SPE 123684

Fuzzy Analysis of ESP System Performance

Donald G. Thornhill and David Zhu, Baker Hughes Incorporated

Copyright 2009, Society of Petroleum Engineers

This paper was prepared for presentation at the 2009 SPE Annual Technical Conference and Exhibition held in New Orleans, Louisiana, USA, 4–7 October 2009.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents of the paper have not been
reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect any position of the Society of Petroleum Engineers, its
officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written consent of the Society of Petroleum Engineers is prohibited. Permission to
reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract
The purpose of this paper is to illustrate how Fuzzy Logic, along with some expert knowledge of how an electrical submersi-
ble pumping system (ESP) performs, can be used to provide near real-time diagnostics using a limited number of signals typ-
ically acquired from an ESP system. Common problems in a running ESP system can be diagnosed by examining patterns in
the measured data compared to values associated with a system that is operating normally. Using Fuzzy Logic, patterns asso-
ciated with specific problems can be detected long before they become obvious. This information can be used to respond to
emerging conditions before they result in downtime, or can identify why a system is not performing as expected. This analy-
sis is applicable to ESP systems operating in a stable state, where normal measurement ranges can be determined. It is not
intended to diagnose problems during system startup or when outside factors are dynamically altering well performance.

Introduction
An electrical submersible pumping system (ESP) is designed to produce fluid at a given rate and pressure at the surface.
When a pump does not produce as expected, it is important to quickly identify the true nature of the problem in order to:
• restore production
• prevent damage to the equipment or wellbore
• avoid or minimize costly down-time

When an ESP is started, the fluid level in the casing drops until it reaches equilibrium with the pressure supplied by the
reservoir. After this, the pump will perform as designed and the pressure at the pump intake will remain relatively constant
unless something in the system changes. For example, an obstruction forming in the pump, tubing or at the surface will in-
crease the pressure seen by the pump. This changes the operating point of the pump resulting in a reduced flow rate. An in-
crease in pump intake pressure is enough information to reveal that something has changed, but what? A gradual change may
indicate that the well’s productivity index (PI) has changed, but more rapid change may point to a problem.

An increase in pump intake pressure is not enough information to conclude that an obstruction exists. The same symptom
could be caused by many things including an increase in pump speed, a tubing leak, a closed or partially closed surface valve,
or a change in fluid properties. More information is required in order to identify a specific problem. An ESP system that sup-
plies measurements including pump intake pressure, pump discharge pressure, surface pressure (before any surface chokes),
motor current and temperature, and flow rate is optimal for diagnostic analysis using fuzzy logic as presented in this article.

Fuzzy Logic
In simple logic, every proposition takes exactly one of two truth values: simply true or simply false – there is no middle
ground, no partial truth. Fuzzy logic is a method of handling inherently imprecise concepts. For example, the concept of
"coldness" cannot be expressed in a simple equation, because although temperature is a quantity, "coldness" is not. Nonethe-
less, people have an idea of what "cold" is, and agree that there is no sharp cutoff between "cold" and "not cold," where
something is "cold" at N degrees but "not cold" at N+1 degrees. 1 The result has no clearly defined answer so it is believed to
be a "fuzzy" answer.

Consider a measuring cup filled halfway. Is the cup full? Is it empty? Using simple logic - neither is true. Using fuzzy
logic, you could say the cup is half-full or, said another way, the cup is full with a 50% degree of truth. Likewise, it could
2 SPE 123684

also be considered half-not-full, or empty with a 50% degree of truth. Using fuzzy logic, the truth of anything becomes a mat-
ter of degree. 2

For ESP system performance, the question could be, “Is the pump performing as expected?” Answering this question is
not hard if you know what to expect. If an ESP is producing the expected amount of fluid for a given wellhead pressure, the
answer is yes. But what if the fluid produced is 2% less than originally thought? Using baseball terminology, the question
becomes, “Is it in the ball park?” For oil applications, the answer would be yes, with an acceptable degree of truthfulness.
Ball park estimations are an inherently fuzzy concept, but they do have boundaries. For oil applications, ±10% may define
the ball park. Because water applications do not have to deal with the effects of gas or viscosity, it may unacceptable for pro-
duction to be off by more than even ±2%.

The result of applying fuzzy logic to diagnose ESP system performance should be considered a "possibility" rather than a
"probability." If the data matches the pattern of a tubing leak with a 70% degree of truth, it is not saying that it is probably a
tubing leak. It says that it looks like a tubing leak. The numerical quantifier is a measure of the strength of similarity based on
how well the data matches the tubing leak pattern, but there may be similarities to other conditions of equal or greater
strength.

Expert Rules
There are many problems that can degrade ESP system performance. Understanding how each of these conditions affects
measured performance enables you to define rules that can be analyzed with fuzzy logic.

Table 1—Diagnostic Conditions

Condition Description
Main Valve is Closed The well is shut in at the surface
Main Valve is Open The surface choke has been opened
High Frequency The controller frequency has been increased
Low Frequency The controller frequency has been decreased
Pump-off The fluid level in the well is too low
Plugged tubing An obstruction is present in the tubing
Plugged pump An obstruction is present in the pump
Tubing Leak There is a hole in the tubing between the pump and surface
Gas Lock The pump is not producing due to gas locking
Pump Worn The pump is worn and no longer producing as expected

Each of these conditions results in a discrepancy in one or more of the measured signals compared to each signal’s ex-
pected value. For example, consider the “Main Valve is Closed” condition. It is not uncommon that an operator may forget to
open the main surface valve before starting an ESP system. In some cases, they will seem certain that they did so, but the data
will tell a different story. It may also be that the valve is only partially opened. So what pattern is associated with this scena-
rio?

• Flow rate is low, because of the obstruction (the valve is closed).

• Intake Pressure is high, because the flow is restricted.
• Surface Pressure (Tubing pressure before the choke) is high, because the pump produces higher pressure at lower
flow rates. (See Fig. 1.)
• Motor Amps is low, because low flow generally results in lower power requirements. (See Fig. 1.)
• Motor Temperature is high, because the cooling effect of fluid flowing past the motor is reduced.
SPE 123684 3

Fig. 1— Pump Performance Curve.

Notice that the language used is vague: "Low" and "High" or even "Normal" do not convey specific values. This is be-
cause these values change from one application to another. However, the rules still apply because they are based solely on
patterns in the data.

Normal Value Ranges

In order to properly detect abnormal conditions using measured signals, the first step is to identify the value of each signal
when the system is operating normally. These can be identified from a graph of historical data. During startup, or when a well
is undergoing treatment, the values will be changing. Once the system reaches a stable state, the traces on the graph will be
relatively flat. A normal value for each signal should be identified from the flat portion of the graph (see Fig. 2).

Fig. 2—“Stable” Operating Conditions.

4 SPE 123684

The second step is to identify the limits of normal operation for each signal. This can be done with a simple rule of
thumb, by adding or subtracting a percentage from the normal value, or by taking into account variations in the data during
stable operation. The purpose of this step is to identify the limits of normal operation. This combination of the normal value
ranges for each signal defines the “ball park” for normal operation, and provides the limits used to detect problems that may
be developing.

Degree of Truth
The result of identifying the normal value range is a membership function as illustrated in Fig. 3 for pump intake pressure.
Based on this membership function, you can calculate a numerical degree of truth for IsHigh or IsLow.

Fuzzy Logic Functions for Pump Intake Pressure

0.75

0.5

0.25

0
0 50 100 150 200
Pump Intake Pressure

IsLow IsHigh

Fig. 3—Fuzzy Logic Functions.

For example, let's say that the "normal" range of pump intake pressure is set as follows:
• Low = 50 psi
• Normal = 100 psi
• High = 150 psi

If the measured value is 40 psi, then the intake pressure is definitely low (IsLow=1.0). It is also definitely not high
(IsHigh = 0.0). If the measured value is 200 psi, then the intake pressure is definitely high (IsHigh=1.0). It is also definitely
not low (IsLow=0.0).

What if the measured reading is 75 psi? With fuzzy logic any value in the normal range has an IsHigh and IsLow compo-
nent and the degree of truth is quantified by a number between 0.0 and 1.0. If the measured value is 75 psi, IsHigh=0.75 and
IsLow = 0.25. Note that IsHigh = 1.0 – IsLow.

Example #1 - Calculations
Let’s take this a step further and apply fuzzy logic to the possibility that a “Main Valve is Closed.” First, we have to define
normal value ranges for each signal. Then we’ll take a measured reading and compute the degree of truth using fuzzy logic.
We’ll apply the IsHigh function to each value that is expected to increase when the valve is closed, and the IsLow function to
each value that is expected to decrease. This will give us a degree of truth for each component of the rule.
SPE 123684 5

Table 2—Main Valve is Closed

Pn Signal RMin RNormal RMax Measured Rule Pi
P1 Surface Pressure 80 100 120 110 IsHigh 0.75
P2 Pump Intake Pressure 50 100 150 120 IsHigh 0.70
P3 Pump Discharge Pressure 1350 1500 1650 1600 IsHigh 0.83
P4 Flow Rate 950 1000 1050 900 IsLow 1.00
P5 Motor Amps 30 40 50 35 IsLow 0.75
P6 Motor Temperature 180 200 220 215 IsHigh 0.86

The next step is to compose the results into a single number that represents the degree of truth for the condition. There is
more than one way to do this. In this example, the possibility of the “Main Valve is Closed” condition is considered valid if
all of the rules match with some degree of truth. Therefore, the result of applying the rules using fuzzy logic is the minimum
of all the factors. 3

⎛ R max − measured ⎞
IsLow = MIN ⎜ ,1.0 ⎟
⎝ R max − R min ⎠

IsHigh = 1.0 − IsLow

Pi is the result of the IsHigh/IsLow test for each signal

P = MIN(P1, P2, P3, P4, P5, P6) degree of truth for the rule

For the measured values in Table 2:

P = MIN(.75, .70, .83, 1.0, .75, .88) = .70

In other words, the entire set of rules share at least a .70 degree of truth for the “Main Valve is Closed” condition, with
IsHigh(Pump Intake Pressure) being the weakest match.

Missing Signals
As mentioned in the introduction, an ESP system that supplies measurements including pump intake pressure, pump dis-
charge pressure, surface pressure (before any surface chokes), motor current and temperature, and flow rate is optimal for
diagnosing problems with ESP systems. What if some of these signals are unavailable? In fact, most installed ESPs are not
equipped with all of the recommended instruments. The effect is that some conditions may not be differentiated from other
similar conditions. For instance, without discharge pressure, our expert rules do not distinguish between a closed surface
valve or a plugged tubing. If either of these conditions exist, the fuzzy logic will return the same degree of truth for both.

Example #2 – Real World

This example demonstrates that fuzzy logic provides useful diagnostics even when some signals are missing. An operator
came to us after an ESP system failure, seeking guidance on how to avoid similar problems in the future. The ESP was in-
stalled in a gas field to improve gas production by dewatering the well. Of the relevant signals used in the fuzzy logic rules,
the ESP system was equipped to monitor pump intake pressure, flow rate, motor temperature and motor amps. Even though
pump discharge pressure and surface pressure would have contributed to the degree of truth calculated for the problem, just
these four measurements—along with some fuzzy logic analysis—would have been sufficient. Early diagnosis would have
saved the operator heavy losses due to down-time, extended rig expenses, equipment replacement, and repair.

The first symptom was a drop in production. The operator responded by increasing the drive frequency from 54.5 to 60
hertz in order to try and boost production. This did not correct the problem. With 200+ wells to monitor, limited manpower,
inexperience with ESP operations, and no apparent reason for the drop in production, the operator continued to run the ESP.
The tubing eventually broke apart and the ESP dropped to the bottom of the well bore. This costly failure could have been
avoided with the help of fuzzy analysis of the supplied data. A tubing leak began to show up on the list of possible problems
40 days before the tubing broke. At first, the degree of truth was around 10%, but as the condition worsened, the value in-
creased steadily until it reached 61%.

Had the operator been alerted, the production engineer could have responded proactively by testing for a tubing leak and
then pulling the equipment before it ended up at the bottom of the well. Doing so would have avoided a costly retrieval and
downtime. As Fig. 4 shows, the operator had plenty of time to act, had they been alerted to the true nature of the problem via
fuzzy analysis.
6 SPE 123684

Fig. 4—Tubing Leak.

Diagnostic Alarming
SCADA systems support alarm notifications based on set points (limit-based alarms). For instance, an alarm may be trig-
gered if a measured value exceeds a set value. This is simple logic applied to data as it comes in. While this type of alarming
brings attention to an abnormal signal (the symptom), it may not provide enough information to diagnose the cause of the
problem (the disease). For example, a set point alarm could tell you that the Pump Intake Pressure has exceeded a predefined
limit, but is not enough information to suggest that the main valve is closed. Identification of the problem would still require
expert analysis.

The use of Fuzzy Logic does not put an end to the need for a real human expert, but it can be used to suggest possible
causes based on patterns in the data. This can significantly reduce the amount of time it takes to identify a problem, or to
detect an emerging condition. Using Fuzzy Logic, a pattern can be detected long before a set point is exceeded, or is visually
apparent from a graph of the data. This provides a framework for an early detection system that identifies possible problems
developing within the well or ESP.

A SCADA system that incorporates fuzzy logic and expert rules can provide real-time diagnostic alarms as the data be-
comes available. But it must be kept in mind that the rules as presented in this paper only apply during stable state conditions.
When the well is shut-in, or drawing down, or being treated, the alarm notifications could be misleading and should be sup-
pressed.

Conclusion
Fuzzy logic is a well established mathematical method of handling inherently imprecise concepts. We have presented fuzzy
logic as a way to quantify the relative highness or lowness of several measurements available on ESP systems equipped with
certain sensors. By applying fuzzy logic to expert knowledge of how an ESP system should perform, it is possible to identify
likely causes of problems associated with ESP systems. When incorporated into a SCADA system, fuzzy logic alarms enable
the operator to respond to adverse developments in a timely manner, thereby optimizing production, minimizing workover
costs and protecting the system run life.
SPE 123684 7

References
1
“Fuzzy Logic,” Economic Expert, www.economicexpert.com/a/Fuzzy:logic.htm.
2
“You fuzzin’ with me?” Surprise 96 Journal, www.doc.ic.ac.uk/~nd/surprise_96/journal/vol1/sbaa/article1.html.
3
fuzzyTECH-MP (DS30238A), Microchip Technology Inc. (1995) 19.