See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/334486460

Remaining Useful Life Prediction of Rolling Element Bearings Using

Supervised Machine Learning

Article in Energies · July 2019

DOI: 10.3390/en12142705

CITATIONS READS

35 1,124

4 authors:

Xiaochuan Li Faris Elasha

De Montfort University Coventry University
41 PUBLICATIONS 657 CITATIONS 49 PUBLICATIONS 804 CITATIONS

SEE PROFILE SEE PROFILE

Suliman Shanbr David Mba

Cranfield University University of the Arts London
6 PUBLICATIONS 401 CITATIONS 372 PUBLICATIONS 7,349 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by David Mba on 18 April 2020.

The user has requested enhancement of the downloaded file.

 energies
Article
Remaining Useful Life Prediction of Rolling Element
Bearings Using Supervised Machine Learning
Xiaochuan Li 1, *, Faris Elasha 2 , Suliman Shanbr 3 and David Mba 1,4
1 Faculty of Computing, Engineering and media, De Montfort University, Leicester, LE1 9BH, UK
2 Faculty of Engineering, Environment and Computing, Coventry University, Coventry CV1 2JH, UK
3 Department of Engineering and Applied Science, School of Water, Energy and Environment,
Cranfield University, Bedfordshire, MK43 0AL, UK
4 Department of Mechanical Engineering, University of Lagos, Nigeria 100213, West Africa
* Correspondence: [email protected]
!"#!$%&'(!
Received: 9 June 2019; Accepted: 11 July 2019; Published: 15 July 2019 !"#$%&'

Abstract: Components of rotating machines, such as shafts, bearings and gears are subject to
performance degradation, which if left unattended could lead to failure or breakdown of the entire
system. Analyzing condition monitoring data, implementing diagnostic techniques and using
machinery prognostic algorithms will bring about accurate estimation of the remaining life and
possible failures that may occur. This paper proposes a combination of two supervised machine
learning techniques; namely, the regression model and multilayer artificial neural network model,
to predict the remaining useful life of rolling element bearings. Root mean square and Kurtosis were
analyzed to define the bearing failure stages. The proposed methodology was validated through two
case studies involving vibration measurements of an operational wind turbine gearbox and a split
cylindrical roller bearing in a test rig.

Keywords: prognostics; vibration measurement; regression model; artificial neural network; rolling
element bearing; remaining useful life

1. Introduction
Rolling element bearings are critical components in rotating machinery. These bearings generally
operate under adverse conditions, making performance degradation unavoidable. Such degradation,
if left unattended, can cause failure or breakdown of the entire system. The purpose of prognostics is
to use prediction techniques to predict the remaining useful life (RUL) of a system and its constituent
components based upon historical information, current usage and future operating conditions so
as to prevent the catastrophe from happening [1]. Historical information includes historical failure
data of similar systems and event logs. Current usage could be in the form of a feature derived
from sensor measurements which indicates the current health status of the system. Future operating
condition refers to operational and environmental factors that could a↵ect the future status of the
system. Such information can be obtained based on expert opinion or by looking at the production
plans. Prognostics o↵ers several benefits, including [2,3]:

• Avoiding catastrophic failure, unscheduled maintenance and production loss.

• Reducing maintenance costs by minimizing the number of unnecessary interventions and
machine overhauls.
• Increasing the lifespan of components by providing advance information on the severity of the
fault to be maintained.

Energies 2019, 12, 2705; doi:10.3390/en12142705 www.mdpi.com/journal/energies

Energies 2019, 12, 2705 2 of 17

The RUL of a bearing is generally defined either as the total number of revolutions before a failure
occurs
Energiesor the12,total
2019, number
x FOR of hours that the bearing can run until the first sign of failure develops
PEER REVIEW [4].
2 of 16

The RUL is estimated based on measured and calculated bearing condition variables such as vibration
develops [4]. The RUL is estimated based on measured and calculated bearing condition variables
amplitude and frequency. As shown in Figure 1, if a certain condition indicator x is calculated or
such as vibration amplitude and frequency. As shown in Figure 1, if a certain condition indicator x is
monitored continuously from t = 0 to t = tB , then a continuous time series y(t) can be obtained,
calculated or monitored continuously from t = 0 to t = 𝑡 , then a continuous time series y(t) can be
which represents the deterioration process of the component under study. This time series consists
obtained, which represents the deterioration process of the component under study. This time series
of two parts, ↵ and , which indicates the healthy running stage and the fault degradation stage,
consists of two parts, α and β, which indicates the healthy running stage and the fault degradation
respectively. Prognostic analysis is usually based on the analysis of the time series from point A to
stage, respectively. Prognostic analysis is usually based on the analysis of the time series from point
point B. Ideally, if the RUL of a bearing (i.e., the total running time between point A to B) can be
A to point B. Ideally, if the RUL of a bearing (i.e., the total running time between point A to B) can be
accurately estimated by using only the past data covered by ↵, then the optimal maintenance schedule
accurately estimated by using only the past data covered by α, then the optimal maintenance schedule
can be made easily.
can be made easily.

.
Figure 1. Bearing life process.
Figure 1. Bearing life process.
Over recent decades, much research has been implemented to develop health monitoring methods
for rotating machinery,
Over recent especially
decades, muchbearings.
researchCompared
has been to fault detection,
implemented to the literature
develop of prognostics
health monitoring
and healthfor
methods management is relatively
rotating machinery, limited, bearings.
especially and e↵ective implementing
Compared to faultofdetection,
prognostic thetechniques
literature is of
still lacking. The
prognostics and increased interest in is
health management machinery
relativelyprognostics
limited, andhas resulted
effective in many successful
implementing tools,
of prognostic
models and applications
techniques is still lacking.in the past
The few years.
increased Basically,
interest there are three
in machinery types of prognostic
prognostics has resulted approaches
in many
that can be employed
successful tools, models to predict the RUL; namely,
and applications data-driven
in the past few years. methods, physics-based
Basically, there are three models typesand of
hybrid models.
prognostic Data-driven
approaches thatapproaches
can be employedutilize thetohistorical
predict thefailure
RUL;datanamely,
of the machine
data-drivenand/or similar
methods,
machines to estimate
physics-based models howandmuch
hybridtime is left until
models. a systemapproaches
Data-driven malfunction occurs.
utilize theThis method
historical does data
failure not
require an in-depth
of the machine and/orunderstanding
similar machinesof theto physics
estimate of systems
how much under
timestudy.
is left Physics-based
until a system approaches
malfunction
predict
occurs.the remaining
This method life doesaccording
not requireto propagation
an in-depthofunderstanding
the damage mechanism
of the physics(i.e., physics
of systems of failure).
under
Astudy.
hybrid approach uses
Physics-based both a data-driven
approaches predict the and physics-based
remaining method to
life according so propagation
as to achieveof antheimproved
damage
predictive
mechanism performance
(i.e., physics in terms of a more
of failure). improved
A hybrid predictive
approach usesaccuracy than that of and
both a data-driven the single method.
physics-based
methodOversorecent years, many
as to achieve e↵orts have
an improved been conducted
predictive performance in developing
in terms ofregression-based
a more improvedprognostic predictive
methods
accuracythat thancanthatbe of used to estimate
the single method.the RUL of rotating machinery. Li et al. [5] improved the
performance
Over recentof the traditional
years, exponential
many efforts have beenregression
conducted model and applied
in developing the developed regression
regression-based prognostic
model
methods to vibration
that can measurements
be used to estimatecollected thefrom
RUL rolling elementmachinery.
of rotating bearings toLi predict
et al. RUL. Wu et al. [6]
[5] improved the
put forward aoftime-to-failure
performance the traditionalprognostic
exponential method
regressionbased on an
model andempirical
applied Bayesian
the developed algorithm and
regression
exponential regression
model to vibration model for rolling
measurements element
collected from bearings. Sutrisno bearings
rolling element et al. [7] investigated
to predict RUL. the accuracy
Wu et al.
of
[6]three
put di↵erent
forward atechniques for predicting
time-to-failure prognostic themethod
RUL ofbasedbearings.
on an Bayesian
empirical Monte Carloalgorithm
Bayesian and moving and
average spectral
exponential kurtosis,
regression support
model vector element
for rolling regression (SVR) and
bearings. an anomaly
Sutrisno et al. [7]detection
investigated algorithm were
the accuracy
compared according
of three different to their performance
techniques for predicting in the
estimating
RUL of abearings.
ball bearing’s
Bayesianremaining
Monte Carlolife. Theandanomaly
moving
detection technique
average spectral was found
kurtosis, to bevector
support the most accurate(SVR)
regression among allan
and methods
anomaly compared.
detectionGoebelalgorithm et al.were
[8]
conducted
comparedaaccording
comparative studyperformance
to their of three prognostic methods:
in estimating relevance
a ball vector
bearing’s machine life.
remaining (RVM),TheGaussian
anomaly
detection technique was found to be the most accurate among all methods compared. Goebel et al.
[8] conducted a comparative study of three prognostic methods: relevance vector machine (RVM),
Gaussian process regression (GPR) and a neural network. The study showed that the three techniques
have resulted in significant different RUL prediction results. Loukopoulos et al. [9] studied the
performance of several machine learning techniques, including linear regression, polynomial
Energies 2019, 12, 2705 3 of 17

process regression (GPR) and a neural network. The study showed that the three techniques have
resulted in significant di↵erent RUL prediction results. Loukopoulos et al. [9] studied the performance of
several machine learning techniques, including linear regression, polynomial regression and K-Nearest
Neighbors Regression. The results showed that an ensemble method based on the weighted average
of the predicted RUL of each individual method o↵ers a higher predictive accuracy. Kim et al. [10]
utilized Support Vector Regression to evaluate the bearing health condition by using real-world
run-to-failure data obtained from bearings of gas pumps. The results showed that the developed
probability estimation based prognostic method is potentially very e↵ective for RUL prediction.
Several other artificially intelligent approaches applied to machinery prognostics have been
considered by researchers. For instance, a self-organizing neural network was employed by Zhang
and Ganesan [11] for extrapolating the fault progression and estimating the remnant life of a bearing.
Loukopoulos et al. [9] applied a Self-Organizing Map (SOM) model to predict the RUL of industrial
pumps using temperature measurements. Dong et al. [12] developed a condition prediction method
based on the grey model and back-propagation neural network. Elasha et al. [13] put forward a life
assessment approach for tidal turbine gearboxes. The method was validated on data generated using
a Blade Element Momentum Theory (BEMT) model. They predicted the RUL of a gearbox based on
the turbine loading conditions. The results of their investigation show life variations between the
gears due to di↵erences in stress cycles and di↵ering rotational speeds. Li et al. [14] put forward
a hybrid method in which a long short-term memory model and a state-space model was combined
to predict the pro-fault performance of a centrifugal compressor. An adaptive neuro-fuzzy inference
system (ANFIS) was used together with the particle filtering (PF) algorithm in [15] to predict the RUL
of a gearbox. The authors concluded that the ANFIS model outperforms the recurrent neural network
through a comparative study. Elforjani and Shanbr [16] employed three supervised machine learning
techniques; Artificial neural network (ANN), SVR and GPR, to correlate Vibration measurement features
with natural wear of bearings. They concluded that the back-propagation neural network model
outperforms the other methods in predicting the RUL of bearings. The aforementioned prognostic
techniques o↵er a tradeo↵ between reliability, speed and applicability. This paper combines two
supervised machine learning techniques; namely, the regression model and artificial neural network
(ANN) model to correlate vibration features with the corresponding fault stages during the natural
run-to-failure process of rolling element bearings. The main contribution of this study is to improve the
fitting of condition indicators obtained from vibration signals of bearings using appropriate regression
models. Furthermore, this study aims to ascertain the feasibility of using ANN models to estimate the
RUL of rolling element bearings, and to explore the feasibility of combining regression models with
ANNs for a better RUL prediction. The e↵ectiveness of the hybrid prognostic method was validated
through two case studies: an operational wind turbine gearbox and a split cylindrical roller bearing
used in an experimental test rig.

2. Methodology

Contributions
In condition monitoring applications, the vibration signals of bearing damage often present
multiple modulation characteristics, and therefore the features extracted by the general methods
from one bearing may not necessarily correlate to fault characteristics extracted from another bearing.
The internal reasons behind this include, for example, di↵erent observed trends from di↵erent cases.
As a result, there is still a need to apply and validate bearing fault indicators such as root mean square
(RMS) and Kurtosis (KU) for di↵erent applications. Furthermore, due to the measurement noise,
variation of operating conditions and stochasticity of the system deterioration, the extracted condition
indicators from the raw vibration signals generally contain fluctuations, which would incur inaccurate
RUL predictions. In this context, this study contributes to the existing literature in that:
Energies 2019, 12, 2705 4 of 17

• It improves the fitting of the features extracted from vibration measurements through the use of
appropriate regression models.
• It explores the possibility of combining regression models with ANNs to improve the predictive
accuracy of regression models.
• A test rig was designed and experimental tests were conducted to generate bearing failure data in
order to validate the proposed prognostic model. The model was also validated through industrial
data provided by a commercial company.

3. Statistical Condition Indicators

Vibration based health monitoring schemes are applicable for monitoring many constituent
components of a gearbox, such as shafts, gears and bearings. To achieve a better signal-to-noise ratio
(SNR), vibration measurements are processed using filtering and amplifying techniques. Two condition
indicators, RMS and KU [17], are often generated from the vibration signals. Then the extracted
condition indicators are fitted using regression methods to provide useful information about the
bearing degradation.

3.1. Kurtosis
Each mechanical failure has an associated “signature” that can be found in the frequency or time
domain representations of vibration signals. Kurtosis is such a “signature” which is referred to as the
fourth statistical moment of a given signal, reflecting the peakness of the histogram [18]. A kurtosis
value greater than three is an indicator of a sharp peak signal. A kurtosis value smaller than three
indicates vibration signals with flat peaks. In some cases, the occurrence of background noise and other
sources of vibration signals may prevent bearing faults from being detected through the observation of
changes in the kurtosis. To solve this problem, the kurtosis value needs to be computed across di↵erent
frequency bands [19]. The Kurtosis of a random signal is computed as:
PN
1
N i = 1 (Xi µ)4
KU = h PN i2 (1)
1
N i = 1 (Xi µ)2

where N is the number of samples in the signal, Xi refers to the amplitude of the signal of the ith
sample, and µ denotes the mean sample amplitude.

3.2. Root Mean Square

The root mean square (RMS) value describes the energy content of a signal. RMS is one of the
most commonly used statistical parameters that describe change in the dynamic of the machine [20].
For the signal of sample size N, the RMS value is calculated using the equation below.
q
X12 + X22 + . . . XN
2
RMS = (2)
N
It is well known that the RMS value is a weak method to detect failure at its early stage because
of the small energy generated by the defect, which makes a small di↵erence in the value of RMS.
However, RMS is capable of reflecting the increase of the vibration energy as the fault progresses [5].
Consequently, RMS is employed as the prediction indicator in this study. In other words, the RUL is
predicted by extrapolating the trajectory of RMS values.

4. Regression Analysis
Regression models, as one of the most popular data-driven techniques for RUL prediction,
make attempts at fitting available data of deterioration by regression functions and then extrapolating
Energies 2019, 12, 2705 5 of 17

the fault propagation until the fitted curve reaches a pre-defined threshold. The objective of regression
analysis is to find an empirical relation to predicting the bearing degradation thought time series. Due to
measurement noise, variation of operational conditions and the stochastic nature of the degradation
processes, the acquired data are usually accompanied by fluctuations that may have a significant impact
on the model’s ability to interpret the degradation trend. In this case, the raw condition indicators
cannot be directly used as the inputs of the prediction models. This is due to the fact that any fluctuations
in the condition indicators will cause the model to follow the randomness, and consequently, its ability
to accurately estimate the health status of the bearings may be very weak [21]. Therefore, in this study,
we first conduct a comparative study of two regression models, namely polynomial and exponential
regression, and then choose the one with the best fitting performance to fit the condition indicators
extracted from the data.
The polynomial models are suitable for situations where the correlation between explanatory and
study variables is curvilinear. Polynomial regression belongs to the least-square curve fitting family.
It takes a set of data as inputs and generates an approximation between the input data and time. To be
specific, it estimates the coefficients of a polynomial function in that the function approximates the
curve closely. The formula of polynomial regression is as follows:

y1 = a0 + a1 x + a2 x2 + . . . + an xni (3)

where y1 is the response variable, is the predictor variable, and a0 , a1 , . . . , an are model coefficients.
The degree of the polynomial function is determined by the number of non-zero coefficients in (3),
which in turn determines how accurate the data can be fit. If the number of coefficients is zero or one,
then the fitted curve is known as a linear regression. If the number of coefficients is larger than one,
a non-linear polynomial regression will be implemented.
The exponential regression model involves a fitting process that finds the parameters of the
exponential function which can present the best fit for a set of data [21]. The function form of the
exponential is shown in Equation (4):
y2 = a ⇥ ebx (4)

where a, b are model constants and is the predictor variable. Figure 2 shows the process of using
the two regression models to find the best fit of the condition indicators. The vibration signals were
first processed to obtain the condition indicators; namely, RMS and KU. Then the coefficients in
Equations (3) and (4) are determined such that the best fit of the exponential and polynomial functions
are reached. The performance of the regression models has been assessed using two statistics: root
mean square error (RMSE) and adjusted R2 . The RMSE is defined as the square root of the variance of
the residuals between the predicted and the actual RUL. RMSE measures how close the measured data
is to the predicted values. The R2 is defined as the ratio between the di↵erence between the sum square
total (SST) and sum square error (SSE) to the SST. SST measures the data deviation from the sample
mean, and SSE measures the deviation of the data from the model’s predicted values. One of the
disadvantages of the R2 metric is that it does not indicate if a regression model provides an adequate
fit to the data. Therefore, the adjusted R2 was instead used in this research as the third performance
metric. Adjusted R2 is defined as the ratio between the residual mean square errors to the total mean
square error (which is the variance of the predicted values).
 between the sum square total (SST) and sum square error (SSE) to the SST. SST measures the data
deviation from the sample mean, and SSE measures the deviation of the data from the model’s
predicted values. One of the disadvantages of the R2 metric is that it does not indicate if a regression
model provides an adequate fit to the data. Therefore, the adjusted R2 was instead used in this
research as the third performance metric. Adjusted R2 is defined as the ratio between the residual
Energies 2019, 12, 2705 6 of 17
mean square errors to the total mean square error (which is the variance of the predicted values).

Energies 2019, 12, x FOR PEER REVIEW 6 of 16

Figure 2. Schematic of Regression Process.

5.5.Multilayer Figure 2. Schematic of Regression Process.

MultilayerArtificial
ArtificialNeural
NeuralNetwork
Network
ANNsANNs belong belong
to theto supervised
the supervised machine
machine learning
learning family.
family. TheyTheyareare inspiredbybybiological
inspired biologicalneural
neural
networksand
networks andeach
eachneuron
neuronisisrepresented
representedby byaanode
node[22].
[22].An
AnANN
ANNgenerally
generallycontains
containsananinput
inputlayer,
layer,
multiplehidden
multiple hiddenlayers,
layers,oneoneoutput
outputlayer,
layer,biases
biasesand andconnection
connectionnodes.
nodes.When
Whenthetheknown
knowninputs
inputsandand
targetoutputs
target outputsare arerepetitively
repetitivelypresented
presentedto toananANN,
ANN,the theconnection
connectionweights
weightsbetween
betweennodes
nodeswill
willbe
be
adjusted automatically
adjusted automatically such such that
thatthe
thedi↵erence
differencebetween
between thethe
network
network outputs andand
outputs the targets is as small
the targets is as
as possible.
small In this
as possible. Instudy, the ANN
this study, that we
the ANN implemented
that we implementedis a multilayer back-propagation
is a multilayer neural
back-propagation
network.
neural FigureFigure
network. 3 illustrates an exemplary
3 illustrates an exemplaryarchitecture of the multiple-layer
architecture neuralneural
of the multiple-layer network model.
network
It is observable
model. from the
It is observable fromfigure that each
the figure thatlayer
each consists of its of
layer consists ownits input and output
own input nodes
and output (x & (x
nodes y),
&weighting coefficients
y), weighting (w) and
coefficients (w)bias
andvector (b). The
bias vector (b).input layer layer
The input doesn’t involve
doesn’t any processing
involve and is
any processing
utilized
and to directly
is utilized feed information
to directly feed informationto the to
subsequent network
the subsequent layers.layers.
network

Multiple-layerneural
Figure3.3.Multiple-layer
Figure neuralnetwork.
network.

In contrast, the output layer involves weighting and biases calculations when producing the
In contrast, the output layer involves weighting and biases calculations when producing the
network outputs. The hidden layers aim at adding additional processing so as to avoid solutions
network outputs. The hidden layers aim at adding additional processing so as to avoid solutions that
that do not converge. As shown in Figure 4, the main purpose of the bias neurons is to prevent the
do not converge. As shown in Figure 4, the main purpose of the bias neurons is to prevent the network
network from generating zero results even if the network inputs are not zero. The exemplary network
from generating zero results even if the network inputs are not zero. The exemplary network
structure is formed of a feed-forward model. The following equation explains how the network
inputs are correlated with the outputs:
𝒉𝒊 = 𝝋𝟎 [𝐂𝝋𝒉 (𝑩𝒖𝒊 + 𝒃𝒉 ) + 𝒃𝟎 ] (5)
where ℎ is the network output vector and the input vector is represented by 𝑢 , C denotes the
Energies 2019, 12, 2705 7 of 17

structure is formed of a feed-forward model. The following equation explains how the network inputs
are correlated with the outputs:

hi = '0 [C'h (Bui + bh ) + b0 ] (5)

where hi is the network output vector and the input vector is represented by ui , C denotes the weighting
matrix between the hidden layer and the output layer. B is the connection matrix from the input layer
to the hidden layer. The bias vectors of the hidden and output layers are represented by bh and bo,
respectively. 'h and 'o denote the activation functions of the nodes in the hidden and output layers,
respectively. Feedforward neural network models also take the form of

hi = f (u) (6)

where f (·) denotes a nonlinear transformation from u to hi . Interestingly, the structure of a feedforward
neural network is similar to that of a nonlinear regression model. Levenberg Marquardt (LM) learning
algorithm [23] was chosen as the network training function in this study for adjusting the weighting
and bias matrices during the training process. Levenberg-Marquardt optimization has been applied
intensively
Energies forxfeedforward
2019, 12, neural network training and has been proven to be able to deal with 7many
FOR PEER REVIEW of 16

difficult and diverse problems in practice. This algorithm minimizes functions that are sums of squares
are sums of squares of nonlinear functions. One of the advantages of this optimization method is that
of nonlinear functions. One of the advantages of this optimization method is that the second-order
the second-order convergence point can be approached without calculating the Hessian matrix.
convergence point can be approached without calculating the Hessian matrix.

Figure 4.
Figure Schematic of
4. Schematic of test
test rig
rig construction.
construction.

6. Data Collection
6. Data Collection
The proposed approach was validated on two datasets: vibration signals captured from
The proposed approach was validated on two datasets: vibration signals captured from a high-
a high-speed shaft bearing of a 2 MW wind turbine and a split cylindrical roller bearing in a test rig.
speed shaft bearing of a 2 MW wind turbine and a split cylindrical roller bearing in a test rig.
6.1. Dataset 1
6.1. Dataset 1
The first vibration dataset was collected from the Green Power Monitoring Systems [24]. The data
The first from
was collected vibration dataset was
a high-speed shaftcollected from the inside
bearing mounted Green aPower
2 MW Monitoring Systems
wind turbine. [24]. The
The high-speed
data was collected from a high-speed shaft bearing mounted inside a 2 MW wind turbine.
bearing which is the subject of this study is housed on the tail end of a gearbox. The bearing defect The high-
speed
was bearing which
efficiently is the
detected by subject of this
the sensor study
which wasis mounted
housed onradially
the tail onto
end of
thea bearing
gearbox.support
The bearing
ring.
An inspection of the bearing later showed that the inner race was cracked. The crack length wassupport
defect was efficiently detected by the sensor which was mounted radially onto the bearing 26 mm
ring.
in theAn inspection
outer race. Theof vibration
the bearing later showedwas
measurement thattaken
the inner race
for 50 was cracked.
consecutive daysThe crack
using length was
MEMS-based
26
accelerometers mounted radially on the bearing support ring [24]. Data was collected at MEMS-
mm in the outer race. The vibration measurement was taken for 50 consecutive days using 10-min
based accelerometers
intervals. The bearing mounted
speed wasradially
1800 rpm.on Athetotal
bearing
numbersupport
of 50 ring
data [24]. Datarecorded
sets were was collected at 10-
for analysis.
min intervals. The bearing speed was 1800 rpm. A total number of 50 data sets were recorded for
analysis. The vibration data was sampled at a sampling rate of 97,656 Hz for 6 s [24]. Table 1 lists the
key parameters of interest.

Table 1. Wind turbine operating details.

Machine State Increasing Inner Race Bearing Fault

Energies 2019, 12, 2705 8 of 17

The vibration data was sampled at a sampling rate of 97,656 Hz for 6 s [24]. Table 1 lists the key
parameters of interest.

Energies 2019, 12, x FOR PEER REVIEW Table 1. Wind turbine operating details. 8 of 16

Machine State Increasing Inner Race Bearing Fault

to measure vibration in both directions. A split cylindrical roller bearing (type 01B 40MM GR) was
employed to perform thePower rating
run-to-failure test. The reason why this 2 MW
typefluxof bearing was selected was
Nominal speed 1800 rpm
that it can be mounted and dismounted
Measurement Channel
simply, permitting regular inspection
Sensor
of the test bearing
throughout the test series.Sample
As shownrate in Figure 5, each element of the testHz
97,656 specimen was split into two
halves. The dimensions Recordof the length
test specimen are shown in Table 2. 6 s A handheld tachometer was
employed to measure theSensor shaft type
speed. The test load on the roller Accelerometer
bearing was applied as normal to
the driven shaft and radially to the designated load point; see Figure 6. A hydraulic ram was used to
6.2. Dataset
apply 2 load, which was powered by a hand-pump; the magnitude of the pressure in the ram
the test
was measured using the fitted pressure gauge.
The second dataset was captured from an experimental test rig. The schematic of the test
rig is shown in Figure 4. It was used to investigate the use of vibration signals to diagnose
Table 2. Specification of Cooper cylindrical roller bearing type (01B 40MM GR).
incipient bearing failure under conditions of high speed and under constant load. An electric
motor (DC Motor—DY1815-B) is partParameter of the rig. The motor drive shaft can rotate at controlled speeds
Measurement
between 10 and 4000 rpm. Two PZE accelerometers, each 84.14
External Diameter (Pitch) with amm useful frequency range of at least
15 kHz, were attached to the test bearing
Internal housing,
Diameter one mounted
(Bore) 40.00radial
mm and horizontal, the other radial
and vertical, to measure vibration in both
Pitch directions.
Circle DiameterA split 62.71
cylindrical
mm roller bearing (type 01B 40MM
GR) was employed to perform the run-to-failure test. The reason why
Roller Diameter 11.91 mmthis type of bearing was selected
was that it can be mounted and dismounted simply, permitting10
Number of Rollers regular inspection of the test bearing
throughout the test series. AsWeightshownofinthe Figure 5, each element
Specimen 1.2 kgof the test specimen was split into
two halves. The dimensions of the test specimen are shown in Table 2. A handheld tachometer was
employed to measure
A hydraulic the ram
cylinder shaftprovided
speed. The thetest load
radial on the
load roller
on the bearinghousing.
specimen was applied as supported
It was normal to
the driven shaft
by an H-frame and radially to the designated load point; see Figure 6. A hydraulic ram was used
had a gauge displaying the load value; see Figure 6. A high-accuracy to
data
apply the test
acquisition DAQload,
waswhich
usedwas powered
to record andby a hand-pump;
monitor the magnitude
the vibration signal. TheofDAQ the pressure
recordedinall
the
ofram
the
was measured using
measurements the fittedrate
at a sampling pressure
of 25.6gauge.
kHz

Specimen as an assembled and in parts.

Figure 5. Specimen

The accelerated life test was performed on the bearing to obtain vibration measurements during
the bearing’s entire lifecycle from beginning to final failure. The Virtual Instrument (VI), consisting
of a computer equipped with LabVIEW programme connected to data acquisition with sensors, was
used to acquire and display vibration data. To speed up the crack initiation, a small amount of grease
Energies 2019, 12, 2705 9 of 17

Table 2. Specification of Cooper cylindrical roller bearing type (01B 40MM GR).

Parameter Measurement
External Diameter (Pitch) 84.14 mm
Internal Diameter (Bore) 40.00 mm
Energies 2019, 12, x FOR PEERPitch Circle Diameter
REVIEW 62.71 mm 9 of 16
Roller Diameter 11.91 mm
amount of lubricant inside Number of Rollers
the bearing 10 to measure and control because
during operation can be hard
Weight of the Specimen
bearing location and geometry can make accessibility difficult. 1.2 kg

Figure 6.
Figure Loading system.
6. Loading system.

A hydraulic cylinder ram provided the radial load on the specimen housing. It was supported by
7. Results
an H-frame and had a gauge displaying the load value; see Figure 6. A high-accuracy data acquisition
DAQ
7.1. was used
Signal to record
Features and monitor the vibration signal. The DAQ recorded all of the measurements
Regression
at a sampling rate of 25.6 kHz
It has been mentioned previously that due to measurement noise, variation of operational
The accelerated life test was performed on the bearing to obtain vibration measurements during
conditions and the stochastic nature of the degradation processes, the raw condition indicators cannot
the bearing’s entire lifecycle from beginning to final failure. The Virtual Instrument (VI), consisting of
be directly used as the inputs of the prediction models. To solve this problem, the obtained raw data
a computer equipped with LabVIEW programme connected to data acquisition with sensors, was used
were fitted by means of appropriate mathematical functions (i.e., exponential and polynomial models
to acquire and display vibration data. To speed up the crack initiation, a small amount of grease was
were utilized to represent trends of the condition indicators). The regression model utilized the two
applied to the roller bearing. It is worth knowing that the quantity of grease was not measured by
equations mentioned above (i.e., Equations (4) and (5)) to find the relationship between condition
weight nor volume. The purpose of doing this was to simulate real-world situations where the amount
indicators and time. Figures 7 and 8 show the actual and fitted values of the (a) high-speed shaft
of lubricant inside the bearing during operation can be hard to measure and control because bearing
bearing case for 50 days of measurements of RMS and KU; and (b) the split cylindrical roller bearing
location and geometry can make accessibility difficult.
for 298 h of testing.
7. Results

7.1. Signal Features Regression

It has been mentioned previously that due to measurement noise, variation of operational
conditions and the stochastic nature of the degradation processes, the raw condition indicators cannot
be directly used as the inputs of the prediction models. To solve this problem, the obtained raw data
were fitted by means of appropriate mathematical functions (i.e., exponential and polynomial models
were utilized to represent trends of the condition indicators). The regression model utilized the two
equations mentioned above (i.e., Equations (4) and (5)) to find the relationship between condition
indicators and time. Figures 7 and 8 show the actual and fitted values of the (a) high-speed shaft
bearing case for 50 days of measurements of RMS and KU; and (b) the split cylindrical roller bearing
for 298 h of testing.

(a)
were fitted by means of appropriate mathematical functions (i.e., exponential and polynomial models
were utilized to represent trends of the condition indicators). The regression model utilized the two
equations mentioned above (i.e., Equations (4) and (5)) to find the relationship between condition
indicators and time. Figures 7 and 8 show the actual and fitted values of the (a) high-speed shaft
bearing2019,
Energies case12,for 50 days of measurements of RMS and KU; and (b) the split cylindrical roller bearing
2705 10 of 17
for 298 h of testing.

Energies 2019, 12, x FOR PEER REVIEW 10 of 16

(a)

(b)
Figure 7.7. Fitted
Figure Fitted condition
condition indicators
indicators for
for dataset
dataset 1:1: (a)
(a) Fitted
Fitted condition
condition indicators
indicators using
using exponential
exponential
functions; (b) Fitted condition indicators using polynomial functions.
functions; (b) Fitted condition indicators using polynomial functions.

Tables 3 and 4 list the optimal fitting parameters and the associated RMSE and adjusted R2 values
for the exponential and polynomial models, respectively. It can be observed that using the exponential
regression to fit the condition indicators is more accurate than the polynomial. For instance, in Table 3
RMS showed the lowest value of root mean square error of 0.05248 and a high adjusted R2 with 0.957.
Whereas the results obtained from polynomial for the same condition indicator showed the lowest
adjusted R2 with 0.5947. Therefore, exponential regression was selected to calculate the RUL, and the
computed results were compared to those of the ANN model.

(a)
 (b)
Energies 7. 2705
2019, 12,
Figure 11 of 17
Fitted condition indicators for dataset 1: (a) Fitted condition indicators using exponential
functions; (b) Fitted condition indicators using polynomial functions.

(a)

(b)
Figure 8. Fitted condition
Figure 8. condition indicators
indicators for
for dataset
dataset 22 (exponential
(exponential and
and polynomial
polynomial functions):
functions): (a)
(a) Actual
Actual
and
and fitted
fitted root
root mean
mean square
square(RMS);
(RMS);(b)
(b)Actual
Actualand
andfitted
fittedKurtosis
Kurtosis(KU).
(KU).

Table 3. Optimal estimated exponential model constants.

Dataset 1 Dataset 2
Condition Indicators Model Constants Model Constants
RMSE Adj. R2 RMSE Adj. R2
a b a b
RMS 2.235 0.0511 0.05248 0.957 0.003835 0.3797 0.000717 0.995
KU 3.439 0.1121 0.9469 0.975 7.87 0.2548 0.2494 0.990

Table 4. Optimal estimated polynomial model constants.

Dataset 1 Dataset 2
Condition Indicators Model Constants Model Constants
RMSE Adj. R2 RMSE Adj. R2
a0 a1 a2 a0 a1 a2
RMS 2.19 0.117 0.0044 0.1644 0.5947 0.000552 0.001365 0.003602 0.0007735 0.853
KU 3.24 0.5718 0.3004 0.2172 0.881 0.244 1.971 7.786 0.3851 0.792
Energies 2019, 12, 2705 12 of 17

7.2. Artificial Neural Network

In this study, we propose use of an ANN model for estimating the RUL for bearing running at
high speed and constant load. The model is a feed-forward back-propagation artificial neural network
model with one input layer, a changeable number of hidden layers and one output layer. The optimal
number of hidden layers, learning rate and algorithm type were determined during the training process
by minimizing the error between target outputs and model outputs. Based on the training results,
the ANN model that was finally adopted consisted of one input layer with two inputs, namely raw
RMS and KU; one output layer with one output (fitted RMS values); and two hidden layers with 9
neurons in the first layer and 7 neurons in the second layer. The neural network was trained with the
resilient back-propagation algorithm and the logistic activation sigmoid function.
The inputs of the ANN model are the raw RMS and KU values, while the target is the best fitted
RMS obtained using fitting tools for the same parameter settings. The network structure maps between
these variables to generate the function of explaining their relationship. The extracted raw RMS
and KU were fed to the model, and the model was trained using the algorithms stated previously.
As mentioned previously, the lowest training errors were obtained by an ANN model containing one
input layer, one output layer and two hidden layers. The layer size was determined according to the
mean square errors between the target output series and the estimated outputs.
The next phase comprised of validation of the trained model. This was achieved by passing the
raw KU and RMS values to the trained model. The ANN model’s output (i.e., estimated fitted RMS) is
then extrapolated to a pre-defined failure threshold to predict the RUL of the system. Figures 9 and 10
presented the final results of model output obtained from 70% of training and 30% of testing data.
The error and the RUL were calculated using the following equations:

Actual (RUL) Estimated (RUL)

Error (%) = (7)
Actual (RUL)

RUL = t f tc (8)

where t f is the time at which the failure occurred; in this study, t f was chosen as the last time instance
of running. The tc is the time point at which the RUL is predicted.
Energies 2019, 12, 2705 13 of 17
Energies
Energies2019,
2019,12,
12,xxFOR
FORPEER
PEERREVIEW
REVIEW 12
12 of
of 16
16

Figure
Figure9. Neural Network
9.Neural Network Training
NetworkTraining
TrainingRegression
Regression(dataset
(dataset1).
1).

Figure
Figure10. Neural
10.Neural Network
NeuralNetwork Training
NetworkTraining Regression(dataset
TrainingRegression (dataset2).
2).

Figures
Figures
Figures 11 11
11 and
and 12
and 12 show
12 show the
show the results
the resultsin
results interms
in termsof
terms ofRUL
of RULprediction.
RUL prediction. The
prediction. The associated
The associated errors
associated errors are
are also
errors are also
also
plotted
plotted
plottedin in the
inthe figures.
thefigures.
figures.Based Based
Basedon on observation
onobservation
observationof of
ofthe the trends
thetrends
trendsofoftheof the predicted
thepredicted
predictedRUL RUL
RULcompared compared
comparedto tothe to the
theactual
actual
actual
one, ititone,
one, is it is evident
is evident
evident that that
that the the proposed
the proposed
proposed approach
approach
approach satisfactorily
satisfactorily
satisfactorily follows
follows
follows thethedeterioration
the deteriorationtrend
deterioration trend of
trend of the
of the
the
bearings
bearings
bearingsunderunder
understudy studywith
study withsmall
with smallerrors.
small errors.To
errors. ToTobebe
be specific,
specific,
specific, Figure
Figure
Figure 11a
11a 11a depicts
depicts
depicts thethe
the estimated
estimated
estimated RUL
RUL RUL
with
withwith
the
the
the
actualactual
actual bearing
bearing
bearing life life
lifeusing
usingusing
the the exponential
the exponential
exponential regression
regression
regression model
model model
as as a predicted
asaa predicted
predicted tool. tool.
tool.It It shows
Itshows
shows that
thatthethat
the the
trend
trend
at
at the
the beginning
beginning of of the
the prediction
prediction initially
initially closely
closely estimated
estimated thethe actual
actual values.
values. However,
However, as as the
the end
end
of bearing life approaches, the prediction deviates from the actual values and oscillates
of bearing life approaches, the prediction deviates from the actual values and oscillates away. Results away. Results
Energies 2019, 12, 2705 14 of 17

trend at the beginning of the prediction initially closely estimated the actual values. However, as the
Energies 2019, 12, x FOR PEER REVIEW 13 of 16
end of bearing life approaches, the prediction deviates from the actual values and oscillates away.
Results obtained from the ANN analysis using two inputs (RMS and KU) are depicted in Figure 11b.
obtained from the ANN analysis using two inputs (RMS and KU) are depicted in Figure 11b. Results
Results observation shows that the prediction made by the proposed model is very accurate. The RUL
observation shows that the prediction made by the proposed model is very accurate. The RUL results
results obtained for dataset 2 show that the errors between the 0 and 250 operating hours of the ANN
obtained for dataset 2 show that the errors between the 0 and 250 operating hours of the ANN model
model are significantly lower than those of the regression model. The mean square error (MSE) was
are significantly lower than those of the regression model. The mean square error (MSE) was
calculated in order to further compare the ANN and the regression models. The MSE of the regression
calculated in order to further compare the ANN and the regression models. The MSE of the regression
model was 14.57, whereas the ANN model only had an MES value of 6.78. Therefore, the prediction
model was 14.57, whereas the ANN model only had an MES value of 6.78. Therefore, the prediction
made by the proposed ANN model is more accurate.
made by the proposed ANN model is more accurate.

(a)

(b)

Figure 11. Remaining useful life (RUL) results (dataset 1): (a) Regression Model RUL results; (b) Artificial
Figure 11. Remaining useful life (RUL) results (dataset 1): (a) Regression Model RUL results; (b)
neural network RUL results.
Artificial neural network RUL results.
Energies 2019, 12, 2705 15 of 17

Energies 2019, 12, x FOR PEER REVIEW 14 of 16

(a)

(b)

Figure 12.Remaining
Figure 12. Remaininguseful lifelife
useful (RUL) results
(RUL) (dataset
results 2): (a)2):
(dataset Regression Model RUL
(a) Regression results;
Model RUL(b) Artificial
results; (b)
neural network RUL results.
Artificial neural network RUL results.
8. Conclusions
8. Conclusions
A data-driven prognostics model has been developed for RUL prediction of rolling element
A data-driven
bearings. Two typesprognostics model
of prediction has have
methods been been
developed for
used to RUL and
model prediction
estimate ofthe
rolling element
remnant life:
bearings. Two
regression andtypes of prediction neural
back-propagation methods have been
network. Theused to model
outputs of theand estimatemodel
regression the remnant life:
have been
regression
used to feedand
theback-propagation neural
neural network. This network.
paper explores The
theoutputs of the
possibility regression regression
of combining model have been
models
used to feed the neural network. This paper explores the possibility of combining regression
with ANNs to improve the predictive accuracy of regression models. This paper also contributes to models
withexisting
the ANNs literature
to improve bythe predictive
designing accuracy
a test rig andofconducting
regression experimental
models. This paper
tests toalso contributes
generate bearingto
the existing literature by designing a test rig and conducting experimental tests to generate
failure data in order to validate the proposed prognostic model. The model was also validated through bearing
failure data
industrial in provided
data order to byvalidate the proposed
a commercial company.prognostic model. The model was also validated
through industrial data provided by a commercial
The obtained results using regression and ANN company.
models have been compared. The performance
The obtained results using regression and ANN
of each regression model was compared using two parameters: models haveRMSEbeen compared. The performance
and adjusted-R 2 . The result
of each regression model was compared using two parameters: RMSE and adjusted-𝑅
showed that the exponential model had better performance than polynomial regression. .The Theresults
result
showed that the exponential model had better performance than polynomial regression.
from the proposed ANN show that the model had good performance in predicting RUL of the bearing The results
from the
failure, andproposed ANNcan
this success show that the model
be attributed to the had
link good performance
created between theinregression
predictingmodel
RUL of the
to the
bearing failure, and this success can be attributed to the link created between the regression model to
ANN through the best fit condition indicator. Comparing performance of the regression model and
the ANN through the best fit condition indicator. Comparing performance of the regression model
and the ANN, it can be seen that the ANN has better performance; however, this performance cannot
Energies 2019, 12, 2705 16 of 17

the ANN, it can be seen that the ANN has better performance; however, this performance cannot be
achieved without the regression model, and so therefore the regression model is considered necessary
to improve the performance of the ANN model.

Author Contributions: methodology, F.E. and S.S.; writing—original draft preparation, S.S.; writing—review and
editing, X.L.; supervision, D.M.
Funding: This research received no external funding.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Saha, B.; Goebel, K.; Poll, S.; Christophersen, J. Prognostics Methods for Battery Health Monitoring Using
a Bayesian Framework. IEEE Trans. Instrum. Meas. 2009, 58, 291–296. [CrossRef]
2. Misra, K.B. Maintenance Enginerring and Maintenability: An Introduction; Springer: London, UK, 2008.
3. Tondon, N.; Choudhury, A. A review of vibration and acoustics measurement methods for the detection of
defects in rolling element bearing. Tribol. Int. 1999, 32, 469–480. [CrossRef]
4. Okoh, C.; Roy, R.; Mehnen, J.; Redding, L. Overview of Remaining Useful Life prediction techniques in
Through-life Engineering Services. Procedia CIRP 2014, 16, 158–163. [CrossRef]
5. Li, N.; Lei, Y.; LiN, J.; Ding, S.X. An Improved Exponential Model for Predicting Remaining Useful Life of
Rolling Element Bearings. IEEE Trans. Ind. Electron. 2015, 62, 7762–7773. [CrossRef]
6. Wu, J.; Wu, C.; Cao, S.; Or, S.W.; Deng, C.; Shao, X. Degradation Data-Driven Time-To-Failure Prognostics
Approach for Rolling Element Bearings in Electrical Machines. IEEE Trans. Ind. Electron. 2018, 66, 529–539.
[CrossRef]
7. Sutrisno, E.; Oh, H.; Vasan, A.S.S.; Pecht, M. Estimation of remaining useful life of ball bearings using data
driven methodologies. In Proceedings of the 2012 IEEE Conference on Prognostics and Health Management,
Denver, CO, USA, 18–21 June 2012; Volume 2, pp. 1–7.
8. Goebel, K.; Saha, B.; Saxena, A. A comparison of three data-driven techniques for prognostics. In Proceedings of the
62nd Society for Machinery Failure Prevention Technology, Virginia Beach, VA, USA, 6–8 June 2008; pp. 1–13.
9. Loukopoulos, P.; Zolkiewski, G.; Bennett, I.; Sampatha, S.; Pilidis, P.; Li, X.; Mba, D. Abrupt fault remaining
useful life estimation using measurements from a reciprocating compressor valve failure. Mech. Syst.
Signal Process. 2019, 121, 359–372. [CrossRef]
10. Kim, H.-E.; Andy, C.C.; Mathew, J.; Eric, Y.H.; Choi, B.-K. Machine Prognostics Based on Health State
Estimation Using Svm. In Proceedings of the Third World Congress on Engineering Asset Management and
Intelligent Maintenance Systems Conference, Beijing, China, 27–30 October 2008; pp. 834–845.
11. Zhang, S.; Ganesan, R.; Xistris, G.D. Self-organising neural networks for automated machinery monitoring
systems. Mech. Syst. Signal Process. 1996, 10, 517–532. [CrossRef]
12. Mandal, D.; Pal, S.K.; Saha, P. Modeling of electrical discharge machining process using back propagation
neural network and multi-objective optimization using non-dominating sorting genetic algorithm-II. J. Mater.
Process. Technol. 2007, 186, 154–162. [CrossRef]
13. Elasha, F.; Mba, D.; Togneri, M.; Master, I.; Teixeira, J.A. A hybrid prognostic methodology for tidal turbine
gearboxes. Renew. Energy 2017, 114, 1051–1061. [CrossRef]
14. Li, X.; Duan, F.; Loukopoulos, P.; Bennett, I.; Mba, D. Canonical variable analysis and long short-term memory
for fault diagnosis and performance estimation of a centrifugal compressor. Control Eng. Pract. 2018, 72, 177–191.
[CrossRef]
15. Cheng, F.; Qu, L.; Qiao, W. Fault Prognosis and Remaining Useful Life Prediction of Wind Turbine Gearboxes
Using Current Signal Analysis. IEEE Trans. Sustain. Energy 2018, 9, 157–167. [CrossRef]
16. Elforjani, M.; Shanbr, S. Prognosis of Bearing Acoustic Emission Signals Using Supervised Machine Learning.
IEEE Trans. Ind. Electron. 2017, 0046, 5864–5871. [CrossRef]
17. Heng, R.B.W.; Nor, M.J.M. Statistical analysis of sound and vibration signals for monitoring rolling element
bearing condition. Appl. Acoust. 2002, 53, 211–226. [CrossRef]
18. Zhu, J.; Nostrand, T.; Spiegel, C.; Morton, B. Survey of Condition Indicators for Condition Monitoring
Systems. In Proceedings of the Annual Conference of the Prognostics and Health Management Society,
Fort Worth, TX, USA, 29 September–2 October 2014; Volume 5, pp. 1–13.
 Energies 2019, 12, 2705 17 of 17

19. Eftekharnejad, B.; Carrasco, M.R.; Charnley, B.; Mba, D. The application of spectral kurtosis on Acoustic
Emission and vibrations from a defective bearing. Mech. Syst. Signal Process. 2011, 25, 266–284. [CrossRef]
20. Wang, L.; Gao, R.X. Condition Monitoring and Control for Intelligent Manufacturing; Springer: London, UK, 2006.
21. Bates, D.M.; Watts, D.G. Nonlinear Regression Analysis; Wiley: New York, NY, USA, 2007; 392p.
22. Mahamad, A.K.; Saon, S.; Hiyama, T. Predicting remaining useful life of rotating machinery based artificial
neural network. Comput. Math. Appl. 2010, 60, 1078–1087. [CrossRef]
23. Levenberg, K.; Arsenal, F. A Method for the Solution of Certain Non-Linear Problems in Least Squares.
Q. Appl. Math. 1943, 1, 536–538. [CrossRef]
24. Bechhoefer, E.; van Hecke, B.; He, D. Processing for Improved Spectral Analysis. In Proceedings
of the Annual Conference of Prognostics and Health Management Society, New Orleans, LU, USA,
14–17 October 2013; pp. 1–6.

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).

View publication stats