sensors

Article
Deep Learning Approach for Vibration Signals Applications
Han-Yun Chen 1,2 and Ching-Hung Lee 3,4, *

1 Department of Mechanical Engineering, National Chung Hsing University, Taichung City 402, Taiwan;
Hamilton.HY.Chen@auo.com
2 AU Optronics Corporation, Taichung 407, Taiwan
3 Department of Electrical and Computer Engineering, National Yang Ming Chiao Tung University,
Hsinchu City 300, Taiwan
4 Department of Electrical and Computer Engineering, National Chiao Tung University, Hsinchu City 300, Taiwan
* Correspondence: chleenctu@nctu.edu.tw; Tel.: +886-3-5712121 (ext. 54315)

Abstract: This study discusses convolutional neural networks (CNNs) for vibration signals analysis,
including applications in machining surface roughness estimation, bearing faults diagnosis, and tool
wear detection. The one-dimensional CNNs (1DCNN) and two-dimensional CNNs (2DCNN) are
applied for regression and classification applications using different types of inputs, e.g., raw signals,
and time-frequency spectra images by short time Fourier transform. In the application of regression
and the estimation of machining surface roughness, the 1DCNN is utilized and the corresponding
CNN structure (hyper parameters) optimization is proposed by using uniform experimental design
(UED), neural network, multiple regression, and particle swarm optimization. It demonstrates the
effectiveness of the proposed approach to obtain a structure with better performance. In applications
of classification, bearing faults and tool wear classification are carried out by vibration signals
analysis and CNN. Finally, the experimental results are shown to demonstrate the effectiveness and
performance of our approach.






Keywords: vibration signal; deep learning; convolutional neural network; hyper parameter; opti-
Citation: Chen, H.-Y.; Lee, C.-H. mization; short time Fourier transform
Deep Learning Approach for
Vibration Signals Applications.
Sensors 2021, 21, 3929. https://
doi.org/10.3390/s21113929 1. Introduction
Vibration signals can be applied for machine diagnosis and help discover problems
Academic Editor: Jongmyon Kim
during machining. By the signal processing methods, the signals can be decomposed and
transformed into different domains for analysis, e.g., fast Fourier transform, wavelet trans-
Received: 10 May 2021
Accepted: 3 June 2021
form, etc. [1–8]. Statistical features and other characteristics related to physical phenomena
Published: 7 June 2021
are then extracted for applications. Based on data analysis, machine learning approaches
model the relationship of features and physical phenomena. The corresponding features
Publisher’s Note: MDPI stays neutral
are usually extracted by statistical analysis in time and frequency domains.
with regard to jurisdictional claims in
In mechanical systems, rolling element bearings (REBs) are one of crucial components
published maps and institutional affil- and the bearing failures can cause safety problems. A lot of the literature has proposed the
iations. diagnosis of bearings or building monitoring systems with machine learning models, e.g.,
support vector machines (SVMs), neural networks (NNs) [9–14]. Recently, deep learning
approaches were proposed to auto extract the characteristics of vibration signals for signals
analysis [9,12–14]. For signals analysis, methods of frequency spectra can also be used for
Copyright: © 2021 by the authors.
prediction or diagnosis [15,16]. The statistical features are usually utilized to be inputs of
Licensee MDPI, Basel, Switzerland.
machine learning for diagnosis model development [17–19]. Herein, the convolutional
This article is an open access article
neural network (CNN) discussed in this paper is also widely applied for bearing diagnosis
distributed under the terms and using raw signals or spectra of signals [20–26].
conditions of the Creative Commons The condition of machine tools affects the quality and the productivity directly. A
Attribution (CC BY) license (https:// blunt tool can cause terrible quality since the magnitude of vibration during machining
creativecommons.org/licenses/by/ increases. Excessive tool wear can even lead to tool breakages. The diagnoses of tool status
4.0/). were proposed by on-line and off-line monitoring [27–31]. For off-line monitoring, the tools

Sensors 2021, 21, 3929. https://doi.org/10.3390/s21113929 https://www.mdpi.com/journal/sensors

Sensors 2021, 21, 3929 2 of 17

are dismounted to measure the worn area. However, the machines need to stop in order
to measure tool wear. In the on-line approach, the status of a tool can be predicted using
vibration, acoustic emission, and force signals of vises and machine tools [27–29]. Due to
the improvement of photographic techniques, on-line monitoring can also be implemented
using high speed cameras in some machines [30,31]. In addition to the status of machines,
predicting the quality of products is a valuable topic for the industries. If the quality can be
estimated, the whole manufacturing process can be controlled easily. Predicting quality
using machining parameters is discussed in many studies. Machine learning algorithms
are applied to model the relation between machining parameters and quality; for instance,
fuzzy logic [32], response surface methodology [33], etc. The main disadvantage of using
machining parameters is that the statuses of tools and machines are not considered. Since
vibrations affect quality, the vibration signals can be analyzed and applied to estimate
quality [2,34,35]. Sensor fusion has also been proposed in other studies; for instance,
multiple vibration sensors [36–38], vibration with acoustic signals [39,40] or load cell [41],
etc. The sensors can be seen as evidence for fault detection. In other words, different types
of sensors can provide different symptoms when the components fail. Fusion in feature
domain and frequency domain are also discussed in other studies [42,43].
Deep learning approaches provide automatic feature extractions; for instance, a con-
volutional neural network (CNN) [40]. Applications of a CNN in vibration signals are
discussed in lots of research, including bearing faults diagnosis, tool wear classification
and machining roughness estimation. By employing convolutional operation, the fea-
tures can be extracted automatically [44–48]. One-dimensional CNNs (1DCNN) and
two-dimensional CNNs (2DCNN) are used in the domain of REB signals prediction. For
1DCNN applications, the inputs are raw signals or other one-dimensional data [20,25].
If 2DCNN is utilized, the inputs should be chosen as time-frequency spectra or other
two-dimensional data or images [21,26,49,50].
In this study, CNNs for vibration signals analysis are discussed. Firstly, 1DCNN with
sensor fusion in parallel structure is introduced for machining roughness estimation. The
model structure (hyper parameters) optimization of the CNN is proposed by experimen-
tal design, data acquisition, neural network modeling, and particle swarm optimization.
Subsequently, CNNs for bearing faults classification and tool wear classification are dis-
cussed later. According to the results of applications, the conclusions for utilizing CNNs in
vibration signals analysis can be presented.
In the rest of paper, the applied techniques are introduced in Section 2, prediction using
CNNs and structure optimization are introduced in Section 3, CNNs for classifications are
discussed in Section 4, and the conclusion of the study is presented in Section 5, finally.

2. Theoretical Background
Herein, techniques utilized in the study are introduced, including short-time Fourier
transform, convolutional neural networks and particle swarm optimization.

2.1. Convolutional Neural Network (CNN)

The CNN was first proposed by Lecun et al. [51] and the structure of the CNN is
shown in Figure 1. The three basic operations in the CNN are convolutional layers, pooling
layers, and fully connected layers. Convolutional layers and pooling layers are adopted
for automatic feature extraction when fully connected layers are general neural networks
which play the roles of classifier or predictor.
At first, the convolutional layer is introduced, and the inputs are convolved by filters
to obtain the corresponding features. The convolutional operation of single filter can be
represented as
zkl = f c (αl ∗ x + b) (1)
where * represents the convolutional operation; x ∈ RW × L denotes the input and f c denotes
the activation function of convolution layer; b and αl are the bias and corresponding kernel
Sensors 2021, 21, 3929 3 of 17

Sensors 2021, 21, x 3 of 18

of the lth filter, respectively; zkl denotes the corresponding output feature map. Herein,
kernel matrix αl are obtained by training and l = 1, . . . , N is the selected kernel size.

(Softmax)

Figure 1. Structure
Figure1. Structure of
ofconvolutional
convolutionalneural
neuralnetwork.
network. Reprinted
Reprintedfrom
fromref.
ref.[47].
[47].

In
Atpooling
first, thelayers, the important
convolutional layer features
is introduced,are reserved,
and theand inputs the are number of features
convolved are
by filters
reduced
to obtain the corresponding features. The convolutional operation of single filter can as
by a max-pooling operation. The operation of a single filter can be represented be
represented as
zkl q,r+1 . . .
  
zk zkl q,r+ L
 k l q,r P
= ( ∗ + ) k (1)

 zl .. z l q+1,r + L P  
q+1,r

k .
pl q,r = max (2)

.. × ..
operation; ∈

where * represents the  convolutional


 . denotes . the 
input
 and fc de-

notes the activation function zof k convolution z k layer; b and . . . z are
k the bias and correspond-
l q+WP ,r l q + W P ,r + 1 l q+WP ,r + L P
ing kernel of the lth filter, respectively; denotes the corresponding output feature map.
Herein, kernel matrix
where q and r are the row and column index of features after pooling, L Pthe
are obtained by training and l = 1, …, N is andselected kernel
WP represent
size.
the length and width of filters in pooling layers.
In pooling
The feature layers, the important
maps after featuresare
feature extraction areflattened
reserved,into andathe number of features
one-dimension array andare
reduced into
inputted by a fully
max-pooling
connected operation.
layers. The Thefeedforward
operation ofoperation a single filter can beneuron
of a single represented
in fullyas
connected layers is represented as
,
…
, ! ,
n
,
= max y = f f,
⋮
∑ wa ha +
⋱b
⋮
, (2)
(3)
a =1
,
…
, ,
where h a is the input of the neuron, wa is weight of h a , a = 1, 2, . . . , n, b is the bias, f f is the
where and
activation r are of
function thethe
row and column
neuron index
in the fully of features
connected after
layer, y ispooling,
the output of andthe CNN. rep-
resent the length and width of filters in pooling layers.
2.2. Short-Time Fourier
The feature mapsTransform (STFT)
after feature extraction are flattened into a one-dimension array
Discrete Fourier
and inputted transform
into fully (DFT)
connected is widely
layers. applied to generate
The feedforward operation frequency spectra
of a single neuronof sig-
in
nals.
fully However,
connectedfrequency spectra do not
layers is represented as contain the information of time domain. In order
to present time domain and frequency domain at the same time, STFT is employed [8,52].
In STFT, signals are divided into short-time segments firstly, and frequency distributions
of segments are computed by DFT. Finally, = +
the time-frequency spectra of signals can be (3)
obtained by stacking the frequency spectra of segments. STFT can be represented as

where is the input of the neuron,

 N −1 of
is weight , = 1, 2, … , , b is the bias,
STFT( x [n]) ≡ X m, e− jω = ∑ x [n]w[n − m]e− jωn (4)
is the activation function of the neuron in the fully
n =0
connected layer, y is the output of the
CNN.
where x is the discrete signal with size N, ω is frequency, n is the index of data points in
x, w Short-Time
2.2. is discrete Fourier
window m is discrete index in the window w. STFT is applied
function,(STFT)
Transform
as theDiscrete
preprocessor of signals in the
Fourier transform (DFT) study. The time-frequency
is widely spectra
applied to generate are the spectra
frequency inputs of
of
signals. However, frequency spectra do not contain the information of time domain. In
order to present time domain and frequency domain at the same time, STFT is employed
Sensors 2021, 21, 3929 4 of 17

convolutional neural networks, which is introduced in the following section. Note that the
axes of spectra are removed when input into the model.

2.3. Particle Swarm Optimization (PSO)

Particle swarm optimization (PSO), simulating the social behaviors of fish and birds
while foraging, was proposed in 1998 [53]. Firstly, the fitness function and the target of
optimization are defined. By fitness function, the score of particles can be evaluated. The
particles adjust their directions and locations according to the best location of the group
and themselves using
   
Vi (t + 1) = w × Vi (t) + random × c1 × Ppbest − Pi (t) + random × c2 × Pgbest − Pi (t) (5)

and
Pi (t + 1) = Pi (t) + Vi (t + 1) (6)
respectively, where Vi is the direction of the ith particle, t represents the index of itera-
tion, w is the weight of inertia, c1 is the weight representing how much Ppbest affects the
optimization, c2 is the weight representing how much Pgbest affects the optimization, Pi (t)
represents the location of the ith particle at the tth iteration. Finally, while reaching the
set maximum of the iteration or the fitness of Pgbest remains the same, the optimization is
complete and Pgbest is the optimized result. In this study, the minimized mean absolute per-
centage error (MAPE) of prediction is adopted to be the objective function for optimization
of hyper parameters.

3. Machining Roughness Estimation Application

In this section, machining surface roughness estimation is achieved using the CNN.
The optimization of the CNN structure is also discussed. Firstly, the dataset is introduced.
Then, the experimental design is carried out and executed. After the experiments are
complete, a simple neural network (NN) is applied to model the relation between hyper
parameters and the performance of model. Optimization using PSO is then discussed. The
optimized results are verified, finally.
At first, the optimization of the model structure is introduced.

3.1. Optimization of Model Structure

Herein, the concept of optimizing the model structure (hyper parameters) is uti-
lized [54]. An improvement by uniform experimental design (UED) [55], a neural network,
and a PSO algorithm is introduced. It preserves the ability of the CNN and optimizes the
performance. The procedure of optimization is introduced. The flow chart of optimization
procedure is shown as Figure 2. The procedures include (1) parameter selection of the
CNN, (2) experimental design using UED, (3) data acquisition, (4) model development, (5)
optimization, and finally, (6) validation.

Optimization Procedure
Step 1. Parameter selection of CNN: Select the main structure (convolution filter size,
pooling, fully connected nodes), the optimized hyper parameters, and levels.
Step 2. Design experiments using UED: Choose the appropriate uniform layout (UL)
of model structure according to the parameter selection and design experiments.
Step 3. Data acquisition: Complete the experiments. The model with the above
structure is trained and the corresponding hyper parameters/trained MAPE are collected
as input/output data.
Step 4. Model development: Modeling the function between hyper parameters and
performance using neural network. The performance applied in this study is MAPE.
Step 5. Optimization: Obtain the hyper parameter combination with better perfor-
mance using PSO. In this study, the goal of optimization is to minimize the MAPE of
the CNN.
Step 6. Verification: Verify the performance of the optimized result.
 Sensors 2021, 21, 3929 5 of 17

Sensors 2021, 21, x In this study, a simple neural network is applied for the model and particle swarm
5 of 18
optimization (PSO) is adopted for optimization to compare with MR and the full-factorial
searching algorithm [54].

Figure 2. Flow chart of the proposed optimization procedure.

Figure 2. Flow chart of the proposed optimization procedure.
3.2. Surface Roughness Estimation Using CNN
Optimization Procedure
Data of milling are proposed by Wu et al. using a tungsten carbide milling cutter to
cutStep
S45C1. Parameter
steel [34]. selection
There areofsix
CNN: Select theaccelerometers
single-axial main structure(Wilcoxon
(convolution filter size,
Research 785A)
pooling,
mounted fullyon connected nodes),
the spindle and the
viseoptimized hyperX-axial,
for measuring parameters, andand
Y-axial, levels.
Z-axial vibration
Step 2.The
signals. Design experiments
signals using
are acquired UED:
using DAQ Choose the appropriate
NI 9234 with 10 kHzuniform
of sampling layout (UL)
frequency.
of model structure according to the parameter selection and design experiments.
The experimental setup can be found in [34]. The surface roughness is measured using
Step 3. Data
Mitutoyo acquisition:
SV-C3200S4. TheComplete
machining theparameters
experiments. The
and model
setup with are:
values the above
spindle struc-
speed
ture
(rpm)—900, 1000, 1800, 1900, 2000, 2100, 2700, 3000 (rpm); feed rate—228, 240, 252,as320,
is trained and the corresponding hyper parameters/trained MAPE are collected
input/output
400, 420, 532, data.
560, 588 (mm/min); cutting depth—0.5, 0.6, 0.7, 0.8, 0.9, 1 (mm); and clamp
Stepof4.vise—18,
force Model development:
30, 75 (N-m). Modeling
There are the function
a total of 153between
data in thehyper parameters
dataset. and
The complete
performance using neural
data are available on thenetwork. The performance applied in this study is MAPE.
website [34].
StepA 5. Optimization: CNN
one-dimensional Obtain the hyper
(1DCNN) withparameter combination
sensors fusion with
in parallel better perfor-
structure, shown in
mance
Figure 3, is applied for machining roughness estimation. The features of vibrationofsignals
using PSO. In this study, the goal of optimization is to minimize the MAPE the
CNN.
in X, Y, Z directions are extracted separately. In order to obtain a CNN structure with
Step performance,
better 6. Verification:theVerify the performance
optimization for hyper of the optimized
parameters result.
combination is applied [52].
TheIn range of optimized
this study, a simplehyper
neuralparameters
network is and the structure
applied of theand
for the model CNN are selected
particle swarm as
shown in Table
optimization (PSO)1.isAccording to Table
adopted for 1, there are
optimization six design
to compare factors:
with MR andFC for
thethe size of filters
full-factorial
in convolutional
searching [54]. FP for the size of filters in pooling layers, NC1 for the filter number
algorithmlayers,
in the first convolutional layer, NC2 for the filter number in the second convolutional layer,
3.2.NSurface
F1 for the numberEstimation
Roughness of nodes in CNNfully connected layer, and NF2 for the number of
the first
Using
nodes in the second fully connected
Data of milling are proposed by Wu et al. layer. The feature
using extraction
a tungsten for three
carbide axialcutter
milling signalstoare
cut S45C steel [34]. There are six single-axial accelerometers (Wilcoxon Research 785A)
mounted on the spindle and vise for measuring X-axial, Y-axial, and Z-axial vibration sig-
nals. The signals are acquired using DAQ NI 9234 with 10 kHz of sampling frequency.
The experimental setup can be found in [34]. The surface roughness is measured using
Mitutoyo SV-C3200S4. The machining parameters and setup values are: spindle speed
(rpm)—900, 1000, 1800, 1900, 2000, 2100, 2700, 3000 (rpm); feed rate—228, 240, 252, 320,
 axial signals are the same. The performance of the model is assumed as a function of hyper
parameters, which is represented as

MAPE = ( , , , , , ) (7)
Sensors 2021, 21, 3929 According to UED [49], four levels are selected for all factors and the corresponding 6 of 17

uniform layout applied here is (4 ), as shown as Table 2. The final experimental de-
sign is introduced in Table 3. The corresponding combinations of parameters and trained
MAPE
the (average
same. testing MAPE
The performance of of
thecorresponding
model is assumed experimental CNNs)
as a function ofare alsoparameters,
hyper introduced.
Every structure has been
which is represented as tested three times and the average MAPEs are computed. The
maximum epoch of each model is 700. In order to reduce the needed time for experiments,
an early stop criterion is set=upf MAPE
MAPE according
( FC , Fto
P , testing
NC1 , NC2experiences:
, NF1 , NF2 )if the loss has not de-
(7)
creased for 15 epochs, the training process is stopped.

Figure 3. The
Figure 3. The sensors
sensors fusion
fusion structure
structure for
for machining
machining surface
surface roughness
roughness estimation.
estimation.

Table 1. Hyper parameters of CNN for machining surface roughness estimation.

Activation
Layers Filter Size Stride Number of Filters or Nodes
Function
Conv. 1 (X, Y, Z) FC (16~25) 2 NC1 (11~20) ReLU
Pool. 1 (X, Y, Z) FP (11~20)
Conv. 2 (X, Y, Z) FC (16~25) 2 NC2 (11~20) ReLU
Pool. 2 (X, Y, Z) FP (11~20)
Flatten
Fully connected 1 NF1 (10~100) ReLU
Fully connected 2 NF2 (10~100) ReLU
Output 1 None

According to UED [49], four levels

are selected for all factors and the corresponding
uniform layout applied here is U28 46 , as shown as Table 2. The final experimental design
is introduced in Table 3. The corresponding combinations of parameters and trained MAPE
(average testing MAPE of corresponding experimental CNNs) are also introduced. Every
structure has been tested three times and the average MAPEs are computed. The maximum
epoch of each model is 700. In order to reduce the needed time for experiments, an early
stop criterion is set up according to testing experiences: if the loss has not decreased for
15 epochs, the training process is stopped.
Sensors 2021, 21, 3929 7 of 17

Table 2. U28 46 uniform layout.

Experiment Factors
Index FC FP NC1 NC2 NF1 NF2
1 1 3 2 3 4 3
2 4 4 3 2 4 2
3 2 3 3 3 3 2
4 1 2 1 4 4 2
5 2 2 3 1 2 3
6 1 4 1 1 1 3
7 3 1 3 4 2 1
8 3 3 3 1 1 4
9 1 2 3 2 1 1
10 3 4 2 2 2 3
11 4 2 4 2 3 3
12 2 1 1 3 1 3
13 4 1 3 4 4 3
14 2 4 4 1 4 1
15 1 1 4 3 2 2
16 3 1 1 2 1 2
17 3 2 1 1 3 4
18 4 3 4 1 2 2
19 1 3 4 4 3 4
20 4 4 1 3 3 4
21 4 2 2 3 2 4
22 4 3 2 4 1 1
23 3 2 4 3 4 1
24 2 3 1 2 2 1
25 2 1 2 2 4 4
26 1 1 2 1 3 1
27 3 4 2 4 3 2
28 2 4 4 4 1 4

After the experiments, the function between hyper parameters and average testing
MAPE is modeled using MR and NN for comparison. The performance of models, op-
timization results, and verifications are compared as follows. The data are normalized
before modeling.
At first, modeling using stepwise MR is obtained as

MAPE = 35.818395 − 1.215402FC − 0.428033FP + 0.758975NC1 + 0.991905NC2

+0.140401NF1 − 0.224964NF2 + 0.001241FC NF2 + 0.053019FC NC2
(8)
−0.046696FP NC2 + 0.01539FP NF2 − 0.070553NC1 NC2 − 0.000967NC1 NF2
−0.010024NC2 NF1 − 0.00065NF1

The corresponding R-squared (R2 ) of MR model is 0.9061 and the normalized root
mean squared error (NRMSE) of MR is 0.0634. The objective function (fitness) is selected
as the MAPE of each structure. The optimization target is to minimize the fitness. The
hyper parameters combination optimized using the full-factorial searching algorithm are:
FC = 25, FP = 20, NC1 = 20, NC2 = 20, NF1 = 100, NF2 = 10. The testing MAPE prediction
of the MR model for the combination is 5.788%. The structure with the optimized hyper
parameters combination has been trained three times. The testing MAPEs are shown in
Table 4. The average MAPE is quite different to the prediction, with an error of 147.06%.
The combination does not perform better compared to the experiments.
Sensors 2021, 21, 3929 8 of 17

Table 3. Experimental design of CNN structure for estimating machining roughness and average testing MAPE of
corresponding experimental CNNs.

Experiment Avg. Testing

FC FP NC1 NC2 NF1 NF2 Parameters
Index MAPE (%)
1 16 17 14 17 100 70 60,230 14.35
2 25 20 17 14 100 40 44,399 13.57
3 19 17 17 17 70 40 49,055 16.00333333
4 16 14 11 20 100 40 87,362 18.42666667
5 19 14 17 11 40 70 30,533 18.3
6 16 20 11 11 10 70 8903 23.83333333
7 22 11 17 20 40 10 69,734 25.16
8 22 17 17 11 10 100 17,399 23.11333333
9 16 14 17 14 10 10 17,504 24.25666667
10 22 20 14 14 40 70 25,325 19.11
11 25 14 20 14 70 70 60,053 15.17333333
12 19 11 11 17 10 70 21,911 25.44
13 25 11 17 20 100 70 148,127 11.33666667
14 19 20 20 11 100 10 31,394 18.82666667
15 16 11 20 17 40 40 57,872 18.46333333
16 22 11 11 14 10 40 19,436 21.03
17 22 14 11 11 70 100 43,769 18.4
18 25 17 20 11 40 40 29,054 16.59333333
19 16 17 20 20 70 100 61,151 13.68333333
20 25 20 11 17 70 100 40,055 18.50333333
21 25 14 14 17 40 100 45,674 18.52333333
22 25 17 14 20 10 10 26,483 19.17333333
23 22 14 20 17 100 10 86,192 16.02333333
24 19 17 11 14 40 10 23,381 19.54333333
25 19 11 14 14 100 100 102,155 15.81
26 16 11 14 11 70 10 52,820 28.36333333
27 22 20 14 20 70 40 43,457 15.21333333
28 19 20 20 20 10 100 28,271 18.87666667

Table 4. Testing MAPEs of the optimized hyper parameters combination using MR model.

Test MAPE 1 Test MAPE 2 Test MAPE 3 Avg. MAPE Standard Deviation
15.74% 13.97% 13.19% 14.3% 1.090%

Then, an NN is applied to model the relation between factors and testing MAPE. The
structure of NN is shown in Table 5. The initial learning rate is 0.005, and the optimizer
is Adam. The R-squared (R2 ) of NN is 0.9999999996 and the normalized root mean
squared error (NRMSE) of the NN is 3.347 × 10−5 . The hyper parameters combination
optimized using the full-factorial searching algorithm are: FC = 25, FP = 11, NC1 = 18,
NC2 = 12, NF1 = 100, NF2 = 50. The testing MAPE prediction of the NN model for the
combination is 10.849%. The combination has also been trained three times. The testing
MAPEs are shown in Table 6. The error between the average MAPE and prediction of the
NN model is much smaller, with an error of 7.337%. The optimized structure improves
the performance by 11.3%. The results show that modeling using NN can also create a
better and more stable hyper parameters combination than the best hyper parameters set
in the experiments. However, the structure, learning rate, and normalization affect the
performance of modeling and optimized result a lot. A simple NN with a smaller learning
rate is recommended in this case. Normalization is also necessary.
Herein, PSO is applied for optimization to compare with the full-factorial searching
algorithm. Modeling using an NN is applied for comparison. The number of particles is
selected as 250, and the number of iterations is set to be 3000. The reason for choosing
this number of particles and iteration is to ensure the optimized result is the same as the
result using the full-factorial searching algorithm. The weights of updating velocity are
Sensors 2021, 21, 3929 9 of 17

adjusted shown in Table 7. If the fitness of Pgbest does not improve for 500 iterations, the
optimization is stopped.

Table 5. Structure of NN for modeling the function between factors and testing MAPE.

Layer Nodes Activation Function Bias

Input 6 None None
Hidden 1 12 Sigmoid None
Output 1 None Yes
Total parameters 85

Table 6. Testing MAPEs of the optimized hyper parameters combination using NN model.

Test MAPE 1 Test MAPE 2 Test MAPE 3 Avg. MAPE Standard Deviation
11.04% 10.68% 8.44% 10.053% 1.150%

Table 7. Adjustment details of weights while updating velocity.

Weights of Updating Velocity Range of Values Adjustment of Weights

w 0.1~2 Decrease while the iteration increases.
c1 0.1~2 Decrease while the iteration increases.
c2 0.1~2 Increase while the iteration increases.

The fitness during optimizing using PSO is shown as Figure 4. The optimized result
is the same as the full-factorial searching algorithm. Moreover, PSO takes 45.435 s to
complete the process, while it takes 146.87 s for the full-factorial searching algorithm. If the
number of particles and iterations are reduced according to the testing results, the time
for optimization can be less than the previous experiment result. When the structure of
Sensors 2021, 21, x the optimized CNN is more complex, the computing time for PSO and other optimization 10 of 18
methods are much less compared to the time for the full-factorial searching algorithm.

Figure 4. Fitness during optimization.

4. Fault
Table Diagnosis details
7. Adjustment Applications
of weights while updating velocity.
4.1. Classification of CWRU Bearing Data
Weights of Updating Velocity Range of Values Adjustment of Weights
Bearing data of CWRU [56] are discussed in many other studies for bearing fault
w 0.1~2 Decrease while the iteration increases.
classification [57–59]. The signals discussed in the study are collected by the accelerometer
0.1~2 Decrease while the iteration increases.
0.1~2 Increase while the iteration increases.

4. Fault Diagnosis Applications

4.1. Classification of CWRU Bearing Data
Sensors 2021, 21, 3929 10 of 17

mounted at the drive end of motor. The sampling frequency is 12 kHz. The bearing statuses
include normal bearings, bearings with inner ring faults, bearings with outer ring faults,
and bearings with ball faults, which are human-made using an electrical-discharge machine
(EDM). The statuses of bearings are labeled according to normal: 0; inner ring fault:1; outer
ring fault: 2; and ball fault: 3, respectively. There are 64 data in the original dataset. In
order to increase the number of data, sliding window is utilized to slice the signals into
one-second signals. The length and the stride of window are 12,000 data points (1 s) and
3000 data points, respectively. The length of window is selected after considering the
completeness of signals in the frequency domain and the testing results. Finally, there are
2368 data; 1657 data (70%) are chosen randomly as training data and the rest (30%) are
applied as testing data.
(a) Bearing Faults Classification Using Vibration Signals
Herein, we introduce the classification of bearing faults using 1DCNN with vibration
signals as inputs. The selected structure of 1DCNN is introduced in Table 8. The initial
learning rate is 0.001, and the optimizer is Adam. The average of training and testing
accuracy of the model are both 100% after testing three times using different training data.
The confusion matrix of the model predicting testing data is shown in Figure 5. The result
shows that 1DCNN can provide excellent performance using vibration signals as inputs
directly for classification. The classifying time of 1DCNN using NVIDIA Tesla V100 32 GB
GPU is 0.00133 s per data.

Table 8. Structure of 1DCNN for bearing faults classification using vibration signals.

Number of Activation
Layer Filter Size Stride
Filters or Nodes Function
Conv. 1 30 1 8 ReLU
Pool. 1 4
Conv. 2 30 1 16 ReLU
Pool. 2 4
Conv. 3 30 1 32 ReLU
Pool. 3 4
Conv. 4 30 1 64 ReLU
Pool. 4 4
Flatten
Fully Conn. 1 128 ReLU
Fully Conn. 2 32 ReLU
Output 4 Softmax
Total parameters 388,488

(b) Bearing Faults Classification Using STFT Time-Frequency Spectra

The time-frequency spectra after STFT of different bearing conditions are shown in
Figure 6. A 2DCNN is applied to classify the bearing faults. The structure of the CNN is
shown as Table 9. The initial learning rate is 0.001 with the Adam optimizer. The average of
training and testing accuracy are both 100% after testing three times. The confusion matrix
of the model for testing data is shown as Figure 7. The result shows that 2DCNN can also
be applied for the classification of bearing faults with great performance. The inputs of
2DCNN can be other types of two-dimensional arrays, e.g., time-frequency spectra using
wavelet transform. The transformation time using STFT is 0.75258 s per data, and the
classifying time of 2D CNN using NVIDIA Tesla V100 32 GB GPU is 0.00419 s per data.
Classification using 2DCNN takes more time due to the input size of the model. 1DCNN
uses raw signals as inputs; the input size is 12,000 × 1. 2DCNN uses STFT time-frequency
spectra as inputs; the input size is 434 × 558 × 3.
 Flatten
Fully Conn. 1 128 ReLU
Fully Conn. 2 32 ReLU
Output 4 Softmax
Sensors 2021, 21, 3929 11 of 17
Total
388,488
parameters

Sensors 2021, 21, x 12 of 18

Figure
Figure 5.5. Confusion matrix of 1DCNN model for classifying
classifying CWRU
CWRU bearing
bearingdata.
data.Reprinted
Reprinted from
from
ref.
ref. [47].
[47].

(b) Bearing Faults Classification Using STFT Time-Frequency Spectra

The time-frequency spectra after STFT of different bearing conditions are shown in
Figure 6. A 2DCNN is applied to classify the bearing faults. The structure of the CNN is
shown as Table 9. The initial learning rate is 0.001 with the Adam optimizer. The average
of training and testing accuracy are both 100% after testing three times. The confusion
matrix of the model for testing data is shown as Figure 7. The result shows that 2DCNN
can also be applied for the classification of bearing faults with great performance. The
inputs of 2DCNN can be other types of two-dimensional arrays, e.g., time-frequency spec-
tra using wavelet transform. The transformation time using STFT is 0.75258 s per data,
and the classifying time of 2D CNN using NVIDIA Tesla V100 32 GB GPU is 0.00419 s per
data. Classification using 2DCNN takes more time due to the input size of the model.
1DCNN uses raw signals as inputs; the input size is 12,000 × 1. 2DCNN uses STFT time-
(a) (b)
frequency spectra as inputs; the input size is 434 × 558 × 3.

(c) (d)
STFTtime-frequency
Figure6.6.STFT
Figure time-frequencyspectra
spectraofofdifferent
differentbearing
bearingconditions,
conditions,(a)
(a)aanormal
normalbearing;
bearing;(b)
(b)aa
bearingwith
bearing withinner
innerring
ringfault;
fault;(c)
(c)aabearing
bearingwith
withouter
outerring
ringfault;
fault;(d)
(d)aabearing
bearingwith
withball
ballfault
fault[47].
[47].

Table 9. Structure of CNN for classifying bearing faults.

Number of Filters or Activation

Layer Filter Size Stride
Nodes Function
Conv. 1 4 ReLU
9×9 2×2
Conv. 2 8 ReLU
Pool. 2 4×4
Conv. 3 16 ReLU
4×4 2×2
Sensors 2021, 21, 3929 12 of 17

Table 9. Structure of CNN for classifying bearing faults.

Number of Activation
Layer Filter Size Stride
Filters or Nodes Function
Conv. 1 4 ReLU
9×9 2×2
Conv. 2 8 ReLU
Pool. 2 4×4
Conv. 3 16 ReLU
4×4 2×2
Conv. 4 32 ReLU
Pool. 4 2×2
Flatten
Fully Conn. 1 64 ReLU
Fully Conn. 2 32 ReLU
Sensors 2021, 21, x Output 4 Softmax13 of 18
Total parameters 63,622

Figure7.7.Confusion
Figure Confusionmatrix
matrixof
ofCNN
CNNfor
forclassifying
classifyingbearing
bearingfaults
faults[47].
[47].

4.2.
4.2.Classification
ClassificationofofTool
ToolWear
WearUsing
UsingSTFT
STFTTime-Frequency
Time-FrequencySpectra
Spectra
The
The experimental setup is introduced in Figure 8; the tool
experimental setup is introduced in Figure 8; the toolwear
weardata
dataofofaatri-axial
tri-axial
milling
milling machine (CHMER HM4030L, Figure 8a) are applied in the study. Themachine
machine (CHMER HM4030L, Figure 8a) are applied in the study. The machine
tools
toolsare
areaatungsten
tungstencarbide
carbidemilling
millingcutter
cutterwith
withtwo
twoblades,
blades,as asshown
shownininFigure
Figure8b.
8b.The
The
diameter of the cutters is 6 mm. The work-pieces are S45C steel. The tri-axial accelerometer
diameter of the cutters is 6 mm. The work-pieces are S45C steel. The tri-axial accelerometer
(CTC
(CTCAC230)
AC230)isismounted
mountedon onthe
thespindle,
spindle,as
asshown
shownininFigure
Figure8c.8c.The
Thevibration
vibrationsignals
signalsare
are
acquired
acquired using DAQ NI PCIe-6361 with 100 kHz of sampling frequency. The toolwear
using DAQ NI PCIe-6361 with 100 kHz of sampling frequency. The tool wearisis
measured using a Deryuan RS-500 industrial camera with ImageJ and PhotoImpact for
measured using a Deryuan RS-500 industrial camera with ImageJ and PhotoImpact for
image processing. The tool worn criteria is selected as 0.4 mm according to ISO.
image processing. The tool worn criteria is selected as 0.4 mm according to ISO.
A 2DCNN with a small structure (shown in Table 10) is adopted for classifying tool
wear using STFT time-frequency spectra. The vibration signals are sliced using sliding
window to increase the size of data. The length and stride of window is 100,000 data points
(1 s) and 30,000 data points, respectively. The STFT time-frequency spectra using Y-axial
vibration signals of an unworn tool and a worn tool are shown in Figure 9. There are a
total of 742 data; half of the data are selected randomly as training data, and the rest are
testing data. Firstly, the classification model is trained. The initial learning rate is 0.001
with the Adam optimizer. The average training and testing accuracy are both 100% after
Sensors 2021,
Sensors 21,21,
2021, x 3929 13 of 17 14 of 18

(a) (b)

(c)
Figure 8.
Figure Experimental setup
8. Experimental setupforfor
tool wear
tool wearmonitoring, (a) CHMER
monitoring, HM4030L
(a) CHMER tri-axialtri-axial
HM4030L milling milling
ma- ma-
chine; (b) tungsten carbide milling cutter for the experiments; (c) setup of CTC AC230 on the
chine; (b) tungsten carbide milling cutter for the experiments; (c) setup of CTC AC230 on the spin-spindle.
dle.
A 2DCNN with a small structure (shown in Table 10) is adopted for classifying tool
wear using STFT time-frequency spectra. The vibration signals are sliced using sliding
window to increase the size of data. The length and stride of window is 100,000 data points
(1 s) and 30,000 data points, respectively. The STFT time-frequency spectra using Y-axial
vibration signals of an unworn tool and a worn tool are shown in Figure 9. There are a total
of 742 data; half of the data are selected randomly as training data, and the rest are testing
data. Firstly, the classification model is trained. The initial learning rate is 0.001 with the
Adam optimizer. The average training and testing accuracy are both 100% after testing three
times. The confusion matrix of the CNN model using testing data is shown in Figure 10.
The result shows that 2DCNN can be applied for not only bearing faults classification but
also other classified problems in vibration signals analysis.

(a) (b)

Figure 9. STFT time-frequency spectra of tools under different conditions, (a) an unworn tool; (b) a
worn tool.
Sensors 2021, 21, 3929 14 of 17
(a) (b)

Table 10. Structure of CNN for classifying tool wear.

Number of Activation
Layer Filter Size Stride
Filters or Nodes Function
Conv. 1 4 ReLU
9×9 2×2
Conv. 2 8 ReLU
Pool. 2 4×4
Conv. 3 16 ReLU
4×4 2×2
Conv. 4 32 ReLU
Pool. 4 2×2
Flatten
Fully Conn. 1 (c) 64 ReLU
Fully Conn. 2 32 ReLU
Sensors 2021, 21, x Figure 8. Experimental setup for tool wear monitoring, (a) CHMER HM4030L tri-axial milling ma-
15 of 18
Output 2 Softmax
chine; (b) tungsten carbide milling cutter for the experiments; (c) setup of CTC AC230 on the spin-
Total parameters 28,360
dle.

Table 10. Structure of CNN for classifying tool wear.

Number of Filters or Activation

Layer Filter Size Stride
Nodes Function
Conv. 1 4 ReLU
9×9 2×2
Conv. 2 8 ReLU
Pool. 2 4×4
Conv. 3 16 ReLU
4×4 2×2
Conv. 4 32 ReLU
Pool. 4 2×2
Flatten
Fully Conn. 1 64 ReLU
Fully Conn. 2 (a) 32 (b) ReLU
Figure9.9.Output
Figure STFTtime-frequency
STFT time-frequencyspectra
spectraof
oftools
toolsunder
underdifferent 2
different conditions,
conditions, (a) an
(a)
Softmax
an unworn
unworn tool; (b)
tool; (b) aa
Total
worntool.
worn tool.parameters 28,360

Figure 10. Confusion matrix

Figure 10. matrix of
of CNN
CNN for
for classifying
classifying tool
tool wear.
wear.

5. Conclusions
In this study, vibration signals analysis using CNN has been discussed, including an
improved optimization method for the structure of a CNN, 1DCNN and 2DCNN with
raw signals and STFT images, respectively. The experimental results were introduced to
illustrate that the CNN can be applied for both prediction and classification. In regression
application, a 1DCNN with parallel feature extracting structure was applied to estimate
Sensors 2021, 21, 3929 15 of 17

5. Conclusions
In this study, vibration signals analysis using CNN has been discussed, including an
improved optimization method for the structure of a CNN, 1DCNN and 2DCNN with
raw signals and STFT images, respectively. The experimental results were introduced to
illustrate that the CNN can be applied for both prediction and classification. In regression
application, a 1DCNN with parallel feature extracting structure was applied to estimate
machining roughness. The optimization of the CNN structure was also introduced and
used to demonstrate the effectiveness of the proposed approach to obtain a structure with
better performance. The most important factor in optimizing the structure of CNN is
to choose the correct method and level for the experimental design. The level can be
comprehended as the resolution experiments. If the level is too large, the number of
experiment results is too little to represent the real situation. On the other hand, the cost
of time will be enhanced due to the large number of experiments. Other experimental
design can also be applied; for instance, the Taguchi method. In classifications, 1DCNN
and 2DCNN are applied according to the inputs. Both 1DCNN and 2DCNN provide
excellent performance. The results also show that CNN can extract features in vibration
signals and time-frequency spectra automatically. While using raw signals as inputs, the
length of signal must be long enough to ensure the information of the signal is complete.
If time-frequency spectra are utilized as inputs, the resolution of STFT affects the model
since time-frequency spectra show the distribution of frequency with respect to time. If the
resolution is not appropriate, the information in the frequency domain will be reduced and
influence the performance of model.

Author Contributions: H.-Y.C. and C.-H.L. initiated and developed the ideas related to this research.
Both of them developed the presented novel methods, derived relevant formulations, and carried
out the performance analyses of simulation and experimental results. H.-Y.C. wrote the paper draft
under C.-H.L.’s guidance and Professor Lee finalized the paper. Both authors have read and agreed
to the published version of the manuscript.
Funding: This work was supported in part by the Ministry of Science and Technology, Taiwan, under
contracts MOST-110-2634-F-009-024, 109-2218-E-005-015, and 109-2218-E-150-002.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The used data of bearing fault can be found in Case Western Reserve
University Bearing Data Center. Available online: http://csegroups.case.edu/bearingdatacenter/pages/
wel-come-case-western-reserve-university-bearing-data-center-website (accessed on 10 March 2019).
Conflicts of Interest: The authors declare no conflict of interests.

References
1. Sharma, G.; Umapathy, K.; Krishnan, S. Trends in audio signal feature extraction methods. Appl. Acoust. 2020, 158, 107020.
[CrossRef]
2. García Plaza, E.; Núñez López, P.J.; Beamud González, E.M. Efficiency of vibration signal feature extraction for surface finish
monitoring in CNC machining. J. Manuf. Process. 2019, 44, 145–157. [CrossRef]
3. He, G.; Ding, K.; Lin, H. Fault feature extraction of rolling element bearings using sparse representation. J. Sound Vib. 2016, 366,
514–527. [CrossRef]
4. Xiao, R.; Hu, Q.; Li, J. Leak detection of gas pipelines using acoustic signals based on wavelet transform and Support Vector
Machine. Measurement 2019, 146, 479–489. [CrossRef]
5. Ren, Z.; Zhou, S.; Chunhui, E.; Gong, M.; Li, B.; Wen, B. Crack fault diagnosis of rotor systems using wavelet transforms. Comput.
Electr. Eng. 2015, 45, 33–41. [CrossRef]
6. Lambrou, T.; Kudumakis, P.; Speller, R.; Sandler, M.; Linney, A. Classification of audio signals using statistical features on time
and wavelet transform domains. Acoust. Speech Signal Process. 1988, 6, 3621–3624.
7. Wu, Z.; Huang, N.E. Ensemble Empirical Mode Decomposition: A Noise-Assisted Data Analysis Method. Adv. Adapt. Data Anal.
2009, 1, 1–41. [CrossRef]
8. Oppenheim, A.V. Applications of Digital Signal Processing; Prentice Hall: Englewood Cliffs, NJ, USA, 1978.
Sensors 2021, 21, 3929 16 of 17

9. Dolenc, B.; Boškoski, P.; Juričić, Ð. Distributed bearing fault diagnosis based on vibration analysis. Mech. Syst. Signal Process.
2016, 66–67, 521–532. [CrossRef]
10. Al-Salman, W.; Li, Y.; Wen, P. K-complexes Detection in EEG Signals using Fractal and Frequency Features Coupled with an
Ensemble Classification Model. Neuroscience 2019, 422, 119–133. [CrossRef] [PubMed]
11. Rukhsar, S.; Khan, Y.; Farooq, O.; Sarfraz, M.; Khan, A. Patient-Specific Epileptic Seizure Prediction in Long-Term Scalp EEG
Signal Using Multivariate Statistical Process Control. IRBM 2019, 40, 320–331. [CrossRef]
12. Yang, Y.; Yang, W.; Jiang, D. Simulation and experimental analysis of rolling element bearing fault in rotor-bearing-casing system.
Eng. Fail. Anal. 2018, 92, 205–221. [CrossRef]
13. Wang, T.; Liang, M.; Li, J.; Cheng, W. Rolling element bearing fault diagnosis via fault characteristic order (FCO) analysis. Mech.
Syst. Signal Process. 2014, 45, 139–153. [CrossRef]
14. Zhao, D.; Wang, T.; Gao, R.X.; Chu, F. Signal optimization based generalized demodulation transform for rolling bearing
nonstationary fault characteristic extraction. Mech. Syst. Signal Process. 2019, 134, 106297. [CrossRef]
15. Liu, Y.; Guo, L.; Wang, Q.; An, G.; Guo, M.; Lian, H. Application to induction motor faults diagnosis of the amplitude recovery
method combined with FFT. Mech. Syst. Signal Process. 2010, 24, 2961–2971. [CrossRef]
16. Lee, W.; Ratnam, M.; Ahmad, Z. Detection of chipping in ceramic cutting inserts from workpiece profile during turning using fast
Fourier transform (FFT) and continuous wavelet transform (CWT). Precis. Eng. 2017, 47, 406–423. [CrossRef]
17. Yan, X.; Jia, M. A novel optimized SVM classification algorithm with multi-domain feature and its application to fault diagnosis
of rolling bearing. Neurocomputing 2018, 313, 47–64. [CrossRef]
18. Abdelkrim, C.; Meridjet, M.S.; Boutasseta, N.; Boulanouar, L. Detection and classification of bearing faults in industrial geared
motors using temporal features and adaptive neuro-fuzzy inference system. Heliyon 2019, 5, e02046. [CrossRef] [PubMed]
19. Liu, J.; Xu, Z.; Zhou, L.; Yu, W.; Shao, Y. A statistical feature investigation of the spalling propagation assessment for a ball bearing.
Mech. Mach. Theory 2019, 131, 336–350. [CrossRef]
20. Wu, C.; Jiang, P.; Ding, C.; Feng, F.; Chen, T. Intelligent fault diagnosis of rotating machinery based on one-dimensional
convolutional neural network. Comput. Ind. 2019, 108, 53–61. [CrossRef]
21. Zhang, J.; Sun, Y.; Guo, L.; Gao, H.; Hong, X.; Song, H. A new bearing fault diagnosis method based on modified convolutional
neural networks. Chin. J. Aeronaut. 2020, 33, 439–447. [CrossRef]
22. Li, C.; Zhao, D.; Mu, S.; Zhang, W.; Shi, N.; Li, L. Fault diagnosis for distillation process based on CNN–DAE. Chin. J. Chem. Eng.
2019, 27, 598–604. [CrossRef]
23. Wang, S.; Xiang, J.; Zhong, Y.; Zhou, Y. Convolutional neural network-based hidden Markov models for rolling element bearing
fault identification. Knowl.-Based Syst. 2018, 144, 65–76. [CrossRef]
24. Lu, C.; Wang, Z.; Zhou, B. Intelligent fault diagnosis of rolling bearing using hierarchical convolutional network based health
state classification. Adv. Eng. Inform. 2017, 32, 139–151. [CrossRef]
25. Dong, S.; Wu, W.; He, K.; Mou, X. Rolling bearing performance degradation assessment based on improved convolutional neural
network with anti-interference. Measurement 2020, 151, 107219. [CrossRef]
26. Islam, M.M.M.; Kim, J.-M. Motor Bearing Fault Diagnosis Using Deep Convolutional Neural Networks with 2D Analysis of
Vibration Signal. Trans. Petri Nets Other Models Concurr. XV 2018, 10832, 144–155. [CrossRef]
27. An, Q.; Tao, Z.; Xu, X.; El Mansori, M.; Chen, M. A data-driven model for milling tool remaining useful life prediction with
convolutional and stacked LSTM network. Measurement 2020, 154, 107461. [CrossRef]
28. Kious, M.; Ouahabi, A.; Boudraa, M.; Serra, R.; Cheknane, A. Detection process approach of tool wear in high speed milling.
Measurement 2010, 43, 1439–1446. [CrossRef]
29. Pandiyan, V.; Tjahjowidodo, T. Use of Acoustic Emissions to detect change in contact mechanisms caused by tool wear in abrasive
belt grinding process. Wear 2019, 436–437, 203047. [CrossRef]
30. Zhang, C.; Zhang, J. On-line tool wear measurement for ball-end milling cutter based on machine vision. Comput. Ind. 2013, 64,
708–719. [CrossRef]
31. García-Ordás, M.T.; Alegre-Gutiérrez, E.; Alaiz-Rodríguez, R.; González-Castro, V. Tool wear monitoring using an online,
automatic and low cost system based on local texture. Mech. Syst. Signal Process. 2018, 112, 98–112. [CrossRef]
32. Srinivasan, R.; Jacob, V.; Muniappan, A.; Madhu, S.; Sreenevasulu, M. Modeling of surface roughness in abrasive water jet
machining of AZ91 magnesium alloy using Fuzzy logic and Regression analysis. Mater. Today Proc. 2020, 22, 1059–1064. [CrossRef]
33. Parida, A.K.; Maity, K. Modeling of machining parameters affecting flank wear and surface roughness in hot turning of Monel-400
using response surface methodology (RSM). Measurement 2019, 137, 375–381. [CrossRef]
34. Wu, T.Y.; Lei, K.W. Prediction of surface roughness in milling process using vibration signal analysis and artificial neural network.
Int. J. Adv. Manuf. Technol. 2019, 102, 305–314. [CrossRef]
35. Rao, K.V.; Murthy, B.; Rao, N.M. Prediction of cutting tool wear, surface roughness and vibration of work piece in boring of AISI
316 steel with artificial neural network. Measurement 2014, 51, 63–70. [CrossRef]
36. Yunusa-Kaltungo, A.; Cao, R. Towards Developing an Automated Faults Characterization Framework for Rotating Machines.
Part 1: Rotor-Related Faults. Energies 2020, 13, 1394. [CrossRef]
37. Cao, R.; Yunusa-Kaltungo, A. An Automated Data Fusion-Based Gear Faults Classification Framework in Rotating Machines.
Sensor 2021, 21, 2957. [CrossRef] [PubMed]
Sensors 2021, 21, 3929 17 of 17

38. Banerjee, T.P.; Das, S. Multi-sensor data fusion using support vector machine for motor fault detection. Inf. Sci. 2012, 217, 96–107.
[CrossRef]
39. Gunerkar, R.; Jalan, A. Classification of Ball Bearing Faults Using Vibro-Acoustic Sensor Data Fusion. Exp. Tech. 2019, 43, 635–643.
[CrossRef]
40. Wang, X.; Mao, D.; Li, X. Bearing fault diagnosis based on vibro-acoustic data fusion and 1D-CNN network. Measurement 2021,
173, 108518. [CrossRef]
41. Safizadeh, M.; Latifi, S. Using multi-sensor data fusion for vibration fault diagnosis of rolling element bearings by accelerometer
and load cell. Inf. Fusion 2014, 18, 1–8. [CrossRef]
42. Luwei, K.C.; Yunusa-Kaltungo, A.; Sha’Aban, Y.A. Integrated Fault Detection Framework for Classifying Rotating Machine Faults
Using Frequency Domain Data Fusion and Artificial Neural Networks. Machines 2018, 6, 59. [CrossRef]
43. Huang, M.; Liu, Z.; Tao, Y. Mechanical fault diagnosis and prediction in IoT based on multi-source sensing data fusion. Simul.
Model. Pr. Theory 2020, 102, 101981. [CrossRef]
44. Cabrera, D.; Sancho, F.; Li, C.; Cerrada, M.; Sánchez, R.-V.; Pacheco, F.; de Oliveira, J.V. Automatic feature extraction of time-series
applied to fault severity assessment of helical gearbox in stationary and non-stationary speed operation. Appl. Soft Comput. 2017,
58, 53–64. [CrossRef]
45. Cintas, C.; Lucena, M.; Fuertes, J.M.; Delrieux, C.; Navarro, P.; González-José, R.; Molinos, M. Automatic feature extraction and
classification of Iberian ceramics based on deep convolutional networks. J. Cult. Herit. 2019, 41, 106–112. [CrossRef]
46. Hung, C.-W.; Zeng, S.-X.; Lee, C.-H.; Li, W.-T. End-to-End Deep Learning by MCU Implementation: An Intelligent Gripper for
Shape Identification. Sensors 2021, 21, 891. [CrossRef]
47. Chan, H.Y.; Lee, C.H. Vibration signals analysis by explainable artificial intelligence (xai) approach: Application on bearing faults
diagnosis. IEEE Access 2020, 8, 134246–134256. [CrossRef]
48. Lo, C.-C.; Lee, C.-H.; Huang, W.-C. Prognosis of Bearing and Gear Wears Using Convolutional Neural Network with Hybrid Loss
Function. Sensors 2020, 20, 3539. [CrossRef]
49. Yildirim, O.; Talo, M.; Ay, B.; Baloglu, U.B.; Aydin, G.; Acharya, U.R. Automated detection of diabetic subject using pre-trained
2D-CNN models with frequency spectrum images extracted from heart rate signals. Comput. Biol. Med. 2019, 113, 103387.
[CrossRef] [PubMed]
50. Cao, X.-C.; Chen, B.-Q.; Yao, B.; He, W.-P. Combining translation-invariant wavelet frames and convolutional neural network for
intelligent tool wear state identification. Comput. Ind. 2019, 106, 71–84. [CrossRef]
51. LeCun, Y.; Bottou, L.; Bengio, Y.; Haffner, P. Gradient-based learning applied to document recognition. Proc. IEEE 1998, 86,
2278–2324. [CrossRef]
52. Sejdić, E.; Djurović, I.; Jiang, J. Time–frequency feature representation using energy concentration: An overview of recent
advances. Digit. Signal Process. 2009, 19, 153–183. [CrossRef]
53. Kenndy, J.; Eberhart, R. Particle Swarm Optimization. In Proceedings of the ICNN’95-International Conference on Neural
Networks, Perth, WA, Australia, 27 November–1 December 1995; Volume 4, pp. 1942–1948.
54. Chou, F.-I.; Tsai, Y.-K.; Chen, Y.-M.; Tsai, J.-T.; Kuo, C.-C. Optimizing Parameters of Multi-Layer Convolutional Neural Network
by Modeling and Optimization Method. IEEE Access 2019, 7, 68316–68330. [CrossRef]
55. Fang, K.-T.; Liu, M.-Q.; Qin, H.; Zhou, Y.-D. Theory and Applications of Uniform Experimental Designs; Springer: Singapore, 2018.
56. Bearing Data Center Seeded Fault Test Data. Available online: https://csegroups.case.edu/bearingdatacenter/pages/apparatus-
procedures (accessed on 10 March 2019).
57. Li, B.; Zhang, P.-L.; Liu, D.-S.; Mi, S.-S.; Ren, G.-Q.; Tian, H. Feature extraction for rolling element bearing fault diagnosis utilizing
generalized S transform and two-dimensional non-negative matrix factorization. J. Sound Vib. 2011, 330, 2388–2399. [CrossRef]
58. Li, X.; Ma, J.; Wang, X.; Wu, J.; Li, Z. An improved local mean decomposition method based on improved composite interpolation
envelope and its application in bearing fault feature extraction. ISA Trans. 2020, 97, 365–383. [CrossRef] [PubMed]
59. Smith, W.; Randall, R.B. Rolling element bearing diagnostics using the Case Western Reserve University data: A benchmark
study. Mech. Syst. Signal Process. 2015, 64–65, 100–131. [CrossRef]
Reproduced with permission of copyright owner. Further reproduction
prohibited without permission.