sensors

Article
Smart Sensor-Based Monitoring Technology for Machinery Fault
Detection
Ming Zhang 1 , Xing Xing 2 and Wilson Wang 2, *

1 Automotive Engineering Department, Weifang Institute of Engineering, Qingzhou 262501, China;

ming-zhang@emechsys.com
2 Department of Mechanical and Mechatronics Engineering, Lakehead University,
Thunder Bay, ON P7B 5E1, Canada; xxing2@lakeheadu.ca
* Correspondence: wilson.wang@lakeheadu.ca

Abstract: Rotary machines commonly use rolling element bearings to support rotation of the shafts.
Most machine performance imperfections are related to bearing defects. Thus, reliable bearing
condition monitoring systems are critically needed in industries to provide early warning of bearing
fault so as to prevent machine performance degradation and reduce maintenance costs. The objective
of this paper is to develop a smart monitoring system for real-time bearing fault detection and
diagnostics. Firstly, a smart sensor-based data acquisition (DAQ) system is developed for wireless
vibration signal collection. Secondly, a modified variational mode decomposition (MVMD) technique
is proposed for nonstationary signal analysis and bearing fault detection. The proposed MVMD
technique has several processing steps: (1) the signal is decomposed into a series of intrinsic mode
functions (IMFs); (2) a correlation kurtosis method is suggested to choose the most representative
IMFs and construct the analytical signal; (3) envelope spectrum analysis is performed to identify the
representative features and to predict bearing fault. The effectiveness of the developed smart sensor
DAQ system and the proposed MVMD technique is examined by systematic experimental tests.

Keywords: smart sensors; data acquisition; bearing fault detection; vibration signal analysis; variational
mode decomposition

Citation: Zhang, M.; Xing, X.; Wang, 1. Introduction

W. Smart Sensor-Based Monitoring Rotating machines are commonly used in various industries, such as electric vehicles,
Technology for Machinery Fault aircraft, and machine tools. Machine component defects will influence machine perfor-
Detection. Sensors 2024, 24, 2470. mance quality and reliability. Based on this investigation, up to 75% of imperfections in
https://doi.org/10.3390/s24082470 small- and medium-size rotating machines are related to defects in rolling element bear-
Academic Editor: Andrea Cataldo ings [1,2]. This work will focus on condition monitoring and fault diagnosis of rolling
element bearings. Bearing defects can not only generate extra vibrations and noise, but
Received: 1 March 2024 also result in the degradation of machine operation accuracy or even the early breakdown
Revised: 5 April 2024
of rotating machines. Thus, reliable machine condition monitoring systems are critically
Accepted: 9 April 2024
needed in industries to detect the bearing fault in its early stage so as to prevent machine
Published: 12 April 2024
performance degradation and improve operation efficiency and reliability.
A real-time machine condition monitoring system includes modules, such as data
acquisition (DAQ) to collect signals, and signal processing for fault detection and diagnosis.
Copyright: © 2024 by the authors.
Vibration signals will be used for bearing fault detection in this work due to its high signal-
Licensee MDPI, Basel, Switzerland. to-noise ratio. A traditional DAQ system uses a DAQ board to collect signals, which is not
This article is an open access article convenient in real industrial applications due to reasons such as the use of pre-conditioning
distributed under the terms and hardware, sensor cable installation limitations, and high expenses. An alternative is to use
conditions of the Creative Commons smart sensor-based DAQ systems. A smart sensor integrates a sensing unit, pre-processing
Attribution (CC BY) license (https:// functionality, and wireless communication in one chip, controlled by a microcontroller unit
creativecommons.org/licenses/by/ (MCU) [3]. Although there are some smart vibration sensors available on the market [3–5],
4.0/). they cannot meet the comprehensive requirements of measurement range, frequency

Sensors 2024, 24, 2470. https://doi.org/10.3390/s24082470 https://www.mdpi.com/journal/sensors

Sensors 2024, 24, 2470 2 of 16

bandwidth and sensitivity, used in this monitoring application. Subsequently, the first
objective of this work is to develop a new smart sensor-based DAQ system to collect
vibration signals wirelessly.
Reliable fault detection in rolling element bearings remains a very challenging task
in the research field of machine health monitoring and fault diagnosis. A bearing is not a
component like a gear or shaft, but it is a system comprising components, such as an outer
ring, an inner ring, a cage, and a set of rolling elements. Bearing defect-related feature
properties depend on bearing geometry, defect location, and operating conditions. In
addition, bearing features are also modulated by other strong vibration signals generated
from vibratory sources such as gear meshing [1,6].
There are many signal processing techniques proposed in the literature for bearing
fault detection [7–10]. The most commonly used approach is based on Fourier transform
(FT) spectral analysis by examining health-related characteristic frequency components on
the spectrums. However, FT analysis cannot be applied to analyze nonstationary signals
whose spectral and statistical properties change with time [11,12]. Nonstationary and
transient signals can be processed by analyzing multiresolution time–frequency information
using techniques such as the Hilbert–Huang transform (HHT) [13]. The HHT uses empirical
mode decomposition (EMD) to extract intrinsic mode functions and residual signature for
signal property analysis; however, its signal decomposition approximation is sensitive to
noise, which can affect the overall processing accuracy [14,15].
The variational mode decomposition (VMD) method decomposes the vibration signal
into several signatures with different center frequencies by using a set of adaptive Wiener
filters [16]. Although VMD has been used in the extraction of nonlinear features and
machine fault detection [17,18], it still has some limitations. For example, it requires pre-
choosing the number of modes and the bandwidth control parameters. Most of the available
research in this area has focused on adaptability improvement and parameter optimization.
For example, a prediction method was proposed in [19] for decomposition prediction
by recursively modifying the related system parameters; however, it was an empirical
algorithm and was difficult to implement in bearing fault detection, as different bearings
have different system parameters and installation dynamics. A spectrum distribution-based
VMD was presented in [20] to detect wheel set-bearing faults; however, its processing
accuracy could degrade due to possible over-decomposition. A particle swarm method
was suggested in [21] to optimize the VMD parameters, but the ratio errors with respect to
the average value and the variance could decrease impact component accuracy. In addition,
some other optimization algorithms were adopted in [22,23] to update the number of
modes in VMD analysis for some specific applications. However, these mode selection
algorithms still have some problems, such as slow convergence of parameter optimization
due to complex spectrum distribution and trapping due to local minima.
To tackle the above limitations in VMD analysis, the second objective of this work
is to propose a new VMD method, called modified VMD or MVMD, for nonlinear signal
analysis and bearing fault detection. It is a great extension of the related work of the
authors’ research team. For example, in comparison with the smart sensors in [4,5], the
smart sensors developed in this work have a wider bandwidth and higher resolution.
On the other hand, the adaptive VMD method in [6] applies the first IMF for bearing
fault detection. Different from our previous related work, the novelties of the proposed
MVMD technique include: (1) a new smart sensor-based DAQ system is developed for
wireless vibration data collection, which has a wider measurement bandwidth and higher
processing resolution accuracy; (2) in the proposed MVMD technique, a new correlation
kurtosis approach is suggested to select the three most representative IMFs and formulate
a new analytical signal. An envelope spectrum analysis is performed to recognize the
representative features for bearing fault detection. The effectiveness of the developed smart
sensor DAQ systems and the MVMD technique is verified by systematic experimental tests.
Sensors 2024, 24, x FOR PEER REVIEW 3 of 17

smart sensor DAQ systems and the MVMD technique is verified by systematic
experimental
smart sensortests.
DAQ systems and the MVMD technique is verified by systematic
Sensors 2024, 24, 2470 The remainder
experimental tests. of the paper is organized as follows: The developed smart3 ofsensor 16
DAQ system is described
The remainder of theinpaper
Section 2. The proposed
is organized MVMD
as follows: technique smart
The developed is discussed
sensor in
Section 3. Theis effectiveness
DAQ system of the 2.MVMD
described in Section technique
The proposed MVMDis examined
technique isindiscussed
Section in
4 by
The remainder
Section 3. The
experimental of the paper is organized as follows: The developed smart sensor
effectiveness of the MVMD technique is examined in Section 4 by
tests. DAQ
system is described
experimental tests. in Section 2. The proposed MVMD technique is discussed in Section 3.
The effectiveness of the MVMD technique is examined in Section 4 by experimental tests.
2. Development of Smart Sensor-Based DAQ System
2. Development of Smart Sensor-Based DAQ System
A DAQ system
2. Development is used
of Smart to collect signals
Sensor-Based DAQ in different forms. A smart sensor node is a
System
A
systemAthatDAQ system is used to collect signals in different forms. A smart sensor node is a
DAQintegrates the sensing
system is used to collectunits andinsignal
signals conditioning
different forms. A smart functions
sensor (e.g.,
node ispower
a
system
supplies, that integrates
amplification, the sensing units and signal conditioning functions (e.g., power
system that integrates theadjustment,
sensing unitsfiltering,
and signaletc.) for datafunctions
conditioning collection [3].power
(e.g., The sup-
related
supplies, amplification, adjustment, filtering, etc.) for data collection [3]. The related
functions are controlled
plies, amplification, by a microcontroller
adjustment, filtering, etc.) for unit (MCU) through
data collection a communications
[3]. The related functions
functions are controlled by a microcontroller unit (MCU) through a communications
are controlled
interface by a microcontroller
[4,5]. Wireless smart sensors unit are
(MCU) throughina industrial
preferred communications interface
applications to[4,5].
reduce
interface [4,5]. Wireless smart sensors are preferred in industrial applications to reduce
Wireless
costs and smart
prevent sensors
DAQarecable preferred in industrial
distribution applications
problems. to reduce costs
The architecture of and
the prevent
developed
costs and
DAQ cableprevent DAQ problems.
distribution cable distribution
The in problems.
architecture of The
the architecturesmart
developed of the developed
sensor DAQ
smart
smart
sensor
sensor
DAQ
DAQ
system
system
is
is
illustrated
illustrated in
Figure1.1.The
Figure
The signals
signals are
are measured
measured
using
using
proper
proper
system
sensing is illustrated
units. After in Figure
proper 1. The signals
pre-conditioning are measured
operations, using
signals proper
are sensing units.
sensing
After units.pre-conditioning
proper After proper pre-conditioning
operations, operations,
signals are signals
digitized by are
the digitized by by
digitized
analog-to-digitalthe the
analog-to-digital
analog-to-digital conversion (ADC) unit integrated
integratedininthe theMCU.
MCU.The The digital vibration
conversion (ADC)conversion
unit integrated (ADC) unit
in the MCU. The digital vibration signalsdigital vibration
are transmitted
signals
signals are transmitted
are transmitted to the receiver wirelessly, and then stored and processed in the
to the receiver wirelessly,toand thethen
receiver
storedwirelessly, and then
and processed in thestored
computerand for
processed
machineinfault
the
computer
computerand for machine
for machine fault detection and diagnosis.
fault detection and diagnosis.
detection diagnosis.

Battery
Battery

Machinery
Machinery Signal
Signal MCU
MCUand
and Wireless
Wireless
SensingUnit
Sensing Unit
System
System Conditioning
Conditioning ADC
ADC Transceiver
Transceiver

The Smart
The Smart Sensor
SensorNode
Node

Wireless
Computer Receiver Wireless
Computer Receiver Transceiver
Transceiver

Figure 1. Block diagram of the smart sensor-based DAQ system.

Figure
Figure1.1.Block
Blockdiagram
diagram of
of the
the smart sensor-basedDAQ
smart sensor-based DAQsystem.
system.
2.1. Sensing Unit
2.1.Sensing
2.1. SensingUnit
Unit
The ICP (Integrated Circuit Piezoelectric) piezoelectric accelerometer is used as the
TheICP
The
sensing ICPin
unit (Integrated
(Integrated Circuit HA1N100
Circuit
this work (model Piezoelectric)
frompiezoelectric
Piezoelectric) piezoelectricaccelerometer is used
accelerometer
Hangjia Technology, Nanjing, as the
isChina).
used as
Its the
sensing
sensing unit in
unitrange this
in this work (model HA1N100 from Hangjia Technology, Nanjing, China). Its
frequency is work (model
over [0.5, HA1N100
10,000] Hz, withfrom Hangjiaof
a sensitivity Technology, Nanjing,
100 mV/g, and China). Its
measurement
frequency range is over [0.5, 10,000] Hz, with a sensitivity of 100 mV/g, and measurement
frequency
range of range
±50 g. is over
This [0.5, 10,000]
sensor requiresHz,
an with a sensitivity
excitation of 100
DC voltage mV/g,
over [18 and
V, 28measurement
V] and a
range of ±50 g. This sensor requires an excitation DC voltage over [18 V, 28 V] and a
constant
range current
of ±50 source
g. This over requires
sensor [2 mA, 20anmA].
excitation DC voltage over [18 V, 28 V] and a
constant current source over [2 mA, 20 mA].
constant current source over [2 mA, 20 mA].
2.2. Signal
2.2. Signal Conditioning
Conditioning Circuitry
Circuitry
2.2. Signal
The Conditioning
The signal Circuitry
signal conditioning
conditioning circuitry is
circuitry is used
used toto support
support the the sensor
sensor requirements
requirements andand
process the
The the
process analog
signal data. The
conditioning
analog smart sensor
circuitry
data. The smart node
is used
sensor node to has a voltage
hassupport
a voltage regulated
the output, a
sensor requirements
regulated boost
output, a boostand
converter,
process the aaanalog
converter, constant
constant current
current
data. source,
The source, anti-aliasing
anti-aliasing
smart sensor filters,
filters,
node has as
as well
well as
a voltage as other
other related
regulatedrelated circuits.
circuits.
output, a boost
Figure 2 shows the block diagram of the developed signal conditioning
Figure 2 shows the block diagram of the developed signal conditioning circuitry. circuitry.
converter, a constant current source, anti-aliasing filters, as well as other related circuits.
Figure 2 shows the block diagram of the developed signal conditioning circuitry.
Voltage DC Boost Constant Biasing
Filter Circuits
Regulator Converter Current Source Circuit
Voltage DC Boost Constant Biasing
Filter Circuits
Regulator
Figure Converter
2. Block diagram Current circuitry
of signal conditioning Source for
circuitry for aa smart
smart vibration Circuit
vibration sensor.
sensor.

Figure The
2. Block diagramsmart
developed of signal conditioning
sensor node is acircuitry for a smartunit;
battery-powered vibration sensor.
its input voltage may
vary with time, which can cause interruptions in operation. To prevent this problem, a
constant voltage supply is needed. A voltage regulator (LD1117, from ST Microelectronics,
Aliso Viejo, CA, USA) is selected to provide a constant 3.3 VDC voltage to the circuit [24].
 The developed smart sensor node is a battery-powered unit; its input voltage may
Sensors 2024, 24, 2470
vary with time, which can cause interruptions in operation. To prevent this problem, a
4 of 16
constant voltage supply is needed. A voltage regulator (LD1117, from ST Microelectronics,
Aliso Viejo, CA, USA) is selected to provide a constant 3.3 VDC voltage to the circuit [24].
The LD1117 is an adjustable voltage regulator, which can reduce the influence of the
The LD1117 is an adjustable voltage regulator, which can reduce the influence of the power
power supply fluctuations on the operation of the smart sensor nodes.
supply fluctuations on the operation of the smart sensor nodes.
A boost converter is used to generate the excitation voltage of 18–24 VDC for the
A boost converter is used to generate the excitation voltage of 18–24 VDC for the
piezoelectric accelerometer sensing unit. In this work, LT1930 (from Linear Technology,
piezoelectric accelerometer sensing unit. In this work, LT1930 (from Linear Technology,
Milpitas, CA, USA) is used as the DC boost converter [25]. The LT1930 has a 1.2 MHz
Milpitas, CA, USA) is used as the DC boost converter [25]. The LT1930 has a 1.2 MHz
switching frequency, and it can convert the 3.3 VDC input to a 24 VDC to signal
switching frequency, and it can convert the 3.3 VDC input to a 24 VDC to signal conditioning
conditioning
circuitry circuitryoperation
to maintain to maintain operation
stability of thestability
sensingof the sensing unit.
unit.
A constant
A constant current
current source
source isis required
required to
to provide
provide aa constant
constant excitation
excitation current
current forfor
piezoelectric accelerometers. The LM234 chip (from Texas Instruments, Dallas,
piezoelectric accelerometers. The LM234 chip (from Texas Instruments, Dallas, TX, USA) is TX, USA)
is selected
selected in in
thisthis
workwork [26],
[26], whichwhich
cancan provide
provide a constant
a constant excitation
excitation current
current between
between 2 mA 2
and 20 mA. It has good current regulation and wide dynamic voltage ranges from 1 V to1
mA and 20 mA. It has good current regulation and wide dynamic voltage ranges from
V to
40 V,40 V, with
with an accuracy
an accuracy of ±3%.
of ±3%.
Figure 3a
Figure 3a shows
showsthethedeveloped
developedtwo-layer
two-layerprinted
printed circuit
circuit board
board (PCB)
(PCB) testing
testing board
board for
for the signal conditioning circuitry.
the signal conditioning circuitry.

(a) (b)
Figure 3.
Figure 3. (a)
(a) Signal
Signal conditioning
conditioning circuitry
circuitry on
on aa testing
testing PCB
PCB board
board (two-layer):
(two-layer): (1)
(1) Voltage regulator
Voltage regulator
LD1117; (2) DC boost converter LT1930; (3) Constant current source LM234. (b) The smart
LD1117; (2) DC boost converter LT1930; (3) Constant current source LM234. (b) The smart sensor sensor
prototype: (4) The signal conditioning circuitry; (5) PIC32 MCU; (6) Pickit programmer; (7) CC1101
prototype: (4) The signal conditioning circuitry; (5) PIC32 MCU; (6) Pickit programmer; (7) CC1101
wireless transceiver.
wireless transceiver.

2.3. Microcontroller PIC32MX

2.3. Microcontroller PIC32MX
The MCU used in this smart sensor is PIC32MX250F128B (from Microchip
The MCU used in this smart sensor is PIC32MX250F128B (from Microchip Technology,
Technology, Chandler, AZ, USA) [27]. PIC32 is a 32-bit MCU with 128 kB reprogrammable
Chandler, AZ, USA) [27]. PIC32 is a 32-bit MCU with 128 kB reprogrammable flash memory,
flash memory, 3 kB dedicated boot flash memory, and 32 kB SRAM. It can provide a 20
3 kB dedicated boot flash memory, and 32 kB SRAM. It can provide a 20 kHz sampling
kHz sampling
frequency, whichfrequency,
can meet which can meetrequirement
the bandwidth the bandwidth requirement
in this in thisHz).
work (i.e., 10,000 workPIC32
(i.e.,
10,000
also hasHz). PIC32 also
a relatively lowhas a relatively
current low current
consumption consumption
of 0.5 mA/MHz of 0.5
(versus 1.5mA/MHz (versus
mA at 4 MHz for
1.5 mA at 4 MHz for Atmega328p MCU). In addition, it can support UART, SPI,
Atmega328p MCU). In addition, it can support UART, SPI, and I2C communications protocols. and I2C
communications protocols.
In general, machine condition monitoring is not conducted continuously, but periodi-
cally, for example,machine
In general, every 4 h.condition
The MCUmonitoring
will provideissleep
not mode
conducted
controlcontinuously,
to shut down but
the
periodically,
power systemforin example,
the smartevery 4 nodes
sensor h. The to
MCU
savewill provide
battery sleep
power mode
when no control to shut
data collection
down the is
operation power system in the smart sensor nodes to save battery power when no data
conducted.
collection operation is conducted.
2.4. SRAM
An extra SRAM is needed in the developed wireless sensor DAQ system to store
the digital data after the ADC and before wireless transmission. The 23LC1024 (from
Microchip) SRAM chip is selected as the extra memory in this work. A single SRAM chip
features 131,072 × 8-bit data points, and its memory for 10-bit ADC is 65,536 data points.
If more data storage capacity is required, multiple SRAM chips can be used as slave devices.
 Microchip) SRAM chip is selected as the extra memory in this work. A single SRAM chip
features 131,072 × 8-bit data points, and its memory for 10-bit ADC is 65,536 data points.
If more data storage capacity is required, multiple SRAM chips can be used as slave
Sensors 2024, 24, 2470
devices.
5 of 16

2.5. Wireless Communication

The wireless
2.5. Wireless transmitter chip CC1101 transceiver (from Texas Instruments) [28] is
Communication
usedThe forwireless
wireless communication.
transmitter chip CC1101 CC1101 is a low-power
transceiver (from Texas and sub-1 GHz
Instruments) transmitter
[28] is used
designed
for wirelessfor different short-range
communication. CC1101 frequency bands
is a low-power andatsub-1
315 MHz, 433 MHz, 868
GHz transmitter MHz, and
designed
915different
for MHz. In this work,frequency
short-range a 915 MHz ultra-high
bands frequency
at 315 MHz, band
433 MHz, isMHz,
868 chosen, andwhich
915 MHz.is assigned
In
for industrial, scientific, and medical applications in North America.
this work, a 915 MHz ultra-high frequency band is chosen, which is assigned for industrial, CC1101 can also
provide extensive
scientific, and medical hardware support
applications for functionalities.
in North America. CC1101 can also provide extensive
The receiver
hardware support forcanfunctionalities.
accept the digital data from different sensor nodes, and transmit the
dataTheto areceiver
computingcan accept
unit, the
such digital
as andata from different
analyzer sensor nodes,
or a computer. It canand alsotransmit
control the
the data to aofcomputing
operations unit, such
the smart sensor nodes as an
foranalyzer or a computer.
data acquisition It can
processes, suchalsoascontrol the
data collection
operations
intervals, of the smartfrequency,
sampling sensor nodesdata for data acquisitionrate,
transmission processes,
sleepsuch
mode as data collection
control, etc. The
intervals,
receiver sampling
unit usesfrequency,
the same data
typetransmission
of MCU and rate, sleep mode
wireless control,
module etc. The
as used receiver
in smart sensor
unit uses the same
nodes in this case. type of MCU and wireless module as used in smart sensor nodes
in this case.
The maximum transmission distance between a smart sensor node and the receiver
The maximum transmission distance between a smart sensor node and the receiver
depends on factors such as obstacles and noise interference. The maximum distance for
depends on factors such as obstacles and noise interference. The maximum distance for
using the C1101 ZigBee wireless communication protocol with 915 MHz is up to 500 m in
using the C1101 ZigBee wireless communication protocol with 915 MHz is up to 500 m in
anopen
an openspace.
space.Figure
Figure4 4shows
shows thethe smart
smart sensor
sensor node
node (a primary
(a primary prototype
prototype for testing).
for testing).

Figure4.4.Experimental
Figure Experimental setup
setup for testing
for testing the smart
the smart sensors:
sensors: (1) Oscilloscope,
(1) Oscilloscope, (2) Tested(2) Tested two-layer
two-layer PCB
board, (3) Power amplifier, (4) Frequency generator, (5) Vibration sensing unit, (6) Excitation(6)
PCB board, (3) Power amplifier, (4) Frequency generator, (5) Vibration sensing unit, Excitation
shaker.
shaker.
2.6. Performance Evaluation of the Smart Sensors
2.6. Performance Evaluation of the Smart Sensors
To evaluate the accuracy of the developed smart vibration sensor nodes, systematic
Toconducted
tests are evaluate the accuracy
using of the developed
the experimental smart vibration
setup, as shown in Figure 4.sensor nodes,
The signal systematic
generator
tests areanconducted
provides using with
excitation signal the experimental setup, After
a specified frequency. as shown in Figure
amplification, the4.signal
The issignal
generator
fed provides
to the shaker an excitation
to generate signal with
the vibration a specified
signal with the frequency. After amplification,
specified frequency and mode the
used
signal inisthe
fedtesting. Figureto
to the shaker 5 shows
generatesome
the spectrum
vibration of the collected
signal vibration frequency
with the specified signals at and
30 Hz, 60
mode usedHz,in2.3 kHz,
the and 3.7
testing. kHz,5respectively.
Figure shows someThe extra spectral
spectrum of the components are due signals
collected vibration to
noise, for example, generated by the parts on the table and shaker connecting
at 30 Hz, 60 Hz, 2.3 kHz, and 3.7 kHz, respectively. The extra spectral components are due frequencies.
It is seen that input signals can be acquired accurately by the smart sensor node.
Figure 6 shows the final prototype of the developed smart sensor DAQ system with
four-layer PCBs, which has been used successfully in industrial applications.
 to noise,
Sensors 2024, 24, x FOR for example,
PEER REVIEW generated by the parts on the table and shaker connecting
frequencies. It is seen that input signals can be acquired accurately by the smart sensor
node.

300
to noise, for example,
300
generated by the parts on the table and shaker conn
Sensors 2024, 24, 2470 6 ofthe
16 smart s
frequencies. It is seen that input signals can be acquired accurately by

Amplitude (g/Hz)
200 node. 200
(a) (c)
100 300 100 300

Amplitude (g/Hz)
0 200 0 200
0 200
(a) 400 600 800 1000 0 500 (c)1000 1500 2000 2500 3000
300 100 300 100
Amplitude (g/Hz)

200 0 200 0
(b) 0 200 400 600 (d) 800 1000 0 500 1000 1500 2000 2500 3000

100 300 100 300

Amplitude (g/Hz)
0 200 0 200
0 200
(b) 400 600 800 1000 0 1000
(d) 2000 3000 4000
Frequency (Hz) Frequency (Hz)
100 100

Figure 5. Amplitude spectra of collected signals using the smart vibration sensor prototype at
different 0supply
0 frequencies:
200 400 (a)600
30 Hz, (b)
800 60 Hz, (c) 2.3 0K0 Hz, and
1000 (d) 3.7 K2000
1000 Hz. The 3000
arrows specify
4000
the excitation frequencies.
Frequency (Hz) Frequency (Hz)

Figurespectra
Figure 5. Amplitude 5. Amplitude spectra
of collected of collected
signals signals
using the smartusing the smart
vibration sensorvibration sensor
prototype at protot
Figure 6 shows the final frequencies:
prototype of the developed smart sensor DAQ system with
different supply frequencies: (a) 30 Hz, (b) 60 Hz, (c) 2.3 K Hz, and (d) 3.7 K Hz. The arrows specify arrows s
different supply (a) 30 Hz, (b) 60 Hz, (c) 2.3 K Hz, and (d) 3.7 K Hz. The
four-layer PCBs, thewhich hasfrequencies.
excitation been used successfully in industrial applications.
the excitation frequencies.

Figure 6 shows the final prototype of the developed smart sensor DAQ system
four-layer PCBs, which has been used successfully in industrial applications.

Figure 6.6. Developed

Figure Developed wireless
wireless smart
smart vibration
vibration sensor
sensor node
node (four-layer):
(four-layer): (1)
(1) Battery,
Battery, (2)
(2) Wireless
Wireless
transmitter, (3) A smart sensor node, (4) Vibration sensing unit, (5) Receiver.
transmitter, (3) A smart sensor node, (4) Vibration sensing unit, (5) Receiver.

3.3.MVMD
MVMDTechnique
Technique for
Figure 6.for BearingFault
Developed
Bearing FaultDetection
wirelessDetection
smart vibration sensor node (four-layer): (1) Battery, (2) W
transmitter, (3) A smart sensor node, (4) Vibrationorsensing unit,is(5) Receiver. in this
AAnewnewsignal
signalprocessing
processingtechnique,
technique, modified
modified VMD or
VMD MVMD,
MVMD, proposed
is proposed in this
work for bearing fault detection. The MVMD technique consists
work for bearing fault detection. The MVMD technique consists of the following processingof the following
processing 3. MVMD
steps: (1) VMD Technique
analysisfor toBearing
generateFault Detection
steps: (1) VMD analysis to generate intrinsic modeintrinsic
functionmode
(IMFs),function (IMFs),
(2) selecting (2)
most
selecting most A new
representative signal
IMFsprocessing
and technique,
reconstruct the modified
signal, VMD
and (3)
representative IMFs and reconstruct the signal, and (3) envelope spectral analysis for or MVMD,
envelope is proposed
spectral i
analysisfault
bearing work
for bearing for
fault
detection. bearing
detection.
Details fault detection.
Details
will be will be
discussed The MVMD
discussed
in this section. technique
in this section. consists of the follo
processing steps: (1) VMD analysis to generate intrinsic mode function (IMFs
3.1. VMD Processing
selecting most representative IMFs and reconstruct the signal, and (3) envelope sp
VMD is ananalysis
adaptive for bearing fault detection.
time–frequency Details will
analysis method thatbe
candiscussed in this section.
process nonlinear and
nonstationary signals [16]. In processing, the VMD decomposes a signal x (t) into a series
of IMFs, each having a finite bandwidth and located around a center frequency. An IMF,
uk (t), can be expressed as:

uk (t) = Ak (t) cos( φk (t)), k = 1, 2, . . . , K (1)

Sensors 2024, 24, 2470 7 of 16

where Ak (t) is the instantaneous amplitude; φk (t) in rad is the phase angle; K is the number
dφ (t)
of IMFs considered. The instantaneous frequency ωk (t) = dtk in rad/sec.
The VMD aims to solve a constrained variational problem represented by:
( )
K    2
j
min{uk },{ωk } ∑ ∂t δ(t) + ∗ uk (t) e − jωk t
(2)
k =1
πt 2

K
where ∑ uk (t) = x (t); ∂t represents the partial derivative of time; δ = 2π/ωk (t) is the
k =1
Dirac delta distribution; j is the complex number; * is the convolution operator.
Equation (2) can be converted into an unconstrained optimization problem using
the augmented Lagrangian method [6,16]. An optimal solution can be obtained using an
alternated direction method of multipliers [16]. Thus, mode uk can be updated by:

(    2 2
j λ ( t )
unk +1 = arguk ∈ F min αk ∂t δ(t) + ∗ uk (t) e− jωk t + x (t) − ∑ ui (t) + (3)
πt 2 i
2 
2

where αk is a penalty factor and λ(t) is Lagrangian multiplier. Using Parseval/Plancherel

Fourier isometry under the L2 norm and the Hermitian symmetry of the FT, Equation (3)
can be solved in the frequency domain:
 
Z ∞ 2
Λ( ω ) 
Ukn+1 = argUk ,uk ∈ F min 4αk (ω − ωk )2 |Uk (ω )|2 +2 X (ω ) − ∑ Ui (ω ) + dω (4)
 0 i
2 

where uk (t) ↔ Uk (ω ) , x (t) ↔ X (ω ) , λ(t) ↔ Λ(ω ) and the related FT pairs.

This quadratic optimization problem can also be represented as [29]:

Λ(ω )
X (ω ) − ∑ Ui (ω ) + 2
i ̸=k
Ukn+1 (ω ) = (5)
1 + 2αk (ω − ωk )2

In addition, the center frequency ωk of IMF mode uk can be updated by:

(  )
  2
j
ωkn+1 = argωk min ∂t δ(t) + ∗ uk (t) e− jωk t (6)
πt 2

Similarly, the center frequency can also be optimized in the frequency domain by:
Z ∞ 
n +1 2 2
ωk = argωk min (ω − ωk ) |Uk (ω )| dω (7)
0

This quadratic optimization problem can be solved by:

R∞
ω |Uk (ω )|2 dω
ωkn+1 0
= R∞ (8)
0 |Uk (ω )|2 dω

The Lagrangian multiplier can be updated by:

!
Λ n +1
= Λ (ω ) + τ X (ω ) − ∑
n
Ukn+1 (ω ) (9)
k

where τ is the iteration step size (τ = 0 is selected in this case). The iteration will be
terminated if the errors meet the following convergence condition:
Sensors 2024, 24, 2470 8 of 16

2
Ukn+1 − Ukn
∑ 2
2
<ε (10)
k Ukn 2
where ε is the convergence threshold (ε = 0.01 is selected in this case).

3.2. Determination of the Penalty Factor

The penalty factor αk is used to determine the bandwidth of the mode component [16].
In general, a higher penalty factor will generate a narrower bandwidth of the mode com-
ponent, and vice versa. According to the spectral characteristics of different bearing fault
representative features, the penalty factor αk will be empirically estimated by [6]:
 
1 fs
αk =  2 f kc
− 0.5 · (11)
log 2
1 + e 10 f s

where fkc is the center frequency of the k-th mode component; fs is the sampling frequency.
Based on Equation (8), the discrete form of frequency fkc in Hz can be calculated by:

N/2h i
2
∑ n · f s /N · |Uk [n]|
n =1
f kc = (12)
N/2
2
∑ |Uk [n]|
n =1

where N is the length of the vibration signal, and Uk [n] is the discrete FT of the k-th mode
component uk.

3.3. A Correlation Kurtosis Method for IMF Section and Signal Reconstruction
VMD analysis will be used to select the IMFs containing the fault characteristic features.
Some IMFs are more sensitive to machine faults than others. A correlation kurtosis method
is proposed to screen the most sensitive IMFs related to machine faults so as to improve
the accuracy of fault diagnosis, while improving processing efficiency by using fewer
representative IMFs.
Rolling element bearings have periodic impact components in the vibration signal
when a bearing component is damaged. The kurtosis is sensitive to tail modulation in
probability distribution functions but less sensitive to machine operating conditions in
terms of speed and load. When the kurtosis value of an IMF is large, it usually indicates
that this IMF would contain more characteristic features related to bearing defects. The
kurtosis KU,k of the IMF uk in the discrete form, uk [n], can be calculated by:

N
1 4
N ∑ (uk [n] − uk )
n =1
KU,k = (13)
σk4

where σk and uk are the standard deviation and mean of uk [n], n = 1, 2, . . ., N and N is the
length of the IMF signature.
In general, the more IMFs that are used for analysis, the higher the processing accuracy,
but it will take a longer time for processing. To improve processing efficiency, based on a
systematic investigation, the first ten IMFs will be selected in this work for analysis (i.e.,
K = 10), without using the general recursive selection algorithm for K selection [16] so as
to further reduce processing time. However, when a bearing is damaged, the representa-
tive features may time-vary, especially considering slip among bearing components [5,6].
In this work, a new correlation kurtosis method will be applied to rank IMF uk [n] as
described below.
 Sensors 2024, 24, 2470 9 of 16

Cross-correlation is a measure of similarity between two signals even if they have

phase differences from one another. In this work, a normalized cross-correlation is used for
analysis. The correlation indicator Ck is computed by

1 N E ( x − µ x ) uk − µuk



N n∑
Ck = (14)
=1 σx σuk

where x is the input vibration signal; uk is the IMF; µ x and µuk represent the mean values
of x and uk , respectively; σx and σuk are the corresponding standard deviation values; E[.]
denotes the expectation operator.
In the proposed MVMD technique, IMFs are ranked based on the weight factor, Wk ,
associated with both IMF kurtosis and its correlation with the original signal:

Wk = Ck KU,k (15)

If three IMFs with the highest weight factor values are selected to reconstruct the
signal such that:
3
x′ = ∑ W m um (16)
m =1
Wm
where W m = 3 . The analytical signal can be formulated by using the Hilbert
∑ Wm
m =1
transform:
x a = x ′ + jH ( x ′ ) = A x e jφx (17)
where H(.) represents
q the Hilbert transform; φ x = arctan( H ( x ′ )/x ′ )
is the instantaneous
2 2
phase; A x = ( x ′ ) + ( H ( x ′ )) is the instantaneous amplitude (i.e., signal envelope).
The envelope spectrum is obtained by taking the discrete FT of the envelope signal.
Bearing fault detection is undertaken by examining characteristic frequency information
Sensors 2024, 24, x FOR PEER REVIEW 10 of 17
on the envelope spectrum. Figure 7 shows the flowchart of the processing of the MVMD
technique, and the processing procedures are summarized as follows:

Input vibration
signal from DAQ
Compute IMF
weight factor Wk
Initialize Uk , ωk ,
αk , Λ
Select three IMFs
based on Wk
Compute Uk, ωk ,
k =1, 2, …, K
Reconstruct the
original signal
Update λ(t) via
Λ(ω)
Formulate the
analytical signal
Formulate IMFs
Conduct envelope
spectrum analysis
Compute the and bearing fault
penalty factor αk detection

Figure7.7.Flowchart
Figure Flowchartofofthe
theprocessing
processing procedures
procedures of the
of the MVMD
MVMD technique.
technique.

4. Experimental Tests and Data Analysis

The effectiveness of the proposed MVMD technique will be evaluated experimentally
in this section.
 αk , Λ
Select three IMFs
based on Wk
Compute Uk, ωk ,
k =1, 2, …, K
Sensors 2024, 24, 2470 Reconstruct the 10 of 16
original signal
Update λ(t) via
Λ(ω)
Step 1: Initialize {Uk1 }, {ωk1 }, Λ1 , and n := 1.
Formulate the
Step 2: Calculate Uk and ωkanalytical signal (5) and (8), respectively, k = 1, 2, . . ., K, (K = 10).
using Equations
Step 3: Update Λ in Equation (9).
Formulate IMFs
Step 4: Repeat Steps 2–3 until the iteration meets the criterion in Equation (10).
Step 5: Compute IMFs ukConduct , k = 1, envelope
2, . . ., K.
Step 6: Compute the penalty spectrum analysis
factor using Equation (11)
Compute the and bearing fault
Step 7. Compute the weight factor Wk through Equation (15).
penalty factor αk detection
Step 8: Reconstruct the signal using Equation (16).
Step
Figure 9: Formulate
7. Flowchart of thethe analytical
processing signal using
procedures of theEquation (17).
MVMD technique.
Step 10: Conduct envelope spectrum analysis and perform bearing fault detection.
4. Experimental Tests and Data Analysis
4. Experimental Tests and Data Analysis
The effectiveness of the proposed MVMD technique will be evaluated experimentally
The effectiveness of the proposed MVMD technique will be evaluated experimentally
in this section.
in this section.
4.1.Experimental
4.1. ExperimentalSetup
Setup
Theexperimental
The experimentalsetupsetupusedusedin inthis
thispaper
paperisisshown
shownin inFigure
Figure8.8.The
Thesystem
systemisisdriven
driven
by aa 33HP
by HPmotor,
motor,andanditsitsspeed
speedranges
rangesfromfrom100 100toto3600
3600r/min.
r/min. The
The shaft
shaft rotation
rotation isis
controlledby
controlled byusing
usinga speed
a speed controller.
controller. AnAn optical
optical transducer
transducer is used
is used to provide
to provide a one-
a one-pulse-
pulse-per-revolution
per-revolution signal signal to measure
to measure the shaftthe speed.
shaft speed. Tworolling
Two ball ball rolling element
element bearings
bearings are
are fitted in the bearing housings for testing. The developed smart
fitted in the bearing housings for testing. The developed smart sensor node is mountedsensor node is mounted
on one
on oneof ofthe
thebearing
bearinghousings
housingstotomeasure
measurevibration
vibrationsignals.
signals. TheThe collected
collected signals
signals are
are
transmitted wirelessly
transmitted wirelessly to
to the
the receiver
receiver that
that is
is connected
connected to to aa computer
computer viavia aa USB
USBport.
port. To
To
makeaacomparison,
make comparison,vibration
vibrationsignals
signalsarearealso
alsocollected
collectedusing
usingthree
threecommercially
commerciallyavailable
available
ICPaccelerometers
ICP accelerometers mounted
mounted along
along threethree directions
directions on another
on another bearingbearing
housing. housing. The
The signals
signals collected
collected by these byICPthese ICP accelerometers
accelerometers and the
and the optical optical transducer
transducer are computer
are fed to the fed to the
computer
through through
a data a data acquisition
acquisition board (NIfor
board (NI PCI-4472) PCI-4472) for further
further signal signal processing.
processing. The static
The static
loading loading
to the to the
bearings bearings
is applied byisload
applied
disks.by load
The disks. load
dynamic The dynamic
is appliedload
by a is applied
magnetic
by a magnetic
brake brake system
system through through a belt drive.
a belt drive.

Figure8.8.Experimental
Figure Experimentalapparatus
apparatusused
usedin inthis
thiswork:
work:(1)
(1)The
Thedrive
drivemotor;
motor;(2)
(2)An
Anoptical
opticaltransducer
transducer
(3)Smart
(3) Smartvibration
vibrationsensor;
sensor;(4)
(4)Bearing
Bearinghousing;
housing;(5)
(5)Shaft
Shaftmisalignment
misalignmentcontroller;
controller;(6)
(6)Shaft
Shaftsystem
system
platform; (7) Small radial load disc; (8) Heavy radial load disc; (9) ICP accelerometers; (10) Dynamic
load drive unit.

4.2. Performance Evaluation

Deep-groove ball bearings (MB ER-10 K) are tested with four bearing conditions:
healthy/normal bearings, bearings with outer race defect, inner race defect, and rolling
element fault. The tested bearings have the following parameters:
• the number of rolling elements: 8;
• rolling element diameter: 7.938 mm;
• pitch diameter: 33.503 mm;
• the angle of contact: 0 degree.
Sensors 2024, 24, 2470 11 of 16

The tests are undertaken with a shaft speed from 600 rpm to 3600 rpm, with the
load at light, medium, and heavy levels. Some typical processing results corresponding
to shaft rotating frequency fr ≈ 30 Hz (i.e., approximately 1800 rpm) at medium load
level (i.e., 6.5 Nm) are used for illustration. The sampling frequency is 32,000 Hz. The
theoretical characteristic frequencies corresponding to different bearing health conditions
are calculated using the related dynamics equations [1] and are summarized in Table 1. For
comparison, the test results of the proposed MVMD technique will be compared with the
related techniques, that is, the HHT [8] and a self-adaptive VMD as proposed in [23]. All
the techniques are implemented in MATLAB.

Table 1. The bearing fault characteristic frequencies in terms of shaft speed fr .

Characteristic Frequency (Hz)

Healthy bearing fH = fr
Outer race fault fod = 3.052 × fr
Inner race fault fid = 4.947 × fr
Rolling element fault fbd = 3.983 × fr

To quantitatively compare the performance of the related fault detection techniques, a

fault indicator FI is introduced for characterize feature properties:

Ah − µ A
FI = ∑ µA
(18)
h

where Ah is the spectral component amplitude of the h-th harmonic of the characteristic
frequency that can be recognized; µ A is the average spectral value over the bandwidth of
interest (400 Hz in this case).
(1) Processing results for a healthy bearing
Sensors 2024, 24, x FOR PEER REVIEW Figure 9 shows processing result comparison for a healthy bearing using 12 of 17the related
techniques. Its characteristic frequency fH = 30 Hz in this case and the updated MVMD penalty factor
is 932, calculated using Equation (11). The selected IMFs are IMF1, IMF2, and IMF4.

(a)

(b)

(c)

Figure
Figure 9. 9. Performance
Performance comparison
comparison ofrelated
of the the related techniques
techniques for a healthy
for a healthy bearing:
bearing: (a) the(b)
(a) the HHT, HHT, (b) the
the VMD,
VMD, (c)(c)
thethe proposed
proposed MVMD.
MVMD. Arrowsindicate
Arrows indicatecharacteristic
characteristic frequency
frequency (30
(30 Hz)
Hz) and
and itsits harmonics.
harmonics. The arrows indicate bearing characteristic frequency (30 Hz) and its harmonics.
The arrows indicate bearing characteristic frequency (30 Hz) and its harmonics.
(2) Processing results for a bearing with outer race defect
Figure 10 shows the processing results and performance comparison using the
related techniques for a bearing with outer race damage. The characteristic frequency is
fod = 90.9 Hz, and the updated penalty factor is 971 in the MVMD technique. The selected
IMFs are IMF1, IMF2, and IMF3.
 Sensors 2024, 24, 2470 12 of 16

It is seen that the dominant spectral component is the third harmonic of the characteris-
tic frequency (approximately 90 Hz). All these three techniques can recognize this harmonic
clearly. However, the proposed MVMD in Figure 9c can also recognize the fundamental
frequency (30 Hz) and its second harmonic (60 Hz), due to its effective IMF processing and
denoising operations. The VMD in Figure 9b performed better than the HHT in Figure 9a
with the recognized second harmonic (60 Hz), due to its more efficient IMF selection and
processing than the HHT. Based on Equation (18), the fault indicator values are 3.75, 4.22,
and 5.09 for the HHT, VMD, and MVMD techniques, respectively.
(2) Processing results for a bearing with outer race defect
Figure 10 shows the processing results and performance comparison using the related
Sensors 2024, 24, x FOR PEER REVIEWtechniques for a bearing with outer race damage. The characteristic frequency is fod13
= 90.9
of 17
Hz, and the updated penalty factor is 971 in the MVMD technique. The selected IMFs are
IMF1 , IMF2 , and IMF3 .

(a)

(b)

(c)

Figure10.
Figure Performancecomparison
10.Performance comparisonofofthetherelated
relatedtechniques
techniques for
for aa bearing
bearing with
with outer
outer race
race defect:
defect:
(a) the HHT,
(a) the HHT, (b) (b) the VMD, (c) the proposed MVMD. Arrows indicate characteristic frequency
VMD, (c) the proposed MVMD. Arrows indicate characteristic frequency (90.9(90.9
Hz)
Hz)
andand its harmonics.
its harmonics.

In this case,
(3) Processing all three
results for techniques can recognize
a bearing with inner race the characteristic frequency (90.9 Hz)
defect
for the bearing outer race fault. This is because when the bearing outer race (fixed ring)
Figure 11 shows the processing results and performance comparison using the
is damaged, the generated impulses and features are usually time-invariant, which are
related techniques for a bearing with an inner defect. The characteristic frequency is fid =
relatively easy to extract using general fault detection techniques. The VMD in Figure 10b
147.9 Hz. The updated penalty factor in the MVMD is 1210, and the selected IMFs are
performs better than the HHT in Figure 10a, with a clear second harmonic due to its more
IMF 1, IMF3, and IMF4.
efficient IMF processing in signal demodulation. However, the fundamental characteristic
As stated
frequency before, bearing
components (90.9 fault
Hz) detection
using both onthe
a rotating
HHT andinner race
the VMDand contain
a rollingadjacent
element
isspectral
more challenging because the defect-related impulse resonance features
components due to modulation of other vibration signals, which may generate vary with time,
especially considering the slip effects among bearing components. Although
false diagnosis especially in automatic bearing health monitoring. On the other hand, the the HHT in
Figure
proposed11a MVMD
and VMD techniques
technique in Figure
provides 11b
the best can recognize
diagnostic the
result in occurrence
this of a bearing
case, as demonstrated
inner race defect,
in Figure 10c, duetheir
to itscharacteristic
more effective frequency (147.9 Hz)
IMF processing andand/or the second
denoising harmonics
operation. In this
do not dominate the spectral maps. However, the proposed MVMD technique
case, the fault indicator values calculated by Equation (18) are 2.94, 3.30, and 5.71 for the provides
the bestVMD,
HHT, performance
and MVMD in this case, which
techniques, can clearly predict the bearing fault condition,
respectively.
with the domain fault characteristic frequency (147.9 Hz) and its second harmonic as
illustrated in Figure 11c. Its higher resolution and better fault diagnostic accuracy are due
to its feature enhancement effects and efficient IMF feature demodulation. In this case, the
fault indicator values are 0.72, 0.65, and 3.87 for the HHT, VMD and MVMD techniques,
respectively.
Sensors 2024, 24, 2470 13 of 16

(3) Processing results for a bearing with inner race defect

Figure 11 shows the processing results and performance comparison using the related
Sensors 2024, 24, x FOR PEER REVIEWtechniques for a bearing with an inner defect. The characteristic frequency is fid = 14
147.9 Hz.
of 17
The updated penalty factor in the MVMD is 1210, and the selected IMFs are IMF1 , IMF3 ,
and IMF4 .

(a)

(b)

(c)

Figure
Figure11. Performancecomparison
11.Performance comparisonofofthe
the related
related techniques for
for aa bearing
bearingwith
withinner
innerrace
racefault:
fault:(a)
(a)the
the HHT,
HHT, (b) (b)
the the VMD,
VMD, (c) proposed
(c) the the proposed
MVMD.MVMD. Arrows
Arrows indicate
indicate characteristic
characteristic frequency
frequency (147.9 (147.9
Hz) and
Hz) and its harmonics.
its harmonics.

(4) Processing results bearing

As stated before, for a bearing with rolling
fault detection on element
a rotating defect
inner race and a rolling element
is more challenging because the defect-related impulse
Figure 12 shows the processing results and performance comparison resonance features vary with the
using time,
related techniques for a bearing with rolling element damage. The theoreticalin
especially considering the slip effects among bearing components. Although the HHT
Figure 11a and
characteristic VMD techniques
frequency is fbd = 91.57in Hz.
Figure
The11b
MVMDcan recognize the occurrence
has an updated of a bearing
penalty factor 288,
inner
and therace defect,
selected theirare
IMFs characteristic
IMF1, IMF2,frequency
and IMF5. (147.9 Hz) and/or the second harmonics do
not In
dominate
general,the spectral
bearing faultmaps. However,
detection the proposed
in a rolling MVMD
element could betechnique provides the
the most challenging
task because the characteristic features are time-varying. In this case, both the HHT with
best performance in this case, which can clearly predict the bearing fault condition, in
the domain
Figure 12a and fault
VMDcharacteristic
in Figure 12b frequency (147.9
have failed Hz) andthe
to identify its characteristic
second harmonic faultas illustrated
frequency
in Figure
(91.57 Hz) 11c. Its higher
clearly, resolution
even though and better
the HHT faultadiagnostic
performs little betteraccuracy
than theareVMD dueintothis
its feature
case.
enhancement effects and efficient IMF feature demodulation. In
They cannot effectively demodulate the representative features from the collected this case, the fault indicator
values are
vibration 0.72, The
signal. 0.65,MVMD,
and 3.87on forthe
theother
HHT,hand,
VMDisand the MVMD techniques,
only technique respectively.
that can recognize
the
(4) fault characteristic
Processing resultsfrequency (91.57
for a bearing Hz)
with in this
rolling case, as
element shown in Figure 12c, even
defect
though it is not the dominant frequency component on the spectral map. In this case, the
Figure 12 shows the processing results and performance comparison using the related
fault indicator values are 0.39, 0.14, and 0.57 for the HHT, VMD, and MVMD techniques,
techniques for a bearing with rolling element damage. The theoretical characteristic fre-
respectively.
quency is fbd = 91.57 Hz. The MVMD has an updated penalty factor 288, and the selected
In addition, the processing results using the vibration signals collected by the smart
IMFs are IMF1 , IMF2 , and IMF5 .
sensor DAQ system are almost identical with those vibration signals collected by using
In general, bearing fault detection in a rolling element could be the most challenging
the commercial ICP accelerometers (along the vertical direction in this case). It
task because the characteristic features are time-varying. In this case, both the HHT in
demonstrates that the developed smart sensors and DAQ system can be applied reliably
Figure 12a and VMD in Figure 12b have failed to identify the characteristic fault frequency
for data collection.
(91.57 Hz) clearly, even though the HHT performs a little better than the VMD in this case.
They cannot effectively demodulate the representative features from the collected vibration
 Sensors 2024, 24, 2470 14 of 16

signal. The MVMD, on the other hand, is the only technique that can recognize the fault
Sensors 2024, 24, x FOR PEER REVIEWcharacteristic frequency (91.57 Hz) in this case, as shown in Figure 12c, even though
15 of it17is
not the dominant frequency component on the spectral map. In this case, the fault indicator
values are 0.39, 0.14, and 0.57 for the HHT, VMD, and MVMD techniques, respectively.

(a)

(b)

(c)

Figure12.
Figure Performancecomparison
12.Performance comparisonofofthe
therelated
relatedtechniques
techniquesfor
fora abearing
bearingwith
withrolling
rollingelement
element
fault: (a) the HHT, (b) the VMD, (c) the proposed MVMD. Arrows indicate characteristic
fault: (a) the HHT, (b) the VMD, (c) the proposed MVMD. Arrows indicate characteristic frequency frequency
(91.57Hz)
(91.57 Hz)and
andits
itsharmonics.
harmonics.

In addition, the processing results using the vibration signals collected by the smart
5. Conclusions
sensor DAQ system are almost identical with those vibration signals collected by using
A new smart sensor-based monitoring technology has been developed in this work
the commercial ICP accelerometers (along the vertical direction in this case). It demon-
for real-time machine condition monitoring and fault detection. As rolling element
strates that the developed smart sensors and DAQ system can be applied reliably for
bearings are commonly used in rotating machines, and most machine imperfections are
data collection.
related to bearing defects, this work focuses on bearing fault detection. Firstly, a smart
sensor-based
5. Conclusions DAQ system is developed for wireless vibration signal collection. Secondly,
a MVMD technique is proposed for nonstationary signal analysis and bearing fault
A new smart sensor-based monitoring technology has been developed in this work for
detection. The proposed MVMD technique takes several processing steps: (1) the signal is
real-time machine condition monitoring and fault detection. As rolling element bearings
decomposed
are commonly intoused
a series of IMFs;machines,
in rotating (2) a correlation kurtosis
and most machinemethod is suggested
imperfections aretorelated
chooseto
the most representative IMFs and construct the analytical signal; (3) envelope
bearing defects, this work focuses on bearing fault detection. Firstly, a smart sensor-based spectrum
analysis is conducted
DAQ system to identify
is developed the representative
for wireless features
vibration signal for bearing
collection. fault detection.
Secondly, a MVMD
The effectiveness of the developed smart sensor DAQ system and the
technique is proposed for nonstationary signal analysis and bearing fault detection. MVMD techniqueThe
isproposed
examinedMVMD by systematic experimental
technique takes severaltests. Test results
processing show
steps: (1)that
the the developed
signal smart
is decomposed
sensor
into a DAQ system
series of IMFs;can
(2) acollect vibration
correlation signals
kurtosis accurately.
method The proposed
is suggested to choose MVMD
the most
technique outperforms the related techniques under tested bearing
representative IMFs and construct the analytical signal; (3) envelope spectrum conditions. Thisis
analysis
monitoring
conducted technology hasrepresentative
to identify the the potential to be applied
features to machine
for bearing condition The
fault detection. monitoring
effective-
and
nessbearing fault detection
of the developed smartinsensor
industrial
DAQapplications.
system and the MVMD technique is examined by
systematic experimental tests. Test results show that the developed smart sensor DAQ sys-
Author Contributions:
tem can Formal
collect vibration analysis,
signals W.W.; Investigation,
accurately. The proposed M.Z. and X.X.
MVMD All authors
technique have read
outperforms
and agreed to the published version of the manuscript.
the related techniques under tested bearing conditions. This monitoring technology has
the potential
Funding: to be applied
This research tono
received machine
externalcondition
funding. monitoring and bearing fault detection in
industrial applications.
Data Availability Statement: Data are contained within the article.
Conflicts of Interest: The authors declare no conflict of interest.
Sensors 2024, 24, 2470 15 of 16

Author Contributions: Formal analysis, W.W.; Investigation, M.Z. and X.X. All authors have read
and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Data Availability Statement: Data are contained within the article.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Wang, W. Analysis of fault detection in rolling element bearings. IEEE Instrum. Meas. Mag. 2021, 24, 42–49. [CrossRef]
2. Mishra, R.K.; Choudhary, A.; Fatima, S.; Mohanty, A.R.; Panigrahi, B.K. A self-adaptive multiple-fault diagnosis system for rolling
element bearings. Meas. Sci. Technol. 2022, 33, 125018. [CrossRef]
3. Gupta, D.; de Albuquerque, V. Smart Sensors for Industrial Internet of Things: Challenges, Solutions and Applications; Springer:
Berlin/Heidelberg, Germany, 2021.
4. Shukla, A.; Mahmud, M.; Wang, W. A smart sensor-based monitoring system for vibration measurement and bearing fault
detection. Meas. Sci. Technol. 2020, 31, 105104. [CrossRef]
5. Mahmud, M.; Wang, W. A smart sensor-based cEMD technique for rotor bar fault detection in induction motors. IEEE Trans.
Instrum. Meas. 2021, 30, 3523811.
6. Xing, X.; Zhang, M.; Wang, W. Adaptive variational mode decomposition for bearing fault detection. J. Signal Inf. Process. 2023,
14, 9–26. [CrossRef]
7. Marsick, A.; André, H.; Khelf, I.; Leclère, Q.; Antoni, J. Restoring cyclostationarity of rolling element bearing signals from the
instantaneous phase of their envelope. Mech. Syst. Signal Process. 2023, 193, 110264. [CrossRef]
8. Bertoni, R.; André, H. Proposition of a bearing diagnosis method applied to IAS and vibration signals: The bearing frequency
estimation method. Mech. Syst. Signal Process. 2023, 187, 109891. [CrossRef]
9. Dibaj, A.; Hassannejad, R.; Ettefagh, M.; Ehghaghi, M.B. Incipient fault diagnosis of bearings based on parameter-optimized
VMD and envelope spectrum weighted kurtosis index with a new sensitivity assessment threshold. ISA Trans. 2021, 114, 413–433.
[CrossRef] [PubMed]
10. He, Y.; Hu, M.; Feng, K.; Jian, Z. Bearing condition evaluation based on the shock pulse method and principal resonance analysis.
IEEE Trans. Instrum. Meas. 2021, 70, 3509212. [CrossRef]
11. Sejdic, E.; Djurovic, I.; Jiang, J. Time–frequency feature representation using energy concentration: An overview of recent
advances. Digit. Signal Process. 2009, 19, 153–183. [CrossRef]
12. He, Y.; Jiang, Z.; Hu, M.; Li, Y. Local maximum synchrosqueezing Chirplet transform: An effective tool for strongly nonstationary
signals of gas turbine. IEEE Trans. Instrum. Meas. 2021, 70, 3516314. [CrossRef]
13. Huang, N.E.; Shen, Z.; Long, S.R.; Wu, M.C.; Shih, H.H.; Zheng, Q.; Yen, N.C.; Tung, C.C.; Liu, H.H. The empirical mode
decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis. Phys. Eng. Sci. 1998, 454, 903–995.
[CrossRef]
14. Pancaldi, F.; Dibiase, L.; Cocconcelli, M. Impact of noise model on the performance of algorithms for fault diagnosis in rolling
bearings. Mech. Syst. Signal Process. 2023, 188, 109975. [CrossRef]
15. Randall, R.B.; Antoni, J. Why EMD and similar decompositions are of little benefit for bearing diagnostics. Mech. Syst. Signal
Process. 2023, 129, 110207. [CrossRef]
16. Dragomiretskiy, K.; Zosso, D. Variational mode decomposition. IEEE Trans. Signal Process. 2014, 62, 531–544. [CrossRef]
17. Kumar, A.; Gandhi, C.P.; Vashishtha, G.; Kundu, P.; Tang, H.; Głowacz, A.; Shukla, R.; Xiang, J. VMD based trigonometric entropy
measure: A simple and effective tool for dynamic degradation monitoring of rolling element bearing. Meas. Sci. Technol. 2021,
33, 014005. [CrossRef]
18. Vashishtha, G.; Chauhan, S.; Yadav, N.; Kumar, A.; Kumar, R. Adaptive MOMEDA model based variational mode decomposition
for Pelton wheel fault detection. Eng. Res. Express 2022, 4, 025044. [CrossRef]
19. Sahani, M.; Dash, P.K. Deep convolutional stack autoencoder of process adaptive VMD data with robust multikernel RVFLN for
power quality events recognition. IEEE Trans. Instrum. Meas. 2021, 70, 9001912. [CrossRef]
20. Li, Z.; Chen, J.; Zi, Y.; Pan, J. Independence-oriented VMD to identify fault feature for wheel set bearing fault diagnosis of high
speed locomotive. Mech. Syst. Signal Process. 2017, 85, 512–529. [CrossRef]
21. Yi, C.; Lv, Y.; Dang, Z. A fault diagnosis scheme for rolling bearing based on particle swarm optimization in variational mode
decomposition. Shock Vib. 2016, 2016, 9372691. [CrossRef]
22. Cui, X.; Huang, J.; Li, C.; Zhao, Y. Three-dimensional instantaneous orbit map for rotor-bearing system based on a novel
multivariate complex variational mode decomposition algorithm. Mech. Syst. Signal Process. 2022, 178, 109211. [CrossRef]
23. Song, Q.; Jiang, X.; Wang, S.; Guo, J.; Huang, W.; Zhu, Z. Self-adaptive multivariate variational mode decomposition and its
application for bearing fault diagnosis. IEEE Trans. Instrum. Meas. 2022, 71, 3503913. [CrossRef]
24. Datasheet: LD1117 Series Voltage Regulator. Available online: https://www.sparkfun.com/datasheets/Components/LD1117V3
3.pdf (accessed on 1 January 2024).
Sensors 2024, 24, 2470 16 of 16

25. Datasheet: LT1930 DC/DC Converters in ThinSOT. Available online: https://www.analog.com/media/en/technical-

documentation/data-sheets/1930f.pdf (accessed on 1 January 2024).
26. Datasheet: LM234 3-Terminal Adjustable Current Sources. Available online: http://www.ti.com/lit/ds/symlink/lm134.pdf
(accessed on 1 January 2024).
27. Microstick II Information Sheet with PIC32. Available online: http://ww1.microchip.com/downloads/en/DeviceDoc/51951B.
pdf (accessed on 1 January 2024).
28. Data Sheet: Wireless Module CC1101110. Available online: http://www.ti.com/lit/ds/symlink/cc1101.pdf (accessed on
1 January 2024).
29. Liu, Y.; Chai, Y.; Liu, B.; Wang, Y. Impulse signal detection for bearing fault diagnosis via residual-variational mode decomposition.
Appl. Sci. 2021, 11, 3053. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.