3480 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 60, NO.

10, OCTOBER 2011

State-of-the-Art Predictive Maintenance Techniques*

H. M. Hashemian, Senior Member, IEEE, and Wendell C. Bean, Senior Member, IEEE

Abstract—This paper discusses the limitations of time-based industrial equipment does not benefit from predictive main-
equipment maintenance methods and the advantages of predictive tenance technologies. The annual calibration of equipment
or online maintenance techniques in identifying the onset of equip- mandated by quality assurance procedures and/or regulatory
ment failure. The three major predictive maintenance techniques,
regulations is a typical example of these time-based meth-
defined in terms of their source of data, are described as follows:
1) the existing sensor-based technique; 2) the test-sensor-based ods. In 2007, Emerson/Rosemount Company published data
technique (including wireless sensors); and 3) the test-signal-based on the performance history of their pressure, level, and flow
technique (including the loop current step response method, transmitters in various industries (excluding the nuclear power
the time-domain reflectrometry test, and the inductance– industry), where time-based and hands-on maintenance tech-
capacitance–resistance test). Examples of detecting blockages in niques are still the norm. The data showed that during peri-
pressure sensing lines using existing sensor-based techniques and odic maintenance actions, technicians in these industries found
of verifying calibration using existing-sensor direct current out-
put are given. Three Department of Energy (DOE)-sponsored
transmitters to be experiencing no problems 70% of the time
projects, whose aim is to develop online and wireless hardware and [1]. By contrast, in nuclear power plants, where online or
software systems for performing predictive maintenance on criti- predictive maintenance is used, this number is over 90% [2].
cal equipment in nuclear power plants, DOE research reactors, The data in [1] also showed that 20% of periodic maintenance
and general industrial applications, are described. actions revealed calibration shift problems (followed by sensing
Index Terms—Inductance–capacitance–resistance (LCR) test- line blockage at 6% and failures at 4%). By comparison, only
ing, loop current step response (LCSR) method, predictive main- 5%–10% of periodic maintenance actions at nuclear power
tenance, time-domain reflectrometry (TDR) test, wireless sensor. plants reveal calibration shift problems (though sensing line
blockages occurred 10% of the time). The lower incidence of
I. I NTRODUCTION calibration faults in nuclear power plant pressure transmitters
suggests that the online monitoring and predictive maintenance
P REDICTIVE maintenance—sometimes called “online
monitoring,” “condition-based maintenance,” or “risk-
based maintenance”—has a long history. From visual inspec-
of equipment can reduce calibration problems in other indus-
tries as well.
tion, which is the oldest method yet still one of the most
powerful and widely used, predictive maintenance has evolved II. L IMITATIONS OF T IME -BASED
to automated methods that use advanced signal processing tech- M AINTENANCE T ECHNIQUES
niques based on pattern recognition, including neural networks,
fuzzy logic, and data-driven empirical and physical modeling. Though time-based and hands-on equipment maintenance is
As equipment begins to fail, it may display signs that can still the norm in industrial processes, these techniques have
be detected if sharp eyes, ears, and noses are used to sense increasingly been seen as flawed and unreliable in recent years
the failure precursors. Fortunately, sensors are now available [3]. Fig. 1 shows data on 30 identical bearing elements tested
to provide the sharp eyes, ears, and noses and identify the under identical conditions by SKF Group, a vibration and pre-
onset of equipment degradations and failures. Integrating these dictive maintenance analysis company. SKF engineers stressed
sensors with the predictive maintenance techniques described in the bearings to cause them to fail, then measured the time to
this paper can avoid unnecessary equipment replacement, save failure. As Fig. 1 shows, some failed in fewer than 15 h, and
costs, and improve process safety, availability, and efficiency. others lasted ten times longer or more; one operated for 300 h.
Despite advances in predictive maintenance technologies, Despite the identical bearings and identical conditions, the
time-based and hands-on equipment maintenance is still the time-to-failure varied by more than an order of magnitude. The
norm in many industrial processes. Nowadays, nearly 30% of conclusion to be drawn is that it is impossible to tell how long
a component may last in an industrial process. It is therefore
imprudent to set maintenance schedules based on failure time-
∗ An incorrect version of this paper first appeared in IEEE Trans. Instrum. based data like SKFs.
Meas., vol 60, no. 1, pp. 226–236, Jan. 2011. The correct version is printed Fig. 2 shows six curves that are currently considered as
here. failure models for industrial equipment such as pressure trans-
Manuscript received March 12, 2009; revised October 21, 2009; accepted
October 22, 2009. Date of current version September 14, 2011. The Associate
mitters, all of which can be managed by periodic time-based
Editor coordinating the review process for this paper was Dr. V. R. Singh. maintenance activities. The three curves on the left-hand
H. M. Hashemian is with the Analysis and Measurement Services (AMS) side—bath tub, wear-out, and fatigue—show the percentage of
Technology Center, AMS Corporation, Knoxville, TN 37923 USA (e-mail:
hash@ams-corp.com). time that the failure characteristics of equipment may follow
W. C. Bean is with Lamar University, Beaumont, TX 77710 USA (e-mail: each of the three failure types. According to U.S. Department
bean@ee.lamar.edu). of Defense (DOD) data, however, these types of failures ac-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. count for only about 11% of all failures [4]. The plots on the
Digital Object Identifier 10.1109/TIM.2009.2036347 right-hand side of Fig. 2 show the following three more recent

0018-9456/$26.00 © 2011 IEEE

Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on August 31,2023 at 18:32:37 UTC from IEEE Xplore. Restrictions apply.
HASHEMIAN AND BEAN: STATE-OF-THE-ART PREDICTIVE MAINTENANCE TECHNIQUES 3481

Fig. 1. Flawed time-based maintenance.

Fig. 2. Failure curves.

failure models that move past the time-based failure models: through a “burn-in” process before they are installed in plants
1) initial break-in period model; 2) random failure model; and and that increased monitoring is desirable during equipment’s
3) infant mortality failure model. The DOD’s data suggest that end-of-life or wear-out period.
89% of failures involve these three types, with infant mortality However, the bath tub curve cannot be used as the dominant
failures constituting by far the single largest type (68% of model for equipment degradation and failure. According to
failures). Moubray, Nolan, and Heap, its accuracy in predicting the aging
Although bath tub curve failures can be managed by periodic and eventual failure of equipment is only about 4%. Like
time-based maintenance activities, these activities are unreli- other time-based failure models, the bath tub curve provides no
able. The bath tub curve is an intuitive concept based on com- precision regarding when end-of-life occurs [5]. The length of
mon sense and empirical knowledge rather than engineering or an equipment’s stable period is never accurately known and will
scientific principles. It was used for many years to formulate clearly vary for different types of equipment and the different
aging management programs in industrial processes and to conditions in which that equipment is used. For an industrial
establish equipment replacement schedules. The idea behind component such as a process sensor, a shaft in a motor, or tubes
the bathtub curve is that the probability of failure of equipment in a heat exchanger, maintenance and replacement schedules
is high when equipment is new (“infant mortality”). This is are not easy to establish. Following the bath tub curve, some
followed by a period of stable performance and lower failures plants replace equipment periodically (e.g., every 5–10 years)
(in which there are only random failures), which eventually to avoid reaching the end-of-life. However, when a piece of
leads to increased failure probability (known as the end-of- equipment is replaced prematurely, the failure rate can actually
life or wear-out period). The bath tub curve is useful in that it increase because the new equipment may experience infant
suggests that to avoid infant mortality, equipment should be put mortality.
Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on August 31,2023 at 18:32:37 UTC from IEEE Xplore. Restrictions apply.
3482 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 60, NO. 10, OCTOBER 2011

Fig. 3. Integrated system employing three categories of techniques for predictive maintenance of industrial sensors and equipment.

only to indicate the pressure, but also to verify the calibration

III. P REDICTIVE OR O NLINE M AINTENANCE
and response time of the sensor itself and to identify anomalies
As noted, predictive maintenance is the preferred mainte- in the process such as blockages, voids, and leaks that can
nance method in 89% of cases, compared to time-based main- interfere with accurate measurement of process parameters or
tenance, which is prudent in only 11% of cases. Because of disturb the plant’s operation, safety, or reliability. The second
the unreliability of time-based maintenance methods, industrial category of predictive maintenance methods uses data from
processes should use online monitoring not only when equip- test sensors such as accelerometers for measuring vibration and
ment is old but throughout the life of equipment to identify the acoustic sensors for detecting leaks. This group of methods is
onset of degradation and failure of equipment. where wireless sensors can play a major role.
One form of predictive maintenance, online calibration mon- These first two classes of predictive maintenance techniques
itoring, provides an example of how predictive maintenance are passive. In contrast, the third category of predictive main-
works. Online calibration monitoring involves observing for tenance technology depends on signals that are injected into
drift and identifying the transmitters that have drifted beyond the equipment to test them. This category includes active mea-
acceptable limits. When the plant shuts down, technicians cal- surements such as insulation resistance tests and inductance,
ibrate only those transmitters that have drifted. This approach capacitance, and resistance measurements, also known as LCR
reduces by 80%–90% the effort currently expended on calibrat- testing. These methods are used to detect defects such as cracks,
ing pressure transmitters. This is according to data published by corrosion, and wear for the predictive maintenance of cables,
the author in a report he wrote for the U.S. Nuclear Regulatory motors, sensors, and other equipment. This third category of
Commission [6]. predictive maintenance techniques also includes the loop cur-
Predictive or online maintenance can be divided into the rent step response (LCSR) method, which was developed in the
following three basic techniques (Fig. 3), based on their data mid-1970s by graduate students, professors, and technicians at
sources: 1) the existing sensor-based maintenance technique; the University of Tennessee (including the author). The LCSR
2) the test-sensor-based maintenance technique; and 3) the test- method is routinely used in nuclear power plants for in situ
signal-based maintenance technique. The first category consists response-time testing of RTDs and thermocouples as well as
of maintenance methods that use data from existing process in other applications [7].
sensors—such as pressure sensors, thermocouples, and resis- Table I lists the typical types of industrial equipment that
tance temperature detectors (RTDs)—that measure variables can benefit from these three predictive maintenance technolo-
like temperature, pressure, level, and flow. In other words, the gies and the parameters that may be monitored. Table II lists
output of a pressure sensor in an operating plant can be used not examples of industries that can use the three types of online
Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on August 31,2023 at 18:32:37 UTC from IEEE Xplore. Restrictions apply.
HASHEMIAN AND BEAN: STATE-OF-THE-ART PREDICTIVE MAINTENANCE TECHNIQUES 3483

TABLE I
PARAMETERS R ELATED TO E QUIPMENT C ONDITION

TABLE II
A PPLICATIONS OF E QUIPMENT C ONDITION M ONITORING

or predictive maintenance and the typical purposes for which plants with performance information. For example, pressure
each industry may use them. In 20–30 years, most industrial sensors can be used to monitor thermal hydraulic effects in
processes will have an integrated online condition monitoring addition to measuring pressure. Furthermore, standing waves,
system for performing automated predictive maintenance. turbulence, and flow-induced vibration effects can be diagnosed
using the normal output of existing pressure sensors in an
operating process.
A. Category 1 Online or Predictive Maintenance Techniques:
Fig. 4 shows how plants can use the output of existing
Existing Sensors
sensors—the first group of online or predictive maintenance
At frequencies usually below 30 Hz, existing process sensors techniques—to verify the performance of the sensors them-
can not only measure a process parameter, but can also provide selves. The figure shows the normal output of a process sensor,
Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on August 31,2023 at 18:32:37 UTC from IEEE Xplore. Restrictions apply.
3484 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 60, NO. 10, OCTOBER 2011

Fig. 4. (a) Typical output of a process sensor. (b) Separating the ac and dc
components of a sensor output. Fig. 5. Effect of sensing line blockage on dynamic performance of a pressure
transmitter.

such as a pressure, level, or flow transmitter, plotted as a

function of time. The dc component of the output is used to Fig. 5 shows an application of this method. Two PSDs are
monitor for drift for online calibration monitoring, and the ac shown in the lower right-hand plot—one for a partially blocked
component of the output (referred to as noise) is used to verify sensing line and the other for the same sensing line after it was
the dynamic performance of the sensor (e.g., response time) and purged of the blockage. These PSDs are taken from an analysis
to identify sensing line blockages. of actual noise data from a nuclear plant pressure transmitter.
Fig. 4 also shows how the ac component of a sensor output is Observing the break frequency, it is clear that the blockage
separated from its dc component and sampled into a computer reduced the dynamic response of the sensor by an order of
through an antialiasing filter and then analyzed. The analysis magnitude.
usually involves performing a fast Fourier transform of the 2) Example—Using Existing Sensor DC Output to Verify
noise data. The results are presented in terms of a spectrum that Sensor Calibration: Another example of existing sensor-based
is known as the power spectral density (PSD) of the ac signal. predictive maintenance techniques is verifying the calibration
The PSD is actually the variance of the noise signal in a narrow of process sensors using the dc output of the sensors. Fig. 6
frequency band plotted as a function of frequency. shows the dc output of four redundant pressure transmitters
After the noise data are analyzed, the PSD is used to calculate from a nuclear power plant. The plot represents the deviation
the response time of the pressure transmitter. An approximate of each transmitter from the average of the four transmitters. It
value for response time can be estimated by noting the break can be seen that the dc output of these transmitters did not drift
frequency. To accurately measure response time, the PSD over the two years shown. This fact could be used to argue that
should be fit to a transfer function model of the sensor from these four transmitters do not need to be calibrated. However,
which the response time is calculated. these data are based on only a one-point calibration check at the
1) Example—Using Existing Sensor AC Output to Detect plant operating point. To verify calibration over the entire range
Blockages in Pressure Sensing Lines: As noted, sensing line of the transmitter, data should be taken not only at the operating
blockages are the second most common problem affecting point, but also at startup and shutdown periods. Fig. 7 shows
pressure transmitters. Using existing sensor-based predictive calibration monitoring results for a pressure transmitter over a
maintenance techniques, one can identify the sensing lines that wide operating range. The limits shown in this figure represent
must be purged and cleared rather than having to purge all the acceptance criteria for the allowable drift of this particular
of them. This technique makes it possible to detect blockages transmitter in this particular plant. These limits are calculated
while the plant is online using the normal output of pressure by accumulating the uncertainty values for the components that
transmitters at the end of the sensing line [8]. are involved in measuring this process parameter. The limits
Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on August 31,2023 at 18:32:37 UTC from IEEE Xplore. Restrictions apply.
HASHEMIAN AND BEAN: STATE-OF-THE-ART PREDICTIVE MAINTENANCE TECHNIQUES 3485

illustrated in Fig. 10, which shows reactor power as a function

of time. Two traces are shown. One is from empirical modeling
of the reactor for the purpose of calculating actual power, and
another is from the plant power-level indication system. The
indicated power trends upward, erroneously, while the actual
power is stable, as it should be. The reason for the discrepancy
is fouling of the venturi flow element. As venturi fouls, the
indicated flow rate rises and causes the power indication to
rise accordingly. Analytical modeling, such as by empirically
modeling the plant process using neural networks, enables
plants to identify the onset of venturi fouling and to see the
extent of error in the indicated power. Fig. 10 shows that in this
Fig. 6. DC output of four redundant pressure transmitters in a power plant.
example, at 500 days of operation, the error is about 2.5% in
power.

B. Category 2 Online or Predictive Maintenance Techniques:

Test Sensors, Including Wireless
The second category of online or predictive maintenance
techniques is, like the first, passive. However, rather than re-
lying on existing sensors for its data, they use data from test
or diagnostic sensors. Typical examples of test sensors are
accelerometers for measuring vibration and acoustic sensors
for detecting leaks. For example, acoustic sensors installed
downstream of valves can establish whether the valve is op-
erating as expected: if a valve is completely open or completely
closed, there is normally no detectable acoustic signal above the
background noise.
When existing test sensors are not available to provide the
Fig. 7. Online monitoring results for calibration check of a pressure transmit-
ter as a function of pressure range. necessary data, wireless sensors can be deployed to fill the
gap. For example, wireless sensors can be implemented in such
a way as to combine vibration, acoustic, and other data with
will differ for different transmitters in different plants, depend- environmental information such as humidity and ambient tem-
ing on the methodology used to calibrate the instruments. perature to yield a comprehensive assessment of the condition
To facilitate the use of existing sensor-based predictive of the process’s equipment and health.
maintenance, the author and his colleagues have developed Wireless sensors can facilitate difficult measurements in
software for measuring sensor calibration online. Fig. 8 shows industrial processes where wiring is a weak link, such as
the software screen, which consists of four windows of data. flame temperature and high furnace-temperature measure-
The window on the top left presents the raw data for nine ments. Wireless sensors also facilitate measurement in haz-
transmitters during a plant shutdown. Below this is a plot of ardous environments and in applications where space for wiring
the deviation of each of the transmitters from the average of installation is limited. Wireless sensors can also be the answer
the nine. The calibration limits are also shown in this deviation to the threat posed by rust, corrosion, steam, dirt, dust, and
plot. The bottom-right window is a plot of calibration error for water to wires in industrial facilities. With wireless sensors,
each transmitter as a function of its operating range. The table in data can be collected from anywhere and routed on to the In-
the upper right shows the results of the calibration monitoring: ternet where they can easily be accessed and analyzed. Finally,
eight transmitters passed the calibration monitoring test; one wiring costs in an industrial process can be as high as $2000
failed it. per foot versus $20 per foot if wireless equipment were used
3) Using Empirical or Physical Modeling to Verify Calibra- for the same applications. Wireless sensors promise to enhance
tion: In some cases, a process will not have enough redundant industrial productivity, lower energy and material costs, and
existing sensors to provide a reference for online calibration increase equipment availability.
monitoring. In these cases, in addition to averaging the redun- According to a 2007 survey of engineering professionals in
dant sensors, the process can be modeled empirically to create the International Society of Automation’s InTech magazine,
analytical sensors [9]. This method is illustrated in Fig. 9. Note wireless technologies are the largest area of technology growth
that both physical and empirical models can be used to create in the near future [1].
analytical sensors. To promote the growth of wireless-based predictive main-
Using empirical modeling, plants can identify not only cal- tenance, the Analysis and Measurement Services (AMS) Cor-
ibration problems, but also other instrument anomalies such poration has developed a software package, called “Bridge,”
as fouling of venturi flow elements in power plants. This is that is able to read data from wireless sensors from different
Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on August 31,2023 at 18:32:37 UTC from IEEE Xplore. Restrictions apply.
3486 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 60, NO. 10, OCTOBER 2011

Fig. 8. Software screen with results of online calibration monitoring.

Fig. 9. Conceptual design of analytical sensor for online calibration monitoring.

manufacturers, bringing all of a plant’s wireless data together

in one place in one format so it can be analyzed and compared.

C. Category 3 Predictive or Online Maintenance Techniques:

Test Signals
The first two categories of predictive maintenance tech-
niques were mostly passive, do not involve perturbations of the
equipment being tested, and can be performed in most cases
while the plant is operating. The third category of predictive
maintenance methods can be described as methods that depend
on active measurements from test signals.
One form of test-signal-based predictive maintenance in-
volves injecting a signal into the equipment to measure their
performance. For example, a method called the LCSR test can
Fig. 10. Detection of venturi fouling. remotely measure the response time of temperature sensors as
Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on August 31,2023 at 18:32:37 UTC from IEEE Xplore. Restrictions apply.
HASHEMIAN AND BEAN: STATE-OF-THE-ART PREDICTIVE MAINTENANCE TECHNIQUES 3487

Fig. 11. Principle of LCSR test for in situ measurement of an RTD response time.

installed in a plant while the plant is online [7]. This method

sends an electrical signal to the sensor in the form of a step
change. Fig. 11 shows the setup for performing LCSR on an
RTD. A Wheatstone bridge is used to send a step charge in
current to the RTD. The current causes heating in the RTD
sensing element and produces an exponential transient at the
bridge output. This transient can be analyzed to give the re-
sponse time of the RTD. The LCSR test can be used for other
purposes such as finding water levels in a pipe. It can also be
used to verify that temperature sensors are properly installed
in their thermowells in a process and that they respond to
temperature changes in a timely manner. Similarly, the LCSR
method can be used to determine if aging causes degradation
in the dynamic performance of temperature sensors so sensor
replacement schedules can thereby be established. In 1979, the
author helped use the LCSR technique during the Three Mile
Island nuclear plant accident to assist the plant to determine
water level in the primary coolant pipes. Finally, the LCSR
method can also be used to detect the bonding of sensors such
as RTDs and strain gauges to solid surfaces (Fig. 12).
Fig. 13 shows the response time of several thermocouples
measured by the LCSR test. There is one outlier, which had a
secondary junction that was not at the tip of the thermocouple,
causing its response time to be much higher than the others.
The LCSR method helped identify that this thermocouple’s
measuring junction is not where it should be and thus indicated
an erroneous temperature.
Test-signal-based predictive maintenance methods also in-
clude the time-domain reflectrometry (TDR) test. It is used Fig. 12. (a) LCSR can determine the degree of bonding of temperature sensors
to solid surfaces. (b) LCSR can determine the degree of bonding of strain
to locate problems along a cable, in a connector, or at an gauges to solid surfaces.
end device by injecting a test signal through the conductors
in the cable and measuring its reflection. The TDR technique
has also served the power and process industries in testing the cable, and its reflection is plotted versus time. The plot will
instrumentation circuits, motors, heater coils, and a variety of show any changes in impedance along the cable, including at
other components. In a TDR test, a step signal is sent through the end of the cable.
Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on August 31,2023 at 18:32:37 UTC from IEEE Xplore. Restrictions apply.
3488 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 60, NO. 10, OCTOBER 2011

Fig. 13. LCSR for in situ testing of a thermocouple problem.

Fig. 14. (a) RTD circuit. (b) TDR test results of the RTD circuit.

Fig. 14 shows the TDR of a three-wire RTD as installed in critical. Fig. 16 shows a TDR of a thermocouple measured in
a plant. It clearly shows all locations where there is a change 1999 and 2008. The 2008 data indicate that the thermocouple
in impedance, including the end of the cable that has an RTD. impedance has changed significantly.
If the TDR is trended, problems that may develop along the A complementary set of cable tests that measure impedance,
cable or at the end device can be identified and located. The called LCR testing, is often used in addition to TDR to identify
simplest application of TDR is locating an open circuit along a whether a circuit problem is caused by an open circuit, a
cable. You can tell if the circuit is open by measuring its loop short circuit, a shunt, a moisture intrusion, or other problems.
resistance, but only a TDR would tell you where the circuit Alternatively, one can perform an LCR test by itself to mea-
is open. Fig. 15 shows how TDR identified a failed RTD in a sure cable inductance and capacitance and trend the values to
nuclear power plant, where knowing if the open circuit is inside identify problems or compare against a baseline. Typical cable
the reactor containment or outside the reactor containment is circuits can be characterized both by series inductance and
Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on August 31,2023 at 18:32:37 UTC from IEEE Xplore. Restrictions apply.
HASHEMIAN AND BEAN: STATE-OF-THE-ART PREDICTIVE MAINTENANCE TECHNIQUES 3489

Fig. 15. Failed RTD detected by TDR.

Fig. 16. Results of TDR monitoring of a thermocouple.

resistance and by parallel capacitance of the dielectric. Fig. 17

shows capacitance measurement results for ten RTDs as well as
dissipation factor calculations. These are all normal values.
Although using TDR and LCR measurements is not compli-
cated and they have been used for many years, their applications
in industrial processes are still new.

Fig. 17. (a) Capacitance measurement results for ten RTDs and dissipation
IV. P REDICTIVE M AINTENANCE R&D P ROJECTS factor calculations (for lead X1 to ground). (b) RTD circuit for the results shown
in this table.
The three predictive maintenance technologies discussed in
this paper are part of three U.S. DOE projects under the Small DOE’s Research Reactors,” is to determine the feasibility of
Business Innovation Research program to subsidize R&D work using wireless sensors to monitor the condition of equipment in
at small (under 500 employees) high-technology firms that can research reactors such as the high-flux isotope reactor (HFIR)
convert the R&D results into commercial products. The goal at the Oak Ridge National Laboratory.
of the first of these projects, namely, “On-Line Monitoring of Since the wireless and HFIR projects are still under way, we
Accuracy and Reliability of Instrumentation,” is to develop will focus here on the progress made in the Online monitoring
systems for online condition monitoring of equipment and (OLM) project. The OLM project is aimed at developing new
processes in industrial plants. As noted previously, the goals of technology that uses signals from existing process sensors to
the second project, namely, “Wireless Sensors for Equipment verify the performance of the sensors and assess the health of
Health and Condition Monitoring in Nuclear Power Plants,” the process. The work involves both software and hardware
are to develop data qualification and data processing techniques development, testing, and validation using existing hardware
for wireless sensors, to identify which parameters these sensors and software that have been adapted to the project. The soft-
should measure, and to determine the correlation between these ware packages we developed are for data acquisition, data
parameters and the actual condition of the equipment being qualification, and data analysis equipment. This equipment is
monitored. The goal of the third project, namely, “Wireless designed for fast data acquisition and will be used for ac and dc
Sensors for Predictive Maintenance of Rotating Equipment in data acquisition. The AMS Corporation is currently adapting
Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on August 31,2023 at 18:32:37 UTC from IEEE Xplore. Restrictions apply.
3490 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 60, NO. 10, OCTOBER 2011

Fig. 18. (a) and (b) Screenshots of OLM software.

V. C ONCLUSION
National Instruments Company’s new data acquisition product
CompactRio to the OLM project. Writing the algorithm and Industrial plants should no longer assume that equipment
software packages and testing and implementing a prototype failures will only occur after some fixed amount of time in
system in an operating plant constitute the bulk of the OLM service; they should deploy predictive and online maintenance
project’s current work. Fig. 18 shows software screenshots of strategies that assume that any failure can occur at any time
an ac signal analysis to detect dynamic anomalies in equipment (randomly). The onset of equipment failure may manifest itself
and processes, and Fig. 19 shows software screens of data in data generated by the methods used to monitor the equip-
qualification results before they are analyzed. Under the OLM ment, providing clues as to whether the equipment should be
project, the AMS Corporation is developing statistical packages repaired, replaced, or left to continue in operation. Ongoing
for data qualification. For example, we calculate the amplitude research into the three major types of predictive or online
probability density of the data and look for nonlinearity and maintenance technologies discussed in this paper promises to
other problems in the data before we send the data through for deliver technologies that may be applied remotely, passively,
analysis. We also calculate signal skewness, kurtosis, and other and online in industrial processes to improve equipment re-
measurements and trend them to identify problems in data. liability, predict failures before they occur, and contribute to
Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on August 31,2023 at 18:32:37 UTC from IEEE Xplore. Restrictions apply.
HASHEMIAN AND BEAN: STATE-OF-THE-ART PREDICTIVE MAINTENANCE TECHNIQUES 3491

Fig. 19. (a) and (b) Data qualification results.

process safety and efficiency. Integrating the predictive mainte- [3] R. K. Mobley, An Introduction to Predictive Maintenance, 2nd ed.
nance techniques described in this paper with the latest sensor St. Louis, MO: Butterworth-Heinemann, 2002.
[4] F. S. Nowlan and H. F. Heap, “Reliability centered maintenance,” Nat.
technologies will enable plants to avoid unnecessary equipment Tech. Inf. Service, Springfield, VA, Rep. A066-J79, 1978.
replacement, save costs, and improve process safety, availabil- [5] J. M. Moubray, Reliability Centered Maintenance. New York: Industrial
ity, and efficiency. Press, 1992.
[6] H. M. Hashemian, “On-line testing of calibration of process instrumenta-
tion channels in nuclear power plants,” U.S. Nucl. Regulatory Commission,
Rockville, MD, NUREG/CR-6343, Nov. 1995.
[7] H. M. Hashemian, Sensor Performance and Reliability. Research
R EFERENCES Triangle Park, NC: ISA, 2005.
[1] G. Hale, “InTech survey: Predictive maintenance top technology challenge [8] H. M. Hashemian, “Accurately measure the dynamic response of pressure
in 2007,” in InTech. Pittsburgh, PA: ISA, Jan. 2007. instruments,” Power, pp. 72–78, 2007.
[2] H. M. Hashemian et al., “Advanced instrumentation and maintenance [9] J. W. Hines and R. Seibert, “Technical review of on-line monitoring tech-
technologies for nuclear power plants,” U.S. NRC, Rockville, MD, niques for performance assessment,” U.S. Nucl. Regulatory Commission,
NUREG/CR-5501, 1998. Rockville, MD, NUREG/CR-6895, Jan. 2006, vol. 1, State-of-the-Art.
Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on August 31,2023 at 18:32:37 UTC from IEEE Xplore. Restrictions apply.
3492 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 60, NO. 10, OCTOBER 2011

H. M. (Hash) Hashemian (SM’07) received the Wendell C. Bean (SM’97) received the Ph.D. de-
B.S. degree in physics from the National University gree in electrical engineering from the University of
of Iran, Tehran, Iran, the M.S. degree in nuclear Pittsburgh, Pittsburgh, PA.
engineering from the University of Tennessee (UT), In 1968, he joined Lamar University, Beaumont,
Knoxville, TN, the D.Eng. degree in electrical en- TX, as Head of the Department of Electrical Engi-
gineering from Lamar University, Beaumont, TX, neering, where he is currently a Professor of elec-
and the Ph.D. degree in nuclear engineering from trical and nuclear engineering. From 1955 to 1967,
Chalmers University of Technology, Göthenburg, he was an Engineer with the Bettis Atomic Power
Sweden. Laboratory, Pittsburgh, PA. From 1961 to 1967, he
He has been with the Analysis and Measurement taught, for the U.S. Naval Reactor Branch under Ad-
Services (AMS) Corporation, Knoxville, TN, since miral Rickover, the Nuclear Reactor Plant Analysis
1977 when the company was founded. He specializes in process instrumen- and Control postgraduate course for naval ROTC graduates located at the Bettis
tation, equipment condition monitoring, online diagnostics of anomalies in Laboratory. During 1967–1968, he was a National Institutes of Health Special
industrial equipment and processes, automated testing, and technical training. Research Fellow, where he did postdoctoral fellowship with the University of
He is the author of two books, six book chapters, and over 200 published Michigan in baroreceptor response for the cardiovascular system.
papers. His books have been translated to Chinese, Japanese, and Russian. He Dr. Bean is a member of Sigma XI, the Honor Society for Engineering Tau
is the holder of nine U.S. patents. Through AMS Corporation, he has worked Beta Pi, and Eta Kappa Nu. He is a registered Professional Engineer in nuclear
for almost all the 104 nuclear power plants that are currently operating in the engineering in Texas. He was the recipient of the Millennium Medal from a
U.S. Furthermore, he has presented seminars, training course, and lectures to local chapter of the IEEE.
nuclear industry professionals in 15 countries. In addition to technical and
business management of AMS Corporation, he has been involved with national
and international organizations concerned with nuclear energy such as the
U.S. Department of Energy, the U.S. Nuclear Regulatory Commission, the
International Atomic Energy Agency, and the International Electrotechnical
Commission.
Dr. Hashemian was elected a Fellow of the International Society of Automa-
tion in 1992 and is a member of the American Nuclear Society and the European
Nuclear Society.

Authorized licensed use limited to: Universidad Federal de Pernambuco. Downloaded on August 31,2023 at 18:32:37 UTC from IEEE Xplore. Restrictions apply.