MODEL BASED CONTROL AND DIAGNOSTIC SYSTEM

FOR CENTRIFUGAL PUMPS

Filippo Attivissimo
Department of Electrical and Information Engineering (DEI), Polytechnic of Bari
Via E. Orabona 4, 70125 Bari, Italy
attivissimo@misure.poliba.it

Massimiliano. Di Febo, Pasquale Paganini and Giacinto di Mango

IPC Industrial Plants Consultants
Via delle Violette, 12,70026, Modugno (BA), Italy
[massimiliano.difebo, pasquale.paganini, giacinto.dimango]@ipc-eng.com
Abstract - Centrifugal pumps are widely used in industrial
applications. Today pumps absorb nearly a quarter of the
electric energy of the European industry. The purpose of this
paper is to present a centrifugal pump model-based control
system that ensures the energy benefits related to variable speed
drive technique and provides the complete protection monitoring
and machine diagnostic using a minimum number of installed
sensors.
Keywords: centrifugal pump; control system; monitoring;
diagnostic

I.

INTRODUCTION

The increasing demand for ordinary comfort, the attention

to environmental pollution and the reduction trend of fossil
fuels, require a more careful management of energy and push
the use of renewable energy [1], [2]; in this scenario, where
the energy usage rises year by year, also the industry should
provide some solutions in order to decrease the consumptions
and contribute to the environment care. Centrifugal pumps,
driven by medium-low motors (50 kW), represent almost all
of the installed machines in Oil & Gas process plants. In
addition to energy costs, also management, maintenance and
downtime costs of pumping systems have high impact on
overall plant costs balance. Despite of all , the idea of control,
protection and diagnostic systems, applied to medium-low
power pumps, is not widely applied yet. This kind of solution
requires external sensors for system and process parameters
acquisition, that might have considerable impact on initial
costs. Modern control systems are able to change the pump
rotational speed. The speed control is actually realized with a
frequency converter Variable Speed Drive (VSD) technique. A
possible strategy for improving outputs of production
processes and reducing overall costs, consists then in the
acquisition of
accurate, precise and reliable real-time
information about how the machinery actually works and
consequent generation of congruent control and protection
actions. These preliminary considerations introduce the
fundamental and central role of the control system
architecture. In the recent literature many fully sensor-less
control techniques are described [3] which determine a

978-1-4799-4988-5/14/$31.00 2014 IEEE

considerable advantage in terms of energy and

implementation costs. However these methods do not allow to
performing the control simultaneously to the continuous
monitoring and diagnostic of the machine. The system
proposed in this paper successfully estimates and controls the
operational state of the pumping system using a minimum
number of field transducers with a consequent reduction of
installation costs; moreover the utilization of a minimum
number of sensors and additional components, also reduces
the possibility of relative malfunctions. The proposed system
architecture also gives a continuous machine monitoring and
diagnostic by an advanced diagnostic algorithm. The
algorithm is based on the field data acquisition and on the
determination of the pump operative "expected" parameters
from the implemented pump model. The system monitoring
and diagnostic capabilities are aimed to increase the machine
availability and reliability with consequent benefits arising
from the reduction of downtimes and maintenance costs. The
process oriented control architecture also permits a reduction
of costs for process operators surveillance. The VFD
technique allows to obtain important energy consumption and
costs reduction. The proposed system obtains these benefits
thanks to the characterization model of the centrifugal pump
performance, as central element.
II. CONTROL TECHNIQUES: STATE-OF-THE-ART
A typical pumping system consists of many elements: a
driver (commonly an induction motor), a centrifugal pump,
valves and tanks. Advanced systems may include a VSD block
and sensors producing the process variable feedback (Fig. 1).

I. plant configuration
II. installation costs
III. energy saving
IV. simplicity
V. efficiency
VI. reduced risk of failure
VII. Precision

Requirements
Figure 1 Typical pumping system structure
s

In each instant of time, the overall proccess status can be

described by the group of all relevant prrocess parameters
(flow rate, liquid level, process pressures annd fluid properties
etc). In this group we can also find thee pump operative
parameters (pump operative status). The acctual value of the
process parameters depend from the cooupling of actual
characteristic of the pump to the actual chharacteristic of the
hydraulic load. When the load changes, the operative state of
the pump changes too. The significant pump parameters
susceptible to variations are the flow ratee, the suction and
discharge pressure, the power and the efficciency. The pump
operation must be then controlled in order too fit the requested
process parameters specifications and simultaneously
s
to
protect the pump itself. The selection of the
t more suitable
control system depends on several factors ass the existing plant
configuration, the required operational proccess state and the
economic factor[5]. The state of the art of the industrial
control systems presents four main differeent techniques for
pumping systems control:
A.
B.

C.

D.

throttle control: simplest and widelyy used method for

flow rate control, its based on thee installation of a
throttle valve on the discharge pipe .
r is sidelined in
by-pass control: a part of the flow rate
the suction tank through a by-passs line. The plant
characteristic is modified by thhe by-pass valve
installed on the by-pass pipe.
on/off control: characterized by thee use of a switch,
this method is extremely sim
mple but highly
inefficient. There are only two operrating states: on,
where the centrifugal pump runs at nominal speed
and off, where the pump is stoppeed.
VSD variable speed drive coontrol: centrifugal
pumps, at different speeds, have different
c
by the
characteristic curves which are connected
affinity laws.

c
techniques,
The brief description of the mentioned control
shows different approaches to flow rate control.
c
The table
below (Tab.1) provides a summary of the benefits and
limitations of the presented techniques according to below
listed requirements:

Control

II

IIII

IV

VI

VII

Throttle

By-pass

+/-

+/-

On/Off

VSD

Table 1 Benefits/limitaations of Control technique

The first three techniques are

a simple and low-cost but they
run the pump at a fixed speedd (nominal speed) and they are
consequently inefficient and inaccurate, producing higher
operating costs (higher enerrgy consumption and higher
machine wear). Only the lastt method is efficient. The VSD
technique has clearly higher im
mplementation costs. This kind
of control system adjusts the puump-motor speed [4] permitting
the pump to match the userss demand. The purpose of this
work is to present a centrifuggal pumps model-based control
system that ensures
energgy saving through the VSD
technique,
machine controol, monitoring and advanced
diagnostic features with a minim
mum number of sensors.
III. PUM
MP MODEL
In a pumping system, the total head of flowing liquid is
defined by the Bernoullis equaation as follows:
(1)
where:
H is the total head, pD and pS are discharge and suction
s
acceleration of gravity,
pressure respectively, g is the standard
is the fluid density, vD and vS are fluid velocities at discharge
and suction respectively, beingg v the ratio between the flow
rate and cross-sectional area of pipeline. Finally, zD and zS
represent the elevations of thhe discharge and suction point
above the reference plane, respectively.
The power required by the pump depends on the
t
head (H), the flow rate (Q)
characteristics of the fluid , the
and the overall pump efficiencyy (), as follows:

(2)
VSD systems are equipped with powerrful control units,
which enable the monitoring of sevveral modulation
parameters: inverter power output and actuall motor speed. The
relationship between these variables and the
t pump process
parameters are defined by pump performannce curves: flowHead (QH) and flow-Power (PQ) (Fig.3).

transformations define the chaanges in the pumping flow rate,

head, and pump absorbed poower following the change of
speed, as follows:
(3)

(4)

(5)
where the subscript 1 and 2 refers to the initial and new
l
of pump modeling is well
condition, respectively. This level
suited to be implemented joinntly to the VSD technique. In
order to perform pump diagnoostic, further characterization of
the pump is necessary; in partticular, the curves NPSHrQ, Q
and PHQ (PH, hydraulic poweer) varying with the speed of
rotation of the pump, shall be determined.
d
IV. MODEL BASED CON
NTROL AND DIAGNOSTIC
SYSTEM: SOLUTIO
ON AND ARCHITECTURE
In order to develop additioonal diagnostic capabilities, the
system described in this papeer implements few elementary
sensors. Compared to the fully
fu
sensorless technique this
choice introduces a very low additional
a
installation costs, but
in return it allows gaining the benefits
b
from enabled diagnostic
capabilities.

Figure 3 Pumps performance curves of Calpeeda NM20/160AE

Both these curves are one of the esseential parts of the

vendors pump documentation. Another releevant parameter is
the net positive suction energy NPSE orr the net positive
suction head NPSH. It is defined as the absolute suction head
Hs,abs minus vapor pressure expressed as heaad pv/(g) [5]. It is
obtained with an experimental procedure, which consists in
lowering progressively suction pressure until
incipient
b a head drop,
cavitation symptoms appear, indicated by
efficiency fall, noise and irregular flow. Experimentally,
h
decreases by
NPSHR is often defined as the NPSH3 (the head
3% at a given flow due to the cavitation onset
o
and reduced
hydraulic performance). The QH and QP curves show the
trend of the head and the power consumpption of the pump
against the flow rate, respectively. These corrrelations are used
in the presented system to determine, senssorlessly, the flow
rate. For variable speed pumps, manufactureers give the pump
performance map as an array of curves plootted for different
speeds. Anyway even when only the nominnal speed curve is
available its always possible to determinne the operational
points at an arbitrary speed by the affinity laws. The affinity

Figure 2 System Proposed

P
Architecture

The choice of the sensors for the implementation of the

proposed method starts from coonsiderations related to the cost:
the flow transducer results too be typically more expensive,
while pressure transducers andd transmitters are relatively less
expensive and hence preferred.. The proposed system uses two
pressure transducers installed near the pump flanges and a
vibration sensor to perform conntrol tasks and to implement the
pump diagnostic feature respectively. The complete
architecture of the system alsoo includes a frequency converter

and a PLC as reported in Figure 2. The PLC is used to provide

multivariable control features, to execute sequences tasks and
provide computation required by the pump dynamic model in
particular the calculation of the actual flow rate. Furthermore
the model-based algorithm allows to calculate the expected
values for Power, Efficiency and NPSHr.

Figure 3 Algorithm employed by the PLC

where:
pD, discharge pressure
pS, suction pressure
Hact, actual total head
Qact, actual flow rate
n, actual speed
PHe, hydraulic power expected
NPSHre, Net Positive Suction Head required expected
e, efficiency expected
V. DIAGNOSTIC ALGORITHM
The proposed diagnostic algorithm (Fig.4) is based on the
real-time monitoring of the main machine and process
parameters. In fact, the system calculates in real-time the
deviation between the "expected" and actual performance
parameters[6], [7]. An expected performance is a parameter
obtained from reference/design characteristic, adjusted to
actual operative condition, while an actual performance is a
value directly determined from field measurements and
acquisitions in a specific operative condition (off design). The
time trend of this deviation is a powerful diagnostic tool able
to give early indication of incoming problems. The diagnostic
system is based on the knowledge of the hydraulic power (Phe)
and the expected pump efficiency (e). These information are

used to determine the expected power (Pwe= Phe/ e) and

then compare this value with the actual power (Pw),
measured by frequency converter on-board sensor. The
difference Pw between the expected and actual power is
then compared to the power threshold, pre-setted during the
system configuration phase. The threshold value (PTH) is also
related to metrological features of transducers used. When the
difference between the two powers parameters is less than the
threshold (Pw < PTH ), the algorithm returns a "System
OK" state. Otherwise, (when Pw < PTH ), the system enters
the "Problem" diagnostic state.
In the specific case of the prototype, the diagnostic signal
generates a message on the HMI panel with the following
indications:
- Mechanical Problem: the system generates this
message when the condition is associated with the
detection of high vibration on the pump case. The
system compares the accelerometer signal (vib) to a
threshold (vibTH) generated by a simplified machine
vibration model, too.
- Mechanical Wear: message related to the detection
of low efficiency state (pW > pWe) without signals
from other conditional blocks (vib, NPSHr, pD).
Typically, the wear determines only low efficiency
problems.
- Instrumentation Problem: message related to the
detection of high efficiency state. Furthermore, this
message is also generated
at low efficiency
conditions with flag signals from the NPSHr or
discharge pressure control blocks.
All thresholds used by the control blocks are determined
by pump models implemented within the PLC, therefore all
thresholds change dynamically as function of the actual
machine operating conditions.
VI. TESTING OF METHOD
In this section, the laboratory test bench implemented for
testing of the proposed solution, and the obtained results will
be presented (Fig.5). Flow rate measurements are carried out
in order to evaluate the precision and accuracy of the flow rate
calculated by the system. During the testing procedure, the
flow rate values were obtained from the calculation algorithm
and compared to the values measured by a flow rate
volumetric transducer. In each test, the flow rate was changed
by throttling a valve installed on the discharge pipeline. The
valve has been used to simulate load variations. In this way,
the full range of the pump flow rate was covered. The
maximum flow rate of the centrifugal pump used during the

Figure 4 Diagnostic algorithm

testing is 6.6 m3/h. The relative errors and their respective

standard deviation values calculated at different speeds are

less than 2% and 1.25 respectively. The measurement points

have been chosen within the pump operating range.

Figure 5 Laboratory test bench plant

The verification of the diagnostic features with respect to

the algorithm logics, was performed by replacing the field
measured
signals
with
artificial
step
signals.
The artificial signals have been applied to the measurement of
power at first and subsequently to the measurement of the
pump body vibrations. Furthermore, in order to verify the
correct indication of the various problems, the discharge
pressure and suction pressure (thus the NPSHa) signals have
been artificially modified.

According to the logic presented in Figure 4, in normal

condition, the diagnostic system does not detect a difference
between measured and expected power than Pw does not
exceeds the power threshold PTH, therefore it sends System
OK message to HMI. In order to simulate a low efficiency
state, a test signal (step) has been used (Fig.6) so that pwe < pw
and |pwe - pw|> PTH. In this condition, the algorithm checks
also the signals relative to the other protection tasks. In case of
absence of flags the diagnostic system returns a "Mechanical
Wear" state indication(Fig. 7).

P
[kW]

1.30
0.98
0.65
Pw
Pwe

0.33
0.00
t [s]

Figure 6 Low efficiency state: power graph

Figure 7 Mechanical Wear state

In order to simulate the Mechanical Problem state, a

vibration test signal (step) has been used with an amplitude
exceeding the actual vibration threshold (Fig.8).

additional costs and the high reliability of the proposed

solution along with advanced diagnostic capabilities make the
system suitable to meet the oil and gas industry centrifugal
pumps control and protection requirements.

7.00
1.30
0.98
3.50
P
[kW]

vib
[mm/s]

5.25

1.75

Vib
VIBth

0.65
Pw
Pwe

0.33

0.00
t [s]

0.00
t [s]

Figure 8 Vibration test signal and threshold

Figure 10 Instrumentation Problem state:

high efficiency condition

Figure 9 Mechanical Problem state

The instrumentation problem state occurs when the high

efficiency case is detected or when low efficiency state is
active along with at least one flag signal coming from
conditional blocks: NPSHr or discharge pressure (pD). Figure
10 shows the test results in the first case. In the second case;
the test procedure is the same cases, and generates the same
result.
The results indicate that the implemented logic is
consistent with the diagnostics algorithm. The prototype
system constantly monitors all signals detected, calculates all
the main pump parameters and provides an overall machine
protection along with the control action requested by process
requirements.
VII. CONCLUSIONS
The proposed control systems allows to obtain a
reduction of operative energy costs by integration with the
VSD technique. Furthermore, the proposed model-based
method allows to perform real-time pump diagnostic, (without
the need of flow measurements and relative sensors) and
consequently increase the machine reliability and availability
(lowering maintenance and downtime costs). The moderate

VIII. REFERENCES
1.

2.

3.
4.
5.
6.

7.

F. Adamo, F. Attivissimo, A. Di Nisio, M. Spadavecchia,

Characterization and Testing of a Tool for Photovoltaic Panel Modeling,
IEEE on Instr. & Meas., vol. 60, pp. 1613-1622, 2011.,
F. Attivissimo, A. Di Nisio, M. Savino, M. Spadavecchia, Uncertainty
Analysis in Photovoltaic Cell Parameter Estimation, IEEE on Instr. &
Meas., vol. 61, pp. 1334-1342, 2012.,
I. Bakman, V. Vodovozov, Sensorless Pressure Control of Centrifugal
Pumps, IEEE Journal, 2013
Europump and Hydraulic Institute, Variable Speed Pumping: A Guide to
Successful Applications, Oxford, UK, Elsevier, Kidlington, 2004
Gulich J. F. Centrifugal Pumps, Springer-Verlag Berlin Heidelberg
2008, Second Edition, 2010
M. Di Febo, P. Paganini, Prediction of centrifugal compressor
performance and application for test, surge protection and machinery
diagnostic, CompressorTech2, May 2012
M. Di Febo, P. Paganini, Influence of inlet parameters on centrifugal
compressor surge limit line, CompressorTech2, June 2013