IFAC MCPL 2013
September 11-13, 2013. Fortaleza, Brazil

the production performance is associated with the annular

fluid level. Therefore a dynamic model of a sucker-rod
pump may be developed by relating the pumping speed
of the unit and the fluid level in the annular well in order
to assist the system monitoring or for control purposes, for
instance. Hence in this model the level in the annular well
(measured in meters) is considered as the process output
y(k), and the pumping speed (measured in CPM) as the
process input u(k).
In Barreto Filho [2001] some relationships between the
pumping speed and the fluid level in the annular well
(and other possible variables and parameters of the whole
system) based on theoretical studies and phenomenological
models of a sucker-rod pump are presented. However, these
studies indicate a flow rate from the annular well to the
production tubing is linear in the head (level), which would
indicate laminar flow by considering a single-phase and Fig. 5. Comparison between the estimated output and the
incompressible fluid. In order to verify this property in the process output.
real plant used in this work some tests are performed and
a step signal is applied to the process input. The process process response. However, it can be seen in some sections
output is shown in Fig.(4). in Fig. (5) there are differences between the estimated
output and process output due to unmodelled dynamics,
for example. Despite of these differences, the obtained
model may meet the main expectations in a control system
design. In other words, the model incorporates the main
features of actual system dynamics and it is sufficiently
representative.

4. ROBUST ADAPTIVE CONTROL TECHNIQUE

In systems that are assembled with sucker-rod pumps,

often it is desired that an operation range is very close to
the pump inlet level. This operation range is characterized
by the complete pump filling with the least bottomhole
pressure possible. That provides the minimum back pres-
sure on the production zone of the reservoir and, in turn,
it increases the oil production Ordoñez et al. [2009].
In this work a robust adaptive controller (called IVS-
Fig. 4. Process output yielded by a step input. MRAC) is applied to a sucker-rod pump. It could be also
desired that the controller operates inside the range. More-
By observing the Fig.(4), the process response with no
over, the controller must be able to adapt the parameters
dead time and it seems that can be well fitted by a linear
and show robustness in case of process changes, uncertain-
first order model. This sounds reasonable because this
ties, variations, unmodelled dynamics, and perturbations
process can be seen as a simplified fluid level system in a
in the system.
SISO tank Ogata [2000]. From Aguirre [2007] a structure
of the ARX model may be proposed as following, The IVS-MRAC was initially proposed in Oliveira and
B(q −1 ) 1 Araújo [2002], and appears as an alternative project to
y(k) = −1
u(k − 1) + e(k) , (1) the direct approach called VS-MRAC Hsu and Costa
A(q ) A(q −1 )
[1989]. The mathematical description, stability analysis
A(q −1 ) = 1 + a1 q −1 , and its proof in presence of unmodelled dynamics and
(2)
B(q −1 ) = b0 . perturbation, for the case of relative degree one, can be
found in Oliveira and Araújo [2008]. A simplified version
A recursive least square (RLS) is used to estimate the with practical application of the IVS-MRAC to the speed
parameters. The results are shown as follow. A comparison control of a three-phase induction can be found in Oliveira
between the estimated output and the process output is et al. [2010].
presented in Fig.(5).
The base that this technique was developed is the indi-
a1 = −9, 307 × 10−1 rect MRAC (called I-MRAC). The classical I-MRAC is
(3)
b0 = 3, 500 × 10−3 , one of main techniques in adaptive control Astrom and
A(q −1 ) = 1 − 9, 307 × 10−1 q −1 Wittenmark [2008], Ioannou and Sun [1996]. The desired
(4)
B(q −1 ) = 3, 500 × 10−3 . response is provided by a reference model and the control
objective is to minimize the error between the system
An analysis of the results and the comparison presented under control and a model reference output. However, in
in Fig (5) show the identified model clearly follows the the IVS-MRAC there are some new features in relation to

262
IFAC MCPL 2013
September 11-13, 2013. Fortaleza, Brazil

I-MRAC. By using the sliding mode control based on VSC pumping speed (measured in cycles per minute - CPM) as
(Variable Structure Control) Utkin [1978], the I-MRAC is the manipulated variable (MV).
associated with faster transitory and robustness to the pa-
rameters uncertainties, variations, and perturbations. The The plant parameters kp and α1 used in the tests are with
IVS-MRAC could estimate the plant parameters instead 10% of uncertainties around the model reference parame-
of using the controller parameters. ters km and αm,1 . The initial conditions of the plant and
the reference model were different to facilitate the observa-
This adaptive controller provides a straightforward design tion of the tracking properties. The tests were performed
for the relays amplitudes used in switching laws of the regarding the reference input r, a set of step signals. It was
controller algorithm Oliveira et al. [2010], since these re- adopted kpnom = 3, 522 × 10−5 and k p = 7, 044 × 10−7 . The
lays will be directly associated with the plant parameters. value of α1 is chosen from the value |α1 | = 6, 973 × 10−4 .
These parameters, in turn, represent the relationships It was adopted 10% plus of uncertainty and the value was
among the physical parameters of the system such as the α1 = 7, 671×10−4 . The time constant τ was adjusted along
resistances, capacitances, moments of inertia, friction co- the tests. It was adopted τ = 0, 01. In Fig.(7) the compared
efficients, etc., that have more easily known uncertainties. response between the reference model output and the
The block diagram shown in Fig.(6) may illustrate the process output is presented. In Fig.(8) and Fig.(9) show
general idea of using the IVS-MRAC. the error signal (in meters) and the control effort (related
to the variation of the CPM - in percent), respectively.

Fig. 6. The general idea of the IVS-MRAC.

The model plant P (θp ) is parametrized in relation with Fig. 7. Comparison between the desired response (refer-
the vector θp . An estimator generates θp (t), by processing ence model signal) and process response.
the input signal u and the output signal y. The estimate
θp (t) specifies a model characterized by P̂ (θp (t)) that, for
the purposes of controller design, is treated as the true
model of the plant at the instant t. The latter is used
to calculate the controller parameters θc (t) by using the
algebraic equation θc (t) = f (θp (t)). The calculation of the
control law C(θ) and the equation θc = f (θp ) are made to
meet the performance requirements for the model P (θp ).
5. EXPERIMENTAL RESULTS

The proposed controller for the level control of the fluid

in the annular well of a sucker-rod pump system are
implemented as follow.
From Eq.(4) one has the continuos-time SISO transfer
functions of the plant dynamical model,
n (s) −5
3.522×10 Fig. 8. Error signal.
PlantP (s) = kp dpp(s) = s+6.973×10−4

⇒ kp = 3.522 × 10 ; α1 = 6.973 × 10−4

−5
It could be observed in Fig.(7) that the reference model
The reference model may be chosen as follow (desired response) output given was tracked by the process
output. The process variable (fluid level in the annular
−5
Reference Model M (s) = km ndm
m (s) 3.874×10
(s) = s+7.671×10−4
well) could be observed in Fig.(7) with no oscillations in
⇒ km = 3.874 × 10 ; αm,1 = 7.671 × 10−4
−5 transitory and stable in steady state. The error in steady
state is small and bounded according to Fig.(8) and the
In these models the level in the annular well (measured in control effort is also bounded in Fig.(9). However, Fig.(9)
meters) is considered as the process variable (PV), and the indicates an input usage for the proposed controller that

263
IFAC MCPL 2013
September 11-13, 2013. Fortaleza, Brazil

Fig. 9. Control effort. Fig. 11. Conventional PID with variation in the parameters
and perturbations.
in some cases would be considered unacceptable. Thus,
some strategies for alleviating the chattering phenomenon
Oliveira et al. [2010] should be further studied. is also increased by the fluid pound occurrence. The
A new set of tests is performed to compare the IVS- parameter variations introduced in kp and α1 can be
MRAC designed with a conventional PID controller. By related to changes in fluid composition (water, oil and/or
this time the objective is to evaluate the robustness and gas), a very common situation in real production fields.
adaptation properties of the controllers. It is introduced a The perturbations on PV may be associated to faults on
15% of variation in the parameters kp and α1 and a step the power supply of the frequency inverter. Perturbations
perturbation of 20% on the MV and PV variables. The on MV may be related to the leaking in the travelling
results is shown in Fig.(10) and Fig.(11). and/or standing valves Takács [2002].

6. CONCLUSIONS

In this paper a modelling and identification tools and a ro-

bust adaptive controller (called IVS-MRAC) were applied
to a sucker-rod system of oil wells. It should be noticed that
one of main challenges that has been repeatedly found is
the validation of models (mostly by using material balance,
energy, etc.) for these systems. Since the components of
pumping, such as, rod string and bottomhole pump, are
located in the deep subsurface and without the possibility
of easy access. It does not happen with the plant used in
this work, whose system is fully instrumented and the bot-
tom is accessible and visible. A contribution of this paper is
that the results were not restricted to theoretical models
Fig. 10. IVS-MRAC with variation in the parameters and (or simulators) and therefore in lacking of evidences in
perturbations. situations close to the real production fields. The support
of the LEA laboratory may confirm the ARX model devel-
It could be seen in these new tests that the references oped within the desired operation range is representative
are tracked by the controllers. However, in the control since incorporates the main features of the dynamics of
system with conventional PID (Fig.(11)), the parameter real plant. About the controller performance, it could be
variations and perturbations cause a performance loss observed that the desired response given by the reference
deviating the process output in relation to the reference model was tracked by the process response. The error
and the desired operation range. In Fig.(10) there is no signal could be seen bounded and small, and the control
performance loss and the IVS-MRAC shows robustness. It effort could be seen also bounded, though alleviating the
is important to remark that outside the operation range chattering should be further studied. The results also show
the oil production is lower. This deviation in the process that the adaptive controller is able to control satisfactorily
output can also cause a fluid pound Ordoñez et al. [2009]. the fluid level in the annular well, in spite of the presence
The fluid pound phenomenon occurs due to entry of air of model uncertainties, unmodelled dynamics, parameter
in bottomhole pump, that is caused when the fluid level variations, and perturbations. Moreover, the results reveal
is below the pump inlet level. This phenomenon has a that the control technique could increase the production
negative effect on the oil production and the maintenance performance and diminishes the maintenance costs (for
costs. The mechanical fatigue of the pump components example, by avoiding the fluid pound).

264
IFAC MCPL 2013
September 11-13, 2013. Fortaleza, Brazil

ACKNOWLEDGEMENTS a improved srp control. In Proceedings of SPE An-

nual Technical Conference and Exhibition, San Antonio,
The research was supported by the CTAI (facilities and USA, 2006. ISBN: 978-1-55563-149-9.
infrastructure) at the Universidade Federal da Bahia and B. Smith, M. Hall, A. Franklin, E. S. Johansen, nalmis, and
CAPES (financial support). H. Field-wide deployment of in-well optical flowmeters
and pressure/temperature gauges at buzzard field. In
REFERENCES Proceedings of SPE Intelligent Energy Conference and
Exhibition, Amsterdam, Netherlands, 2008.
L. A. Aguirre. Introduction to System Identification - G. Takács. Sucker-Rod Pumping Manual. PennWell
Linear and Non-Linear Techniques Applied to Real Sys- Books, Tulsa, USA, 2002.
tems. (in portuguese). Editora UFMG, Belo Horizonte, V. I. Utkin. Sliding Modes and Their Application in
Brazil, 2007. Variable Structure Systems. MIR, Moscow, Russia,
K. Astrom and B. Wittenmark. Adaptive Control. Dover, 1978.
New York, USA, 2008.
M. de A. Barreto Filho. Estimation of avarage reser-
voir pressure and completion skin factors of wells that
produce using sucker rod pumping. PhD thesis, The
University of Texas at Austin, Austin, USA, 2001.
M. A. D. Bezerra, L. Schnitman, M. A. Barreto Filho,
and J. A. M. Felippe de Souza. Pattern recognition
for downhole dynamometer card in oil rod pump sys-
tem using artificial neural networks. In Proceedings of
11th International Conference on Enterprise Informa-
tion Systems, pages 351–355, Milan, Italy, 2009.
L. Hsu and R. R. Costa. Variable structure model
reference adaptive control using only input and output
measurements part i. International Journal of Control,
49(2):399–416, 1989.
P. A. Ioannou and J. Sun. Robust Adaptive Control.
Prentice-Hall, New Jersey, USA, 1996.
W. L. Lake. Petroleum Engineering Handbook, volume 04.
Society of Petroleum Engineers, Richardson, New York,
USA, 2006.
G. V. Moisés, T. A. Rolim, and J. M. Formigli. Gedig:
Petrobras corporate program for digital integrated field
management. In Proceedings of SPE Intelligent Energy
Conference and Exhibition, Amsterdam, Netherlands,
2008. ISBN: 978-1-55563-166-6.
K. Ogata. Modern Control Engineering. Prentice Hall,
New Jersey, USA, 2000.
J. B. de Oliveira and A. D. Araújo. Indirect variable
structure model reference adaptive control: Indirect vs-
mrac. In in Proceedings of XIV Brazilian Control
Conference (CBA), pages 2557–2562, 2002.
J. B. de Oliveira and A. D. Araújo. Design and stability
analysis of an indirect variable structure model reference
adaptive control. International Journal of Control, 81
(12):1870–1877, 2008.
J. B. de Oliveira, A. D. Araújo, and S. M. Dias. Controlling
the speed of a three-phase induction motor using a sim-
plified indirect adaptive sliding mode scheme. Control
Engineering Practice, 18(6):577–584, June 2010.
B. Ordoñez, A. Codas, and U. F. Moreno. Sucker-rod
pumping system: Simulator and dynamic level control
using downhole pressure. In Proceedings of 13th IEEE
International Conference on Emerging Technologies and
Factory Automation, Hamburg, Germany, 2008.
B. Ordoñez, A. Codas, and U. F. Moreno. Improving
the operational conditions for the sucker-rod pumping
system. In Proceedings of IEEE International Confer-
ence on Control Applications, pages 1259–1264, Saint
Petersburg, Russia, 2009.
R. Peterson, T. Smigura, C. Brunings, W. Quijada, and
A. Gomez. A production increases at pdvsa using

265