Tribology International 109 (2017) 473–484

Contents lists available at ScienceDirect

Tribology International
journal homepage: www.elsevier.com/locate/triboint

Lubricating oil conditioning sensors for online machine health MARK

monitoring – A review
⁎
Xiaoliang Zhu, Chong Zhong, Jiang Zhe
Department of Mechanical Engineering, University of Akron, Akron, OH 44325, USA

A R T I C L E I N F O A BS T RAC T

Keywords: Analysis of lubricating oil is an effective approach in judging machine's health condition and providing early
Oil condition monitoring warning of machine's failure progression. Many studies from both academia and industry have been conducted.
Wear debris This paper presents a comprehensive review of the state-of-the-art online sensors for measuring lubricant
Lubricant properties properties (e.g. wear debris, water, viscosity, aeration, soot, corrosion, and sulfur content). These online sensors
Online sensor
include single oil property sensors based on capacitive, inductive, acoustic, and optical sensing and integrated
sensors for measuring multiple oil properties. Advantages and disadvantages of each sensing method, as well as
the challenges for future developments, are discussed. Research priorities are defined to address the industry
needs of machine health monitoring.

1. Introduction the machine). In comparison, online sensors, installed on the machine,

monitor a portion of the lubricating oil from the continuously
Machine condition monitoring has become essential in maintaining circulated lubricant flow in-situ during operation. Currently, off-line
and extending the health of high speed rotating and reciprocating measurement methods including spectrography and ferrography still
machinery used in many of the nation's key industries including remain the most commonly used strategy for diagnosing oil conditions
aerospace, manufacturing, and energy [1]. The real time condition [1]. Although offline methods can provide comprehensive and detailed
monitoring of a machine's health status can significantly reduce the information about oil conditions, the test procedures are time-con-
operation cost by eliminating the need for costly machine shutdowns suming and often require expensive equipment and skilled analysts [2].
for inspection, which would otherwise be needed to avoid the possibi- Furthermore, this method cannot provide real-time information about
lity of catastrophic component failure during operation. Accurate machine health that could be used for avoiding the catastrophic
condition monitoring methodologies are being sought to facilitate the component failure during operation and providing a prognosis of
effective scheduling of maintenance and repair downtime, particularly pending machine fault. Online, portable devices that can perform real
in military and the air transport industries, to ensure the safety of long- time analysis of lubricating oil with no need for complicated setup and
range operations. Although vibration analysis [2] and thermography skilled analysts are highly desired.
[3] have been used in detecting severe faults of rotating machinery, During the last two decades, a wide range of studies, from both
these methods require sophisticated data acquisition and computa- academia and industry, have been conducted in developing online oil
tional procedures, which are difficult to implement. In addition, little or conditioning sensors, which can provide real time and continuous
no vibration/temperature change can be noticed at the early stage of a monitoring of machine's health condition. Although several review
fault developing. More importantly, these methods are usually ineffec- papers have been published in recent years, the majority of the papers
tive in identifying machine's wear conditions and providing a prognosis to date were focusing on sensors to test one kind of oil property or
of pending machine failure [4]. sensors (mainly offline sensors) to be used in a special machine, e.g.
Lubricating oil analysis has become an effective mean to provide wind turbine. In this article, we intend to provide a comprehensive
early warnings in the failure progression because it contains valuable review of the state-of-the-art online sensors for measurement of the oil
information regarding the aging and damage of oil-wetted moving properties including wear debris, water, viscosity, aeration, soot,
components [5]. For lubricating oil property sensors, offline sensors corrosion, sulfur content, etc., with a particular focus on the develop-
require a physical sample of the lubricating oil be taken from the ments in the past five years. Advancements on online oil property
lubrication system and then be analyzed in the laboratory (away from sensors based on capacitive, inductive, acoustic, and optical methods

⁎
Corresponding author.
E-mail address: [email protected] (J. Zhe).

http://dx.doi.org/10.1016/j.triboint.2017.01.015
Received 5 October 2016; Received in revised form 10 January 2017; Accepted 13 January 2017
Available online 18 January 2017
0301-679X/ © 2017 Elsevier Ltd. All rights reserved.
X. Zhu et al. Tribology International 109 (2017) 473–484

[17–19]
[16]

[13]

[23]

[24]

[25]
Ref.

[8]

[9]
Low sensitivity, may identify multiple debris particles as

complicated structure, cannot differentiate metallic and

Low throughput, complicated system, affected by oil

cannot detect 1) individual debris particles, 2) non-

Low sensitivity, water influence, large sensing zone

Low throughput, complicated flow recess structure

one large debris particle due to large sensing zone

Extremely low throughput, water influence

Fig. 1. Relationships of wear debris size, concentration, and machine conditions.

ferrous debris particles

for both single oil property and multiple properties measurements are

non-metallic debris
Low throughput
reviewed. The remaining of this paper is organized as follows: In

Disadvantage

transparency
Section 2, the online sensor techniques used for measuring one oil
property are reviewed. Section 3 reviews the integrated sensors for
measurement of multiple oil properties. The future work to fulfill the
industry need is discussed in Section 4.

Differentiate ferrous, nonferrous, solid debris

Differentiate ferrous and non-ferrous debris,

Can 1) differentiate ferrous and non-ferrous

2. Single oil property sensors

debris, 2) individual wear debris particles

Detect debris particle shape and material

Clear up ferrous debris, high throughput
2.1. Wear debris sensors

Detect solid debris and air bubbles

Detection of small wear debris is critical to identify abnormal wear
conditions for prognosis of pending machinery failure. A typical
correlation between time, wear debris size and wear debris concentra-
tion is shown in Fig. 1 [6], although wear debris size range at specific

High throughput
high throughput

High sensitivity

and air bubbles

machine conditions varies with different machines. During initial/
Advantage

normal operation of a new engine or gearbox, the wear debris size is

typically between 1 µm and 10 µm, and the concentration remains
approximately constant. When the machine is in the abnormal condi-
tion, larger debris particles between 20 µm and 100 µm (between
Detection range

10 µm and 50 µm for wind turbines) are seen. The wear debris size
100–3360 µm

170–1000 µm
20–1000 µm

50–310 µm
and concentration gradually increase with machine operation time 10–40 µm

5–160 µm
until machine fails [7]. Studies [8,9] have shown that, when debris size
is larger than 100 µm (comparable to the diameter of human hair), the
NA

NA

machine is in critical conditions and an immediate machine main-

Solid debris, air bubble

tenance is needed to avoid machine failure. Thus, to give an early

ceramic, air bubble
warning of the machine conditions, debris size larger than 20 µm
Ferrous and non-

Ferrous and non-

Material detected

Metallic debris,
Metallic debris
Metallic debris
Ferrous debris

(10 µm for wind turbines) is needed to be monitored.

ferrous debris

ferrous debris

Solid debris

Several methodologies have been adopted since wear debris present

in lubricating oil changes the oil's permittivity, permeability, conduc-
tivity, and optical properties. By monitoring these parameters, the
presence, size, shape, and concentration of wear debris can thus be
obtained. To be noted that, some of the online sensors reviewed in the
Ultrasonic oil debris sensor

following sections are already commercialized, e.g. Gill Sensor and

Microfluidic capacitance
Bulk capacitance sensor

MetalSCAN sensor, which can be implemented in a fleet of engines or

Inductive pulse sensor

Integrated ultrasonic–
inductive pulse sensor
Active pixel sensor
MetalSCAN sensor

other machines to measure lubricating oil conditions during operation.

Other online sensors, such as the microfluidic capacitance sensor
Specific sensor

Gill sensors

(Section 2.1.2) and the 2-layer planar coil sensor (Section 2.1.3) are
still tested in bench-top experimental setups. A summary of the wear
sensor

debris sensors has been listed in Table 1.

As summarized in Table 1, many wear debris sensors can be
3-D solenoid structure

adapted to detect wear debris within a specific size range, including

Summary of the wear debris sensors.

Magnetic collection
Monitoring methods classification

Acoustic amplitude
Dielectric constant

microfluidic capacitance sensor (Section 2.1.2), MetalSCAN sensor

2-layer planar coil

(Section 2.1.3), 2-layer planar coil sensor (Section 2.1.3), ultrasonic-

Wear debris
morphology

based sensor (Section 2.1.4) and optical-based sensor (Section 2.1.5).

structure

Taking the inductive sensor with a 2-layer planar coil for example [19],
change

it can detect ferrous debris particles between 20 µm (lower detection

limit) and 1000 µm (upper detection limit). A physical filter can be
Capacitance
Inductance

added in the upstream to only allow particles below a certain size to

Ultrasonic

pass through and be detected. The size and count of all debris particles
Optical
Table 1

can be measured. Debris and counts of debris particles with a certain

size range can thus be obtained.

474
X. Zhu et al. Tribology International 109 (2017) 473–484

Fig. 2. Schematic of the ferrous debris sensors developed by Gill Sensors.

2.1.1. Wear debris sensors based on bulk measurement Although this device can detect wear debris as small as 10 µm, it
Wear debris includes ferrous, non-ferrous, and ceramic debris. has the following disadvantages. First, the sensor is unable to
Among wear debris, ferrous debris is the largest portion because many differentiate ferrous and nonferrous wear debris since these debris
machine components are made of steel. Therefore, it is an important particles have a similar permittivity. Second, the measured capaci-
information carrier for machine's wear condition. Gill Sensors has tances also reflect the change in acid number content and viscosity,
developed a ferrous debris sensor, which has been utilized in monitor- creating difficulties in detecting debris. Third, while air bubbles
ing the health of gearbox and transmission systems of land-based (εair=1.0, compare with εoil ~2.0) have no noticeable influence on
vehicles, to detect the ferrous debris in lubricating oil [10]. The sensor's the output of the sensor, water droplet could generate a large capacitive
structure is shown in Fig. 2. It consists of two identical sensor units; pulse because water's relative permittivity (εwater~80.0, under room
each unit has an inductance sensing coil and a permanent magnet for temperature) is much larger than that of lubricating oil. The ultra-low
collecting ferrous debris to the sensor tip. The two units, one for throughput is another challenge for this device owing to the use of a
detecting fine ferrous debris build-up (Sensor 1 in Fig. 2) and the other micro-sized microchannel. Therefore, its capability for real time wear
for detecting large ferrous debris (Sensor 2 in Fig. 2), are able to debris monitoring is limited.
determine the amount of ferrous debris build-up based on the sensing
coil's inductance change. Furthermore, because of the permanent 2.1.3. Wear debris sensors based on inductive method
magnet, this sensor also prevents debris from continuing to travel In an attempt to differentiate ferrous and non-ferrous debris as well
through the oil circulation system, reducing further damage. However, as to eliminate the influences caused by air bubbles and water droplets,
this sensor relies on the bulk measurement of a large amount of ferrous inductive sensors based on 3-D solenoid coils were developed. Its
debris attracted to the sensor tip over a period of time. Therefore, it general working principle is illustrated in Fig. 4. The inductance change
cannot provide real time information about individual ferrous debris of the sensor is determined by two factors, namely magnetic perme-
particles’ size and concentration. Additionally, it is unable to detect the ability and eddy current [15,16]. If a non-ferrous conductive metallic
non-ferrous debris, which is also critical to judge the condition of debris particle is present in the magnetic field (Fig. 4(a)), an eddy
machines with the nonferrous coating. current is induced inside the debris to oppose the existing magnetic
field; this decreases the total magnetic flux, leading to a decrease of the
2.1.2. Wear debris sensors based on capacitive method coil's equivalent inductance. On the other hand, if a ferrous conductive
In order to detect each individual wear debris particle, Murali et al. have metallic debris particle is introduced into the coil (Fig. 4(b)), the two
demonstrated the a device for detecting and counting micro metallic debris aforementioned factors work in a competing way. While the high
in nonconductive lubricating oil inspired by Coulter counting principle, an relative permeability increases the magnetic flux, the eddy current
established technique for counting and sizing debris in electrolyte solution causes a decrease of the magnetic flux. At low frequencies, however, the
[11]. The microfluidic device, as shown in Fig. 3, consists of an inlet induced eddy current is small, making the total magnetic flux to be
reservoir, an outlet reservoir, and a single fluidic channel with dimensions dominated by the debris’ magnetic permeability. Thus, the coils’
of 40 µm (H)×100 µm (W)×300 µm (L). A pair of coplanar electrodes with inductance is increased by the passage of a ferrous debris particle. In
a gap of 40 µm are located in the middle of the fluidic channel. When a comparison, the coils’ inductance is decreased by the passage of a
metallic debris particle is passing through the microchannel, a capacitance nonferrous debris particle because of the dominance of the eddy
change is detected owing to the permittivity differences between the wear current effect at higher frequencies. In this way, the ferrous and non-
debris and the lubricating oil. Compared with Gill Sensors and other bulk ferrous debris particles can be differentiated at an appropriate fre-
capacitance sensing methods [12–14], which has a simple structure but a quency, which is important for condition monitoring of rotating and
very large sensing zone, this microfluidic capacitive sensor can measure reciprocating machinery. Air bubble and water droplet do not cause an
wear debris (10–25 µm) individually. observable inductance change because 1) the permeabilities of air

Fig. 3. Schematic of a capacitive oil debris sensor with a microfluidic channel [11]. (a) top view. (b) sectioned side view.

475
X. Zhu et al. Tribology International 109 (2017) 473–484

tive-type oil debris sensors is that they cannot detect non-conductive

wear debris (e.g. ceramic debris from ceramic coatings), which is also
true for the capacitive wear debris sensors.

2.1.4. Wear debris sensors based on acoustic method

Acoustic detection method uses single or dual ultrasonic transdu-
cers to radiate a large amplitude radio frequency pulse and receive the
echoes generated by the debris particle in the flow stream. Debris size
can be obtained based on the echo signal's amplitude; the larger the
amplitude, the larger the debris size. The main advantage of the
acoustic detection method is that it is capable of detecting and
discriminating solid debris from air bubble [23–25]. As shown in
Fig. 4. Illustration of the sensing mechanism for inductive sensors. (a) Magnetic flux Fig. 6(a), when an incident acoustic pulse encounters a debris particle,
(black lines) induced in the solenoid coil is attenuated owing to the eddy current a small echo (a reflection of the incident acoustic pulse) is generated by
generated magnetic flux (red lines) in the non-ferrous debris. (b) For ferrous debris, the debris particle. Because of the differences in reflection coefficient
magnetic flux is enhanced owing to high relative permeability (blue lines) but also (negative for air bubble and positive for solid debris), echoes reflected
attenuated owing to eddy current.
by air bubbles invert the incident pulse while echoes from solid debris
particles remain non-inverted, as shown in Fig. 6(b). Therefore, air
(1.0000037) and water (0.99999992) are similar to oil (~1.0) [15], and bubbles and solid wear debris can be separated by observing whether
2) the electrical conductivities of air and water are approximately 10–21 the maximum positive peak value occurs before the maximum negative
and 10−9 times that of metallic debris; thus, their eddy current effects peak value. However, the drawback of this method is that it is difficult
are negligible. to differentiate metallic debris from dielectric debris because all solid
Several online wear debris sensors have been developed based on debris particles have similar acoustic reflection coefficient (~0.93) [25].
the 3-D solenoid structure, which can detect ferrous debris as small as More importantly, because the acoustic sensing field is non-uniform,
100 µm [16,17]. For example, the MetalSCAN oil debris sensor of the size of a debris particle may be underestimated if it does not pass
GasTOPS has been utilized in the wind turbine gearbox and aero- the focal zone of the ultrasonic pulse sensor.
derivative gas turbine engines. However, the 3-D solenoid structure To overcome the above two drawbacks, Du and Zhe demonstrated a
does not have the ability to detect wear debris below 100 µm (e.g., 20– proof-of-concept wear debris sensor with the integration of an induc-
50 µm), which is critical to provide early warnings for abnormal tive pulse sensor and an ultrasonic pulse sensor in series [25]. The
machine conditions [7]. On the other hand, because of the large ultrasonic pulse sensor was used to count all solid debris particles while
sensing zone, 3-D solenoid structure has a difficulty in detecting the inductive pulse sensor was used to detect only the ferrous and non-
individual wear debris particles. If multiple small debris particles are ferrous debris. Fig. 7 shows the schematic of the integrated sensor. The
present in the sensing area at the same time, they are most likely combination of the inductive and ultrasonic pulses allows the differ-
identified as one large debris particle, which may induce possible false entiation of dielectric, ferrous and nonferrous debris. The inductive
positive of machine failure. In order to solve this problem as well as to sensor is made of 2-layer planar coils wound around a 1 mm diameter
detect smaller debris, Du et al. [18–20] developed a 2-layer planar coil glass tube; the ultrasonic pulse sensor uses a spherically focused
based inductive sensor, as illustrated in Fig. 5(a). The 2-layer planar ultrasonic transducer for generating a focused beam and receiving
coil was assembled around a glass tube (1 mm inner diameter and the reflected acoustic echoes. To minimize the debris sizing error due to
1.2 mm outer diameter) between two glass slides. The use of the 2-layer the non-uniform acoustic field in the fluidic pipe, a unique flow recess
planar coil dramatically reduced the sensing zone; the sensitivity was structure was used to force all debris particles to pass the focal zone, as
also improved comparing to the 3-D solenoid structure with the same shown in Fig. 7.
turns of the coil. Fig. 5(b) illustrates the sensing zone of the 3-D
solenoid and the 2-layer planar coil. Because the 2-layer coil concen- 2.1.5. Wear debris sensors based on optical/imaging methods
trates the magnetic flux to a smaller volume (increasing the magnetic
Debris present in the lubricating oil affects the oil's light transmis-
flux density), this structure can detect iron debris as small as 50 µm [7] sion property. Thus, wear debris can be measured in terms of
However, one limitation is its low throughput: the sensor can only
measuring light transmission. Based on the light transmission mea-
process 3 mL/min with a single channel. To increase the throughput, surement, different techniques, including microscopy and laser scatter-
oil debris sensors with multiple sensing channels have been developed
ing, have been developed [27–29]. Hamilton et al. [26] developed a
by Zhe's research group [21,22]. Frequency/time division multiplexing wear debris image system to detect the wear debris size and shape. As
has been demonstrated effective in measuring signals from multiple
Fig. 8 shows, it consists of a light source, a flow channel, and a high
sensing channels without much increasing the complexity of measure- speed web camera. The light source is used to illuminate the flow
ment circuit. The highest throughput achieved was reported to be
channel; the webcam constantly takes images with a high speed as oil
460 mL/min, which can be potentially increased to 4.1 L/min by using with wear debris passes the light source. These images are then
additional sensing channels. Such a high throughout is suitable for
processed with MATLAB to get individual debris particles’ size and
online machine health monitoring [22]. Another limitation of induc- shape information, which are critical to identify mechanical faults and

Fig. 5. (a) schematic of the 2-layer planar coil inductive sensor and (b) comparisons of sensing zone between 3-D coil and planar coil structure [18].

476
X. Zhu et al. Tribology International 109 (2017) 473–484

Fig. 6. (a) illustration of a single transducer ultrasonic sensor and (b) sensor response for air bubble and solid debris [25].

hardware and efficient image processing algorithms to distinguish wear

debris when debris particles move at a high speed or overlap in the
detection window.

2.2. Viscosity sensors

Viscosity, which indicates the lubricating oil's capability to provide

a sufficient thickness of oil film between two moving surfaces, is
another critical parameter that indicates the status of the lubricant
Fig. 7. Schematic of the inductive-ultrasonic integrated wear debris sensor with a flow and the machine. Changes in lubricant properties due to oil oxidation,
recess structure [25]. fuel contamination, and additive depletion cause an abnormal viscosity
change. The inappropriate viscosity of lubricating oil leads to excessive
wear, unnecessary fatigue or even catastrophic engine failure [30,31].
Dynamic viscosity (η) and kinematic viscosity (ν) are usually used to
define the lubricant viscosity, where dynamic viscosity is the ratio of
the shear stress between the lubricant layers to the transversal gradient
of velocity, and kinematic viscosity is the ratio of dynamic viscosity to
lubricant density [32]. It is worth noting that viscosity is significantly
affected by temperature; viscosity decreases with an increase in
temperature.
The widely used methods for online viscosity measurement can be
divided into three categories based on: 1) sensing body displacement
[33], 2) acoustic [34–38], and 3) vibrational [39,40] methods. Fig. 9(a)
shows a typical online viscosity sensor based on sensing body
displacement. A ferromagnetic piston is placed inside a chamber where
Fig. 8. Diagram of the webcam-based wear debris sensor developed by Hamilton et al.
two electromagnetic coils are located at top and bottom. The two coils
[26].
are switched on and off alternatively; this produces a force causing the
piston moving between the two coils. Lubricant viscosity can be
worn components. Wu et al. also developed a similar system to extract
obtained from the piston's moving time between the two coils. The
the wear debris particles’ material information [28]. However, this
reliability of this sensor, however, is decreased by the great complexity
method has many drawbacks when used for online applications. Firstly,
and the presence of the moving piston. An alternative approach for
lubricating oil may quickly become nontransparent after the operation,
online viscosity measurement is based on acoustic wave. For these
causing difficulties in taking clear pictures and errors in counting and
viscometers, piezoelectric materials are used to generate acoustic waves
sizing wear debris. Secondly, the system demands high performance
by an alternating electrical field [33]. Fig. 9(b) illustrates the simplest

Fig. 9. Schematic of the viscosity sensors based on (a) sensing body movement, and (b) acoustic wave [32].

477
X. Zhu et al. Tribology International 109 (2017) 473–484

yet widespread viscometers based on quartz crystal microbalance 2.3. Water/moisture content sensor
(QCM), consisting of an AT-cut quartz disk (~500 µm thickness) with
two circular electrodes deposited on both of the substrate surfaces. The According to a locomotive engine problem audit, water/moisture
application of an external alternating voltage (1–10 MHz) on the two contamination, typically caused by coolant leakage or water condensa-
electrodes produces an alternating shear deformation of the substrate tion from the ambient environment, is the number one contaminant in
and induces an acoustic wave that propagates perpendicularly to the engine lubrication system [41,42]. There are three different forms that
substrate. The lubricant viscosity exerts an influence on the QCM's water presents in lubricating oil: dissolved, emulsified, and free [43].
resonant frequency by the following equation [35]: Water/moisture in emulsified or free form not only causes metal-on-
metal contact and thus increases wear debris concentration, but also
ρηf03 oxidizes the lubricant quickly, making the machine more susceptible to
∆f 2 =−
π μsρs (1) corrosion and wear [43].
At the present time, real-time water content sensors based on
where ρ is the lubricant density, η is the viscosity, f0 is the QCM's different sensing mechanisms, including resistive method and capaci-
resonant frequency without loading, µs is the crystal shear modulus, ρs tive method, have been exploited widely [44–47]. A moisture sensor
is the crystal density, and Δf is the frequency shift due to the viscous based on ZnO nanomaterial has been reported to sense the humidity of
loading. By measuring the resonant frequency shift, the viscosity can be lubricating oil at different temperatures [44]. The capacitive method
measured. Although QCM viscometer is remarkable for its simple shows its extraordinary advantages, including low costs, wide measure-
structure and thermal stability [32], the top electrode being in contact ment range, and simplicity, etc. Raadnui et al. [46] have developed a
with the lubricant might get corroded by the acid components in low cost capacitive sensor, which consists of two series of parallel poles,
lubricant, decreasing its accuracy and service time. Micro-acoustic for measuring the water content and other pollutants. A single
solid-state viscometers have a good potential to be embedded in electrode capacitive probe (SeCaP) has been used to measure the water
lubricating oil condition monitoring system because the lubricating content from 0% to 100% in lubricating oil while circumventing the
oil's viscosity and density are highly correlated to the oscillatory capacitive short-circuiting problem [47].. Although these sensors have
parameters [36–38]. Heinisch et al. [40] developed an online viscosity simple structures and low costs, they can be easily affected by the
sensor using commercially available tuning forks (75.1 mm (length) by environment, e.g., temperature change or influenced by other con-
13.7 mm (width)). As shown in Fig. 10, the tuning forks, made of steel taminants. To overcome this, Holzki et al. [48] have demonstrated the
and corrosion-resistant gold coating, are sealed in a glass tube. An feasibility of a fiber-optical evanescent-field absorption sensor with a
electromagnet, placed outside the glass tube and close to the end of one radiation wavelength of 2.95 µm to analyze water content in lubricating
prong, is used for generating magnetic force and hence the mechanical oil. Water in lubricant absorbs the transmitted light and thus leads to
excitation. At the end of the other prong, an electrodynamic pick-up an attenuation loss of the received light. A diagram of the sensor setup
coil with a permanent magnet in the center of the coil is placed. When a is shown in Fig. 11. A wire electrode was helically wound around the
sinusoidal voltage with a DC offset is applied, the electromagnet sapphire fiber and a cylindrical counter electrode was placed concen-
attracts the tuning forks periodically. The right prong, as shown in trically at the inner surface of the flow channel in order to improve the
Fig. 10, vibrates along the pick-up coil's axis direction, generating an sensitivity for measuring the water content because of application of
induced voltage in the pick-up coil. Viscosity can thus be measured inhomogeneous fields. A cladding-less sapphire fiber was placed inside
because it affects the vibration of the tuning fork. The tuning fork the glass flow channel; one end of the fiber was connected to an
viscosity sensor is low cost, robust, easy to install and use, and is infrared (IR) source and the other connected to an optical detector to
unaffected by the presence of air bubbles. However, its bulky size and measure the attenuation loss. Although this method can detect water
multi-component structure limit its usage for online measurements. content as low as 150 ppm in real time, the sensor not only consists of
It is worth mentioning here both acoustic and vibrational visc- multiple components but also needs a cooling system for the signal
ometers detect the change in resonant frequency and/or amplitude detector to minimize the noise level, making it difficult for being used
change, and the liquid's viscosity can be calculated from the change of in real applications.
resonance frequency or amplitude. However, a large difference between
the two types of sensors is their resonance frequency ranges: ultrasonic
viscometers have resonance frequencies ranging from 1 MHz to several 2.4. Soot sensors
hundreds of MHz, while vibrational viscometers generally have lower
resonance frequency (1–100 kHz). Although prone to pick up noise Soot particles are microscopic carbonaceous particles that are the
caused by capacitive coupling [39], vibrational viscometers (e.g. tuning product of incomplete combustion of hydrocarbons. Most soot particles
fork) allow individual measurements of mass density and viscosity are absorbed by lubricant and relatively few are exhausted [49,50].
[40]. Within comparison, the QCM resonant viscometers typically Although individual or primary soot particles from diesel combustion
provide a viscosity-density product (as shown in Eq. (1)) [35]. are around 40 nm in diameter, they can agglomerate up to a maximum
of approximately 500 nm because of its colloidal properties, causing an
increase in the viscosity, and eventually leading to a significant increase
of engine wear [51–53]. A high concentration of soot content can also
increase the local acidity level [54], or even worse, cause oil starvation
on tribological surfaces [55].
Currently, however, very little is reported on using online sensors to
directly measure the soot content. This is because soot content

Fig. 10. Schematic of the tuning fork viscosity sensor [40]. Fig. 11. Setup of the evanescent-field fiber sensor developed by Holzki et al. [48].

478
X. Zhu et al. Tribology International 109 (2017) 473–484

Fig. 12. Schematic of the EIS measurement setup for determining soot and diesel contents by Ulrich et al. [56].

Fig. 13. (a) Top view and cross section view of the working electrode, reference electrodes with glass Ag/AgCl, Polymer Ag/AgCl, and bare silver, and (b) schematic drawing of the test
setup [59].

measurement is mainly in terms of monitoring the change in lubri- these electrochemical sensors cannot pinpoint the soot content change
cant's conductivity and/or permittivity. However, because of the over- individually. To solve this problem, Ulrich et al. [56] explored a
lapping problem, i.e. the changes in conductivity and permittivity can combination of electrochemical impedance spectroscopy (EIS) and
be caused by changes of many oil properties, rather than a specific one, multivariate data analysis to simultaneously predict the soot concen-

479
X. Zhu et al. Tribology International 109 (2017) 473–484

tration and diesel in engine oil. EIS, also known as dielectric spectro-
scopy, measures the impedance as a function of frequency and the
dielectric property of the medium [53]. Fig. 12 shows the schematic of
the measurement setup. Two stainless steel electrodes, separated by
two glass slides, are connected to an impedance analyzer. Oil samples
with different soot content (1–4.3%) and diesel content (1–5.1%) were
then added to the space between the two electrodes. The resistance and
reactance were measured at 24 different frequencies ranging from
24 Hz to 600 kHz. After this, partial least squares (PLS), a general-
Fig. 14. Schematic of the CNT oil sensor for total acid number measurement [61].
ization of multiple linear regression method, was used to construct two
models from the reactance and resistance data; one model for predict-
2.6. Aeration sensor
ing soot concentration and the other for diesel concentration. Although
the authors claimed that this method can differentiate soot and diesel
Air can get into the oil circulation system (oil aeration) as
contents, other contaminants in an oil sample, such as water, wear
lubricating oil passes through the bearings and gears [62]. The study
debris, and sulfur etc., may also introduce impedance change, and
shows that the lubricating oil can contain as much as 15% (v/v)
possibly cause large prediction errors.
dispersed air on its way back to the oil sump, where most of the air
bubbles rise and escape from the oil [63]. The degree of aeration is even
severe for high speed engines because of the high air ingestion and low
2.5. Oil acidity/basicity sensor
residual time in the oil sump. Entrained air, however, if not reduced to
1% or less before it gets into the oil pump, leads to a loss of lubrication
Lubricating oil degrades gradually as a result of its exposure to high
in moving components, and increase wear and oil oxidation. Therefore,
temperature, water, air, and acid contaminants (combustion bypro-
there is an increasing demand to monitor the oil aeration rate before oil
ducts such as sulfuric acid, nitric acid) [57], causing oxidation of the
gets into the oil pump.
oil. When engine oil starts to oxidize, alkyl radicals are formed and then
Kimura et al. [64] developed a method to measure both aeration
further transformed into peroxy radicals in the presence of oxygen. The
rate and bubble diameters in lubricating oil using image analysis.
peroxy radicals further react with oil and form hydroperoxides and
Fig. 15(a) shows the mechanism of the developed system. A small
alkyl radicals. As this chain reaction continues, a high concentration of
transparent viewing window was opened on the engine wall. Pictures of
hydroperoxides is formed, which accelerates the oxidation [58].
the lubricating oil inside the viewing window were taken using a metal
Generally, the lubricating oil's total acid number (TAN) increases and
halide light source and a CCD camera. The software was implemented
the total base number (TBN) decreases during use. The increase in
to recognize bubbles and calculate aeration rate automatically. While a
TAN is related to the continued oxidation and acid contamination of
high accuracy has been reported in a laboratory environment, this
the oil while the decrease of TBN is caused by the degradation of
method relies on picture quality. Large errors are likely to be included
antioxidants (lubricant additives) [57]. The acid contamination and
by transparency, light intensity, and camera/engine vibration. DSI-
loss of antioxidants typically cause a loss of performance and eventual
Deltabeam introduced an X-ray absorption method to perform online
failure of the lubricating oil. Therefore, development of an online
aeration measurement [65]. The operation principle, as shown in
sensor capable of monitoring changes of TAN becomes important in
Fig. 15(b), is based on the fact that air has negligible X-ray absorption
ensuring more efficient and safer machine operation. It is worth
in comparison to engine oils. For the fixed volume between the two
mentioning here that traditional glass pH sensors cannot be used here
viewing windows, the aeration can thus be calculated by
because of their slow response and low stability in the hydrocarbon
environment [57]. I
log I
0
Solid state ion selective electrode (ISE), invented over 50 years ago, x= I
has shown promise in sensing pH of the aqueous solutions. Recently, log I1 (2)
0
thick-film and thin-film ruthenium oxide (RuO2) ISEs have been
explored for monitoring engine oil acidity [59,60]. The sensor consists where x is the aeration, I0 and I1 are X-ray intensities measured at full
of a RuO2 electrode as the working electrode and three different types oil (x=0) and full air (x=1) for the fixed volume. This method has been
of reference electrodes using glass-based Ag/AgCl, polymer-based Ag/ demonstrated to have a high accuracy. Additionally, less than 0.5 L of
AgCl, and bare silver conductor, as shown in Fig. 13(a). The working oil is needed for the analysis, which would unlikely cause an oil
electrode senses the H+ concentration change while a stable potential is shortage of machinery on the original machine oil circulation system.
kept on the reference electrodes. The measurement setup is shown in However, the accuracy is also affected by other contaminants present in
Fig. 13(b) where a water bath is used to regulate the oil's temperature the oil, such as soot particles and water, which have different X-ray
at 50 °C and 80 °C, representing engine oil operating temperatures absorption behaviors in comparison to oil. In addition, the system is
within the oil sump. A commercial reference electrode was used as a bulky (60 cm×50 cm×40 cm) and heavy (40 kg), and has high power
comparison to the other three reference electrodes. Although test consumption (1 kW), making it impractical for automobiles and wind
results showed that the RuO2 ISE had a linear response and a fast turbines.
response time to the acid concentration in oils, it had low stability at
high temperature and was sensitive to light and external electromag- 2.7. Grease lubricant condition monitoring
netic field [54]. Moon et al. [61] investigated a resistive sensor utilizing
carbon nanotubes (CNTs) as sensitive films for measuring the oil's Grease lubricants, typically consisting of 85% base oil, 10%
TAN. The sensing mechanism is that change in TAN affects oil's thickener, and 5% additives [66], have been widely used in lubricated
electrochemical reactivity and conductivity. A schematic of the CNT rolling bearings. Grease also suffers from severe physical and chemical
oil sensor is shown in Fig. 14. Using a 3 mm×3 mm CNT sensing film, degradations caused by thermal and mechanical effects, as well as the
the sensor's outputs were correlated to the TAN change in a laboratory presence of wear debris and excess water content [67]. However, grease
environment. However, this sensor was unable to accurately predict condition cannot be monitored by most of the online sensing methods
TAN when being used in actual oils because other oil properties such as developed for lubricating oil because 1) grease does not flow uniformly
water content or soot content also had an influence on the conductivity or circulate in bearings due to its non-flow, semi-solid property, and 2)
change. it is difficult to install sensors and have them function in the bearing

480
X. Zhu et al. Tribology International 109 (2017) 473–484

Fig. 15. Illustrations of the aeration sensors based on (a) image analysis [64] and (b) X-ray absorption [65].

chambers with harsh operation conditions (e.g., high pressure, high 3. Integrated oil properties sensors
temperature, large shocks etc. [68,69].) in order to monitor the grease
condition. To date, offline methods including wet chemical analysis and Lubricating oil degradation is a very complex and complicated
spectroscopic analysis still remain dominant in analyzing the grease process and it is inadequate to assess its condition just monitoring a
properties. For example, Cann et al. [70] used Fourier transform single property [82], such as viscosity, wear debris concentration,
infrared spectroscopic analysis to measure grease degradation in deep water, and soot contents, etc. In an attempt to provide a comprehensive
groove ball bearings. evaluation of the lubricating oil, a few multi-property sensors were
Recently Schaeffler Group developed a sensor to monitor the online developed. Preethichandra and Shida [83] developed a multifunctional
condition of grease in rolling bearings. Four properties of the grease sensor, as shown in Fig. 17, using a phototransistor to measure oil
(water content, opacity, wear and temperature) [71] were measured. cleanliness, a rotating disk to measure viscosity, and the capacitance
The sensor is based on the near-infrared reflection principle. The between the fixed and rotating aluminum plate to measure oil
sensor was installed on the inner surface of rolling bearings. As shown permittivity. Note it is difficult to detect wear debris and individual
in Fig. 16, near-infrared light is emitted from two light-emitting diodes oil properties using the capacitance measurement methods because the
(LED), reflected by the grease, and is measured by the measurement capacitance is influenced by a combination of soot, wear debris, acid/
detector. At the same time, a reference detector measures the reflected base depletion etc. Additionally, using a rotation part driven by a motor
light that does not have any contact with the grease at the same makes the system bulky and impractical for online measurement.
temperature. From the absorption and transmission changes of the Duchowski and Mannebach developed an oil condition monitoring
near infrared spectrum, the four parameters can be evaluated by device consisting of four separate components: a quartz crystal
comparing the signal from the measurement detector with the signal microbalance (QCM) component for viscosity measurement, two
from the reference detector. However, the sensor head must be capacitive sensors for measuring total acid number (TAN) and moist-
embedded in the grease. It is a challenge to install the sensor in the ure, and a thermo-resistor for temperature measurement [84]. Due to
appropriate position in the bearing as the measurement results are the complex nature of oil chemical properties, the measured capaci-
sensitive to the position of the sensor. Furthermore, this method is tance change or dielectric constant change are affected not only by
likely to have difficulties in differentiating contaminants that have moisture and total acid number, but also by other oil properties (wear
similar absorption characteristics [72]. For example, the infrared debris, soot component, etc.). Hence the measurement may be
absorption is recorded in the 3600–3400 cm−1 range and may not be inaccurate for actual lubricating oil where many contaminants are
used to differentiate water, glycol or antioxidant additives, because present. Additionally, the output of each sensor component has to be
their absorption characteristics are usually similar and overlapping. measured separately. If a large number of sensor components are used,
Vibration analysis in the frequency domain [73] can be used to online implementation of detection electronics would be complex and im-
monitor wear and health conditions of the grease lubricated bearings. practical for online machine health monitoring.
Vibration analysis has been used for machine health monitoring for decades Zhu et al. developed an integrated oil conditioning sensor based on
[74–77]. Bearings generate vibration during operation. When abnormal frequency division multiplexing to monitor wear debris, viscosity, and
wear or damage occurs, the vibration signal differs from standard vibration water/moisture content [21]. The equivalent circuit of the integrated
signal under normal condition. By analyzing the vibration signal in the sensor is shown in Fig. 18. It consisted of eight parallel wear debris
frequency domain, possible defect frequencies could be present and thus sensing channels, where the 2-layer inductive coils were used as
the bearings condition could be obtained. Although vibration analysis has sensing elements, one piezoelectric viscosity sensing element and one
been utilized for years, the method 1) needs installation of multiple capacitive water sensing element. Frequency division multiplexing was
vibration sensors (accelerometers) in several positions around the bearing applied on the sensing elements such that only one signal input and
[78], 2) relies on complicated signal processing and data analysis methods two signal outputs were needed, and the signal-to-noise ratio was also
(e.g. wavelet transform [79], multivariate statistics [80], etc.), and 3) is improved. This sensor is capable of measuring wear debris ( > 50 µm),
insensitive to early signs of abnormal wear or grease starvation, making it viscosity ( > 12.4 cSt), and water content ( > 50 ppm) with a throughput
unsuitable for early fault detection [81].

Fig. 16. Schematic design of the online grease sensor [71]. Fig. 17. Schematic of the multi-property sensor using a rotating disk [84].

481
X. Zhu et al. Tribology International 109 (2017) 473–484

implemented in an engine or vehicle, external engine-induced vibration

can possibly add environmental noise to the measurement results.
Studies have shown that during engine operation, the structural
vibration frequency can be up to 200 Hz [90,91]. This vibration noise
is picked up by the measurement circuits and induce measurement
noises. However, for sensors using DC or AC excitation signals (in the
kHz or MHz), this vibration frequency can be eliminated by Fast
Fourier Transform (FFT) during signal processing. For sensors based
on optical/imaging methods, specialized de-blurring algorithms [92]
and/or vibration compensation techniques [93,94] can be used to
minimize the external vibration influence.

4. Future trends and remarks

For online lubricating oil condition monitoring, single property

Fig. 18. Measurement setup of the integrated oil property sensor by Zhu et al. [21]. sensors like wear debris sensor and viscosity sensor have achieved
Eight wear debris sensing channels were installed to increase the throughput; signal certain success. However, to meet the industrial needs in machine
multiplexing was used to simplify the measurement circuit (only one signal input and two health monitoring, we believe the future trends of researches are:
signal outputs were needed).
1) Development of high sensitivity wear debris and oil property
sensors. Certain machines generate finer debris particles even at
of 200 mL/min. However, only up to 10 wear debris sensing channels abnormal wear conditions. For example, for wind turbines, their
(equivalent to a throughput of 250 mL/min) can be installed; crosstalk abnormal wear begins to generate approximately 10 µm wear debris
between channels would arise if a larger number of channels are used. [95]. Furthermore, in addition to lubricating oil, fuel has been used as a
Additionally, complex post signal processing or cumbersome circuit special lubricant for fuel injection pumps. Common contaminants and
parameter adjustment becomes inevitable for reducing the crosstalk. properties, including wear particle, water content, or abnormal viscos-
It is worth mentioning here that to date most electrochemical ity change in fuel-lubricated systems are greatly smaller than these in
sensors for oil properties have not solved the cross sensitivity problem oil-lubricated systems. Current sensors that can measure debris as
(called overlapping), i.e., the changes in permittivity and conductivity small as 20 µm and viscosity as low as 10 cSt may be insufficient for
respond to changes in many properties including soot content, oxida- wind turbines and fuel-lubricated machines. Thus, development of
tion, fuel dilution, water/moisture content, metallic wear debris, and wear debris sensors, water content sensors, and viscosity sensors with
charged ions, etc., rather than to a specific property. Hence it is difficult higher sensitivity is needed to provide an accurate evaluation of oil or
to accurately determine the value of each individual property with these fuel lubricant conditions. On the other hand, large engines need a
sensors. monitoring system with a throughput larger than 1 L/min. New sensor
To overcome the cross sensitivity problem, Zhu et al. [85] devel- structure and/or advanced measurement circuit (e.g. high- order LCR
oped an electrochemical sensor array and used back propagation resonant circuit) need to be explored to increase the throughput
artificial neural network (BPNN) for detection and quantification of without sacrificing the sensitivity.
four oil properties, namely water content, TAN, soot content, and 2) Development of special oil property sensors that are only
sulfur content. The artificial neural network has been proved effectively sensitive to the designed properties, e.g. soot content, sulfur content,
in processing the responses from sensors that are affected by more than and oil oxidation, etc. Current online technologies for these properties
one factor, and determining the relationship between the sensors and either have a low sensitivity or have a large error rate because of the
each factor [86,87]. The back propagation neural network (BPNN) is a low selectivity [95]. Possible solutions might be the development of a)
widely used artificial neural network in untangling the overlapping of highly selective sensing film that only responds to the designed
sensor responses [88,89]. Fig. 19 shows the design of the microsensor property, and b) online and miniature, portable sensors based on
array. While all the four sensing components respond to all properties, element analysis techniques such as Fourier transform infrared spec-
each sensor responds to the four properties differently. With an troscopy (FTIR).
optimized BPNN, the sensor array can detect contents of water, TAN, 3) Integration of multiple sensors to collect more information
soot, and sulfur at different concentration levels with a maximum regarding oil degradation (e.g. wear debris, water content, viscosity,
prediction error of 18.8%. soot content, TAN, etc.). However, with more and more sensors
To be noted that, if any of the introduced online sensors are installed in the monitoring system, cross sensitivity among sensors
becomes an issue. Effective algorithms or models for data processing
and information extraction to identify damage/failure for a variety of
machines are needed to provide predictive maintenance and prognosis
of the machine's remaining service life. Artificial neural network (ANN)
[86–88] as well as principle component analysis (PCA) [96] becomes
promising to effectively obtain useful information from multiple
sensors because of their abilities in solving the cross sensitivity
problem. Because of complex nature of lubricant in operation, new
algorithm is needed to accurately pinpoint various properties of
lubricant if unknown contaminants are present in the oil.
Additionally, algorithm toward self-training and validation of the
artificial neural network would be of great benefit to the end users.
4) Development of an efficient and fast onboard data collecting and
processing algorithm/circuit associated with the integrated sensors to
Fig. 19. Schematics of the microsensor array developed by Zhu et al. [85]. Sensor 1 realize real time processing. One of the challenges is that raw data size
primarily responds to water, sensor 2 has no preference, sensor 3 primarily responds to from multiple sensors must be reduced. Possible solutions to reduce
soot, and sensor 4 primarily responds to sulfur content. the data size as well as simplify the measurement circuits have been

482
X. Zhu et al. Tribology International 109 (2017) 473–484

demonstrated based on Frequency/Time Division Multiplexing com- based on QPSO-matching pursuit for online wear debris discrimination. Mech Syst
Signal Process 2015;60:301–15.
bined with Fast Fourier Transform (FFT) or Synchronized Sampling [25] Du L, Zhe J. An integrated ultrasonic–inductive pulse sensor for wear debris
[21,22]. detection. Smart Mater Struct 2013;22:25003.
5) Development of wireless sensor networks (WSN) for wind [26] Hamilton A, Cleary A, Quail F. Development of a novel wear detection system for
wind turbine gearboxes. IEEE Sens J 2014;14:465–73.
turbines in remote wind farms, air planes, and marine ships, etc [27] Peng Y, Wu T, Wang S, Peng Z. Oxidation wear monitoring based on the color
[97,98]. For these machines, using cables to connect multiple sensors extraction of on-line wear debris. Wear 2015;332–333:1151–7.
increases the sensor system's overall complexity, cost, and weight. [28] Wu T, Wu H, Du Y, Kwok N, Peng Z. Imaged wear debris separation for on-line
monitoring using gray level and integrated morphological features. Wear
Wireless sensor networks can provide new and retrofitted systems by
2014;316:19–29.
eliminating much of the cabling, interconnects, and ultimately reduce [29] Du L, Zhe J. On-line wear debris detection in lubricating oil for condition based
overall system weight, and tremendously reduce the running cost. WSN health monitoring of rotary machinery. Recent Pat Electr Electron Eng 2011;4:1–9.
[30] Brouwer MD, Gupta LA, Sadeghi F, Peroulis D, Adams D. High temperature
also have the advantage of more tolerance in the presence of water,
dynamic viscosity sensor for engine oil applications. Sens Actuators A Phys
smoke, and fire, etc. However, data transmission reliability, data 2012;173:102–7.
collisions between multiple sensors, and data security from the sensor [31] Agoston A, Ötsch C, Jakoby B. Viscosity sensors for engine oil condition monitoring
networks to the remote station should be considered before cable - application and interpretation of results. Sens Actuators A Phys
2005;121:327–32.
replacement. [32] O’Sullivan CK, Guilbault GG. Commercial quartz crystal microbalances - theory and
applications. Biosens Bioelectron 1999;14:663–70.
Acknowledgement [33] Markova LV, Makarenko VM, Semenyuk MS, Zozulya a P. On-line monitoring of
the viscosity of lubricating oils. J Frict Wear 2010;31:433–42.
[34] Markova LV, Myshkin NK, Kong H, Han HG. On-line acoustic viscometry in oil
This work is partially supported by National Science Foundation via condition monitoring. Tribol Int 2011;44:963–70.
grants CMMI-0968736 and ECCS-1625544. [35] Saluja A, Kalonia DS. Measurement of fluid viscosity at microliter volumes using
quartz impedance analysis. AAPS PharmSciTech 2004;5:e47.
[36] Durdag K. Solid state acoustic wave sensors for real-time in-line measurement of
References oil viscosity. Sens Rev 2008;28:68–73.
[37] D. Billson, Ultrasonic liquid viscosity sensor using mode conversion. United States
patent US 20070101804, 2007.
[1] Yan R, Gao RX. Complexity as a measure for machine health evaluation. IEEE
[38] Jakoby B, Scherer M, Buskies M, Eisenschmid H. An automotive engine oil
Trans Instrum Meas 2004;53:1327–34.
viscosity sensor. IEEE Sens J 2003;3:562–8.
[2] Khan MA, Starr AG. Wear debris: basic features and machine health diagnostics.
[39] Salzer S, Jahns R, Piorra A, Teliban I, Reermann J, Höft M, et al. Tuning fork for
Insight Non-Destr Test Cond Monit 2006;48:470–6.
noise suppression in magnetoelectric sensors. Sens Actuators A Phys
[3] Younus AMD, Yang BS. Intelligent fault diagnosis of rotating machinery using
2016;237:91–5.
infrared thermal image. Expert Syst Appl 2012;39:2082–91.
[40] Heinisch M, Voglhuber-Brunnmaier T, Reichel EK, Dufour I, Jakoby B. Application
[4] García Márquez FP, Tobias AM, Pinar Pérez JM, Papaelias M. Condition mon-
of resonant steel tuning forks with circular and rectangular cross sections for
itoring of wind turbines: techniques and methods. Renew Energy 2012;46:169–78.
precise mass density and viscosity measurements. Sens Actuators A Phys
[5] Kumar M, Shankar Mukherjee P, Mohan Misra N. Advancement and current status
2015;226:163–74.
of wear debris analysis for machine condition monitoring: a review. Ind Lubr Tribol
[41] Mortier RM, Fox MF, Orszulik ST. Chemistry and technology of lubricants, 3rd ed..
2013;65:3–11.
London: Springer; 2010.
[6] Flanagan IM, Jordan JR, Whittington HW. Wear-debris detection and analysis
[42] Lancaster JK. A review of the influence of environmental humidity and water on
techniques for lubricant-based condition monitoring. J Phys E Sci Instrum
friction, lubrication and wear. Tribol Int 1990;23:371–89.
1988;21:1011–6.
[43] P.A. von Guggenberg and J.R. Melcher, An immersible relative saturation moisture
[7] Du L, Zhu X, Han Y, Zhe J. High throughput wear debris detection in lubricants
sensor with application to transformer oil. In: Proceedings of the 3rd international
using a resonance frequency division multiplexed sensor. Tribol Lett
conference on properties and applications of dielectric materials: IEEE; 1991; p. 8–
2013;51:453–60.
11.
[8] Cao W, Dong G, Chen W, Wu J, Xie YB. Multisensor information integration for
[44] Yadav BC, Srivastava R, Dwivedi CD, Pramanik P. Moisture sensor based on ZnO
online wear condition monitoring of diesel engines. Tribol Int 2015;82:68–77.
nanomaterial synthesized through oxalate route. Sens Actuators B Chem
[9] Shen MX, Dong F, Zhang ZX, Meng XK, Peng XD. Effect of abrasive size on friction
2008;131:216–22.
and wear characteristics of nitrile butadiene rubber (NBR) in two-body abrasion.
[45] Balde M, Vena A, Sorli B. Fabrication of porous anodic aluminium oxide layers on
Tribol Int 2016;103:1–11.
paper for humidity sensors. Sens Actuators B Chem 2015;220:829–39.
[10] Controls, Gill Sensors & Controls. Oil debris sensor. 〈https://www.gillsc.com/
[46] Raadnui S, Kleesuwan S. Low-cost condition monitoring sensor for used oil
assets/Uploads/GScondition-CD1599-Iss-5.pdf〉 [accessed 20.12.16].
analysis. Wear 2005;259:1502–6.
[11] Murali S, Xia X, Jagtiani AV, Carletta J, Zhe J. Capacitive Coulter counting:
[47] Schüller RB, Gundersen T, Halleraker M, Engebretsen B. Measurement of water
detection of metal wear particles in lubricant using a microfluidic device. Smart
concentration in oil/water dispersions with a circular single-electrode capacitance
Mater Struct 2009;18:37001.
probe. IEEE Trans Instrum Meas 2004;53:1378–83.
[12] Wen Z, Yin X, Jiang Z. Applications of electrostatic sensor for wear debris detecting
[48] Holzki M, Fouckhardt H, Klotzbücher T. Evanescent-field fiber sensor for the water
in the lubricating oil. J Inst Eng Ser C 2013;94:281–6.
content in lubricating oils with sensitivity increase by dielectrophoresis. Sens
[13] Liu Y, Liu Z, Xie Y, Yao Z. Research on an on-line wear condition monitoring
Actuators A Phys 2012;184:93–7.
system for marine diesel engine. Tribol Int 2000;33:829–35.
[49] C.J. Mueller, W.J. Pitz, L.M. Pickett, G.C. Martin, D.L. Siebers and C.K. Westbrook,
[14] Keller MA, Saba CS. Monitoring of ester base lubricants by dielectric constant. Lubr
Effects of oxygenates on soot processes in DI diesel engines: experiments and
Eng 1989;45:347–51.
numerical simulations. SAE Technical Papers; 2003; 1:1791.
[15] Du L, Zhe J, Carletta J, Veillette R, Choy F. Real-time monitoring of wear debris in
[50] Laget O, Malbec L-M, Kashdan J, Dronniou N, Boissard R, Gastaldi P.
lubrication oil using a microfluidic inductive Coulter counting device. Microfluid
Experimental and numerical investigations on the mechanisms leading to the
Nanofluidics 2010;9:1241–5.
accumulation of particulate matter in lubricant oil. SAE Int J Engines
[16] Du L, Zhe J, Carletta JE, Veillette RJ. Inductive Coulter counting: detection and
2016;9:2016–NaN.
differentiation of metal wear particles in lubricant. Smart Mater Struct
[51] Hu E, Hu X, Liu T, Fang L, Dearn KD, Xu H. The role of soot particles in the
2010;19:57001.
tribological behavior of engine lubricating oils. Wear 2013;304:152–61.
[17] Gastops Long Live Equipment. MetalSCAN MS4000. 〈http://www.gastops.com/
[52] Storey JM, Curran S, Dempsey AB, Lewis SA, Sr, Reitz R. Investigation of the
wp-content/uploads/2016/09/C008850_001.pdf〉 [accessed 20.12.16].
potential impact of lubricant contribution to the formation of particulate matter
[18] Du L, Zhe J. A high throughput inductive pulse sensor for online oil debris
with reactivity controlled compression ignition on light-duty diesel engines. Emiss
monitoring. Tribol Int 2011;44:175–9.
Control Sci Technol 2014:1.
[19] Du L, Zhe J. Parallel sensing of metallic wear debris in lubricants using under-
[53] George S, Balla S, Gautam M. Effect of diesel soot contaminated oil on engine wear.
sampling data processing. Tribol Int 2012;53:28–34.
Wear 2007;262:1113–22.
[20] Du L, Zhu X, Han Y, Zhao L, Zhe J. Improving sensitivity of an inductive pulse
[54] Clague AD, Donnet J, Wang T, Peng JC. A comparison of diesel engine soot with
sensor for detection of metallic wear debris in lubricants using parallel LC
carbon black. Carbon 1999;37:1553–65.
resonance method. Meas Sci Technol 2013;24:75106.
[55] Jun HB, Kiritsis D, Gambera M, Xirouchakis P. Predictive algorithm to determine
[21] Zhu X, Du L, Zhe J. An integrated lubricant oil conditioning sensor using signal
the suitable time to change automotive engine oil. Comput Ind Eng
multiplexing. J Micromech Microeng 2015;25:15006.
2006;51:671–83.
[22] Zhu X, Du L, Zhe J. A 3×3 wear debris sensor array for real time lubricant oil
[56] Ulrich C, Petersson H, Sundgren H, Björefors F, Krantz-Rülcker C. Simultaneous
conditioning monitoring using synchronized sampling. Mech Syst Signal Process
estimation of soot and diesel contamination in engine oil using electrochemical
2017;83:296–304.
impedance spectroscopy. Sens Actuators B Chem 2007;127:613–8.
[23] Nemarich CP, Whitesel HK, Sarkady A. On-line wear particle monitoring based on
[57] Smiechowski MF, Lvovich VF. Iridium oxide sensors for acidity and basicity
ultrasonic detection and discrimination. Mater Eval 1988;50:525–30.
detection in industrial lubricants. Sens Actuators B Chem 2003;96:261–7.
[24] Xu C, Zhang P, Wang H, Li Y, Lv C. Ultrasonic echo waveshape features extraction

483
X. Zhu et al. Tribology International 109 (2017) 473–484

[58] Blaine S, Savage PEP. Reaction pathways in lubricant degradation. 3. Reaction conference on advanced computer control (ICACC 2010) conference; 2010;
model for n-hexadecane autoxidation. Ind Eng Chem Res 1992;31:69–75. 192:441–4.
[59] Soleimani M, Sophocleous M, Wang L, Atkinson J, Hosier IL, Vaughan AS, et al. [80] Baydar N, Chen Q, Ball A, Kruger U. Detection of incipient tooth defect in helical
Base oil oxidation detection using novel chemical sensors and impedance spec- gears using multivariate statistics. Mech Syst Signal Process 2001;15:303–21.
troscopy measurements. Sens Actuators B Chem 2014;199:247–58. [81] Raj AS, Murali N. Early classification of bearing faults using morphological
[60] Soleimani M, Sophocleous M, Glanc M, Atkinson J, Wang L, Wood RJK, et al. operators and fuzzy inference. IEEE Trans Ind Electron 2013;60:567–74.
Engine oil acidity detection using solid state ion selective electrodes. Tribol Int [82] Wu T, Wu H, Du Y, Peng Z. Progress and trend of sensor technology for on-line oil
2013;65:48–56. monitoring. Sci China Technol Sci 2013;56:2914–26.
[61] Moon S-I, Paek K-K, Lee Y-H, Kim J-K, Kim S-W, Ju B-K. Multiwall carbon [83] D.M.G. Preethichandra and K. Shida, Actual condition monitoring of engine oil
nanotube sensor for monitoring engine oil degradation. Electrochem Solid-State through an intelligent multi-functional sensing approach. In: Proceedings of the
Lett 2006;9:H78. 26th annual conference of the IEEE industrial electronics society, 2000. IECON:
[62] L. Ippoliti, J. Steimes and P. Hendrick, Investigation on an oil aeration measure- 2000; p. 2383–7.
ment technique for the study of pump performance in an aircraft engine lubrication [84] Duchowski JK, Mannebach H. A novel approach to predictive maintenance: a
system. Vol. 5C Heat Transfer. ASME; 2015, p. V05CT15A016. portable, multi-component MEMS sensor for on-line monitoring of fluid condition
[63] Fowle TI. Aeration in lubricating oils. Tribol Int 1981;14:151–7. in hydraulic and lubricating systems. Tribol Trans 2006;49:545–53.
[64] Kimura Y, Kato N, Kataoka T, Takahashi K, Kikuchi T. Bubble behavior in engine [85] Zhu X, Du L, Liu B, Zhe J. A microsensor array for quantification of lubricant
lubricant. Int J Automot Eng 2011;2:149–53. contaminants using a back propagation artificial neural network. J Micromech
[65] Delta Services Industrials. Lubricant aeration. 〈http://www.deltabeam.net/en/ Microeng 2016;26:65005.
product/lubricant-aeration〉 [accessed 20.12.16]. [86] Sadegh H, Mehdi AN, Mehdi A. Classification of acoustic emission signals
[66] Miettinen J, Andersson P. Acoustic emission of rolling bearings lubricated with generated from journal bearing at different lubrication conditions based on wavelet
contaminated grease. Tribol Int 2000;33:777–87. analysis in combination with artificial neural network and genetic algorithm. Tribol
[67] Tandon N, Ramakrishna KM, Yadava GS. Condition monitoring of electric motor Int 2016;95:426–34.
ball bearings for the detection of grease contaminants. Tribol Int 2007;40:29–36. [87] Rai A, Upadhyay SH. A review on signal processing techniques utilized in the fault
[68] Zhao M, Lin J, Xu X, Li X. Multi-fault detection of rolling element bearings under diagnosis of rolling element bearings. Tribol Int 2016;96:289–306.
harsh working condition using imf-based adaptive envelope order analysis. Sensors [88] Mimendia A, Gutiérrez JM, Alcañiz JM, del Valle M. Discrimination of soils and
2014;14:20320–46. assessment of soil fertility using information from an ion selective electrodes array
[69] He Y, Zhao Z, Luo T, Lu X, Luo J. Failure analysis of journal bearing used in and artificial neural networks. Clean Soil Air Water 2014;42:1808–15.
turboset of a power plant. Mater Des 2013;52:923–31. [89] Kotzalas MN, Doll GL. Tribological advancements for reliable wind turbine
[70] Cann PM, Doner JP, Webster MN, Wikstrom V. Grease degradation in rolling performance. Philos Trans A Math Phys Eng Sci 2010;368:4829–50.
element bearings. Tribol Trans 2001;44:399–404. [90] Seshadrinath J, Singh B, Panigrahi BK. Vibration analysis based interturn fault
[71] Industrial Technology. Online condition monitoring of grease in rolling bearings. diagnosis in induction machines. IEEE Trans Ind Inform 2014;10:340–50.
〈http://www.industrialtechnology.co.uk/products–online-condition-monitoring- [91] Bohn C, Cortabarria A, Härtel V, Kowalczyk K. Active control of engine-induced
of-grease-in-rolling-bearings.html〉 [accessed 20.12.16]. vibrations in automotive vehicles using disturbance observer gain scheduling.
[72] Aranzabe A, Aranzabe E, Marcaide A, Ferret R, Terradillos J, Ameye J, et al. Control Eng Pract 2004;12:1029–39. http://dx.doi.org/10.1016/j.coneng-
Comparing different analytical techniques to monitor lubricating grease degrada- prac.2003.09.008.
tion. NLGI Spokesm 2006;70:17–30. [92] McCarthy DMJ, Chandler JH, Palmeri A. Monitoring dynamic structural tests using
[73] Tondon N, Choudhury A. A review of vibration and acoustics measurement image deblurring techniques. Key Eng Mater 2013;569–570:932–9.
methods for the detection of defects in rolling element bearing. Tribol Int [93] Zhu F, Zhang X, Xia L, Guo Z, Zhang Y. Active compensation method for light
1999;32:469–80. source frequency drifting in Φ -OTDR sensing system. IEEE Photonics Technol Lett
[74] Boškoski P, Petrovičič J, Musizza B, Juriičić D. Detection of lubrication starved 2015;27:2523–6.
bearings in electrical motors by means of vibration analysis. Tribol Int [94] Maret Y, Member S, Angelosante D, Steiger O, Member S. Vibration sensitivity
2010;43:1683–92. reduction of photoacoustic gas analyzers. IEEE Sens J 2014;14:2841–50.
[75] Gutowski P, Leus M. The effect of longitudinal tangential vibrations on friction and [95] Jakoby B, Vellekoop MJ. Physical sensors for liquid properties. IEEE Sens J
driving forces in sliding motion. Tribol Int 2012;55:108–18. 2011;11:3076–85.
[76] Kiral Z, Karagülle H. Simulation and analysis of vibration signals generated by [96] Wang J, Wang X. A wear particle identification method by combining principal
rolling element bearing with defects. Tribol Int 2003;36:667–78. component analysis and grey relational analysis. Wear 2013;304:96–102.
[77] Yoo SS, Kim DE. Effects of vibration frequency and amplitude on friction reduction [97] Singh S, Galar D, Baglee D, Björling SE. Self-maintenance techniques: a smart
and wear characteristics of silicon. Tribol Int 2016;94:198–205. approach towards self-maintenance system. Int J Syst Assur Eng Manag
[78] Sheng S, Zhang L, Gao RX. A systematic sensor-placement strategy for enhanced 2014;5:75–83.
defect detection in rolling bearings. IEEE Sens J 2006;6:1346–54. [98] J. Cho and S. Park, Capacitive sensor for automotive engine oil degradation using
[79] Y. Cai, Y. He, A. Li, J. Zhao and T. Wang, Application of wavelet to gearbox wireless network. In: Proceedings of the 2010 international symposium on
vibration signals for fault detection. In: Proceedings of the 2nd international advanced packaging materials: microtech. 2010; p. 88–91.

484