The International Journal of Advanced Manufacturing Technology (2025) 136:1577–1585

https://doi.org/10.1007/s00170-024-14949-y

ORIGINAL ARTICLE

Innovative defect detection in surface‑modified gears using

time–frequency domain analysis and deep learning models
Shu‑Hsien Huang1 · Ting‑En Wu2 · Chia‑Hung Lai3

Received: 12 August 2024 / Accepted: 16 December 2024 / Published online: 27 December 2024
© The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2024

Abstract
Gears play a pivotal role in power transmission systems. Over prolonged periods of high-load transmission, the issue of gear
lifespan has become a challenging one to address. Monitoring systems are vital in today’s smart manufacturing landscape,
contributing to safety and sustainable development. Presently, there is a dearth of research focused on the vibration of gears
with surface modification. Consequently, this study has devised a novel defect detection model for surface modification
gears. It employs time-domain, frequency-domain, and discrete wavelet transform (DWT) analysis in combination with
deep neural network (DNN) deep learning to validate the effectiveness of this approach. The study deals with gears in two
states: undamaged and wear. Vibration data was sampled for 800 s at a rate of 1024 Hz. Leveraging artificial intelligence, this
research opens the door to innovative applications and services in the field of gear manufacturing. Notably, DWT exhibits
a significant reduction in the amplitude of vibrations between surface modification helical gears and surface modification
spur gears. Furthermore, from the DNN training and accuracy, it is evident that there is no substantial difference in accuracy
between epochs 30 and epochs 100.

Keywords Gear monitoring · Vibration analyze · Gear modification · Deep learning

1 Introduction Monitoring systems play a crucial role in today’s indus-

trial landscape, contributing to safety and sustainable
Gears play a crucial role in transmission systems, especially development [1]. By using digital twins to monitor gear
during prolonged high-load operations. The issue of gear transmissions, it was discovered that the proposed monitor-
lifespan becomes a challenging problem to address. Wear ing system can simulate real-world gear performance and
and tear of gears can lead to transmission errors, increased provide high-precision monitoring [2]. The use of vibra-
vibrations, and elevated noise levels. Monitoring gear per- tion indicators for assessing gear wear revealed the ability
formance through a gear monitoring system allows for the to quickly detect gear damage [3]. Analyzing motor signal
recording of the gear transmission process, ensuring the sta- modulation signal bispectrum for gear wear monitoring, it
bility of gears during the transmission. was found to be accurate in detecting gear wear [4]. The
application of transfer learning with convolutional neural
Shu-Hsien Huang and Ting-En Wu contributed equally to this work. networks in gear transmission monitoring eliminated the
need for extensive data preprocessing and achieved high
* Chia‑Hung Lai accuracy with minimal data [5]. An intelligent prediction
[email protected] system was introduced, capable of identifying gear wear
1
Department of Information Management, National Chin-Yi through gearbox vibrations and predicting the remaining
University of Technology, No. 57, Sec. 2, Zhongshan Rd., lifespan of gears, showing a high level of accuracy [6].
Taiping Dist., Taichung 411030, Taiwan (R.O.C.) Analyzing twist vibration signals for planetary gear wear
2
Department of Industrial Education and Technology, provided a more accessible and effective method for data
National Changhua University of Education, No. 2, Shi‑Da extraction and prediction [7]. Vibration-based monitor-
Road, Changhua City 50074, Taiwan (R.O.C.) ing techniques for gear wear, including changes in surface
3
Department of Intelligent Automation Engineering, National contour and pitting density, proved effective in identifying
Chin-Yi University of Technology, No. 57, Sec. 2, Zhongshan gear damage [8]. Vibration-based testing for gear pair wear
Rd., Taiping Dist., Taichung 411030, Taiwan (R.O.C.)

Vol.:(0123456789)
1578 The International Journal of Advanced Manufacturing Technology (2025) 136:1577–1585

demonstrated high accuracy in wear prediction under vari- networks for classification and prediction. This process
ous lubrication conditions [9]. An intelligent manufacturing constructs a model for detecting gear surface defects, ena-
system for gear health management, utilizing gear vibration bling efficient detection and prediction of wear and defects
detection, displayed a high level of predictive accuracy [10]. in surface-finished gears.
Finally, in cases of lubricant contamination and transmis-
sion errors, the use of vibration monitoring for gear wear 2.1 Gear tooth surface modification
detection has been found to be effective [11]. Research on
the Profile Modification and Meshing Analysis of Hyper- Gears offer the advantage of high transmission efficiency
bolic Circular Arc Teeth aims to effectively reduce vibra- and are commonly used components in the industrial sector.
tion and noise while enhancing load capacity [12]. Using However, gear transmissions often encounter issues such as
the fish-swarm algorithm and support vector machine for noise and vibration. Therefore, it is necessary to enhance the
bearing detection, it was found that this method achieved gear lifespan through gear tooth surface conditioning. This
high accuracy [13]. Predicting turning precision through research adopts a gear tooth surface conditioning method
XGBoost, it was discovered that this method can optimize [14]. Through this method, it was observed that an effec-
turning parameters. tive reduction in vibration during gear transmission and an
From the literature review, it has been found that no improvement in stress distribution could be achieved [15].
research has focused on defect detection for gears with Utilizing gear tooth surface conditioning for bevel gears, it
surface-finished teeth. This research introduces a novel was found to effectively reduce surface stresses and signifi-
method for detecting gear wear. Signals collected are pro- cantly decrease transmission errors [16]. Gear tooth surface
cessed through time-domain, frequency-domain, discrete conditioning was also found to prevent edge contact situa-
wavelet transform, and spectral analysis techniques. These tions by introducing longitudinal module variation, resulting
processed signals are used to train a deep neural network in a convex gear surface that transforms line contact into
(DNN) model, enabling effective monitoring of gear wear. point contact, maintaining contact at the center of the gear
The significance of this study lies in proposing a gear defect surface (Fig. 1).
detection model that combines time-domain, frequency-
domain, and discrete wavelet transform (DWT) with deep 2.2 Fast fourier transform (FFT)
neural networks (DNN), effectively enhancing the accuracy
and efficiency of gear lifespan prediction and fault detection. Fourier transform is frequently used for signal process-
This study only studies the defects of longitudinal module ing. It can decompose signals into sine and cosine compo-
variation tooth surface modification and evaluates its feasi- nents, revealing periodic features in the signal, as shown
bility. If it is applied to other tooth types, further research in Fig. 2. In the time domain, the signal evolves over time,
is needed. The primary contributions of this research are as while in the frequency domain, as depicted in Fig. 3, it
follows: represents amplitudes at different frequencies. This study
uses a time-accelerometer (t-a) curve. The experiment is to
I. Propose a vibration-based gear condition monitoring detect defects due to natural wear during gear transmission.
system to broaden the practical applications of gear The peak at 260 Hz is irregular wear or defects on the gear
inspection systems. surface. When gears have uneven wear or damage, these
II. Introduce a detection system tailored for inspecting defects can produce significant vibration signals at specific
gear tooth surface defects. frequencies [17]. By employing the fast Fourier transform
III. Validate the effectiveness of this approach based on method to analyze the time and frequency domains, it has
a deep neural network model. been observed that this approach is effective in detecting
IV. Achieve a gear tooth surface defect recognition rate gear defects [18]. Through wavelet spectrogram analysis of
of 97.8%. gear defects, it was found that this method achieves 100%
accuracy. The fast Fourier transform is more efficient com-
pared to the discrete Fourier transform. The complexity of
2 Gear tooth surface vibration detection discrete Fourier transform is O (n2 , whereas fast Fourier
( )

system transform has a complexity of O nlog2 n , where “n” repre-

)

sents the signal length, y(n) is the nth sample, and N is the
The gear vibration detection system utilizes fast Fourier length of the signal. The fast Fourier transform formula is
transform (FFT), discrete wavelet transform (DWT), and as follows:
deep neural network (DNN) methods. These methods extract −j2π×n
(1)
∑N−1
features by converting vibration signals into the frequency Y(k) = n=0
y(n)e N , (0 ≤ k ≤ N − 1)
and time–frequency domains and then use deep neural
The International Journal of Advanced Manufacturing Technology (2025) 136:1577–1585 1579

Fig. 1  a Modification tooth

surface. b Unmodified tooth
surface

Fig. 2  Time-domain signal

Fig. 3  Frequency-domain signal

1580 The International Journal of Advanced Manufacturing Technology (2025) 136:1577–1585

2.3 Discrete wavelet transform (DWT) defects, demonstrating its superiority over other methods
[23]. Using the LSTM-DNN method for gear prognosis, it
Discrete wavelet transform (DWT) is commonly used to was observed that this approach reduces the impact of data
analyze the time and frequency characteristics of signals, while increasing recognition accuracy [24]. By combining
enabling localized analysis in both the time and frequency SAE, softmax, and DNN for gear defect recognition, it was
domains [19]. Through the application of discrete wavelet found that this method becomes more robust as the number
transform and adaptive neuro-fuzzy inference, gear defect of DNN layers increases. [25] Monitoring of spindle cutting
identification was carried out, revealing that this method vibration using LSTM and DNN revealed that this method
achieves a high level of accuracy compared to traditional achieves very high precision. DNN [26] has very good accu-
visual inspection methods [20]. Using discrete wavelet racy in deep learning training.
transform and autoregressive modeling for gear vibration
analysis, it was found that this approach can serve as an
indicator of gear health [21]. The DWT(m, n) represents the 3 Experimental setup
wavelet coefficients at scale and position (m, n), where x(t)
is the time-domain signal, and 𝜓( t−n ) represents the wavelet This research designs a novel defect detection model for
2m
function. The formula for discrete wavelet transform is as gear tooth surface conditioning. The gear and motor speci-
follows: fications used in this research are presented in Table 1. The
number of teeth on the large gear is 30, and the number
1 ∑ t−n
� �
DWT(m, n) = √ k × (t)ψ (2) of teeth on the small gear is 20. The modulus is 3. The
2m 2m addendum is 1 m, and the dedendum is 1 m. The pressure
angle is 2π
9
. The unmodified half tooth thickness is 0.25𝜋m ,
and the modified half tooth thickness is 0.25𝜋m × Cos𝛽 (𝛽
2.4 Deep neural network (DNN) range is − 6π ∼ π6 ). The motor speed is 625 rpm. An accel-
erometer is placed on the bearing connecting the motor
A deep neural network (DNN) is a model composed of an
and the gear to record vibrations during gear transmission.
Input Layer, Hidden Layer, and an Output Layer (Fig. 4).
The recorded values are subjected to Fourier transforma-
The Input Layer takes the input data as labels. The Hidden
tion and subsequently analyzed through time-domain,
Layer performs calculations on the data received from the
frequency-domain, discrete wavelet transformation, and
first layer, and it forms the model based on feature classifi-
spectral analysis methods. Deep learning using a DNN is
cation and weight loss. The Output Layer outputs the data
then employed. The model consists of three hidden layers
from the Hidden Layer [22]. The experiment employs 8 hid-
containing 128, 64, and 32 neurons, respectively, and each
den layers for deep learning training. By employing DNN
layer uses a ReLU activation function to capture nonlin-
to analyze gearbox current, it can identify planetary gear
ear features. The output layer uses a single neuron with a
sigmoid activation function for binary classification tasks.
And do it using Adam optimizer. The model uses ReLU,
which is a function that is effective in solving the vanishing
gradient problem and has high computational efficiency.
Data standardization processing, scaling the data to a range

Table 1  Parameters of the gear tooth surface conditioning gear

inspection system
Number of gear teeth,z2 30

Number of pinion teeth,z1 20

Modulus,m 3
Addendum,at 1m
Dedendum,ac 1m
Pressure angle,𝛼1 2𝜋
9
Unmodified half the tooth thickness,bc 0.25𝜋m
Modified half the tooth thickness,bc 0.25𝜋m × Cos𝛽
(𝛽 range is −𝜋∕6 ∼ 𝜋∕6)
Motor speed, rpm 625
Fig. 4  Deep neural network architecture diagram
The International Journal of Advanced Manufacturing Technology (2025) 136:1577–1585 1581

with a mean of 0 and a standard deviation of 1, helps accel- Table 2  Training and testing data
erate model convergence and avoid the adverse effects of Count
different feature scales on training. The Adam optimizer
combines the advantages of momentum gradient descent Tooth Surface Modification Spur Gear Training Database 640
(SGD) and RMSProp. It has the characteristics of adaptive Tooth Surface Modification Spur Gear Test Database 160
learning rate and fast convergence, making sparse gradi- Wear Tooth Surface Modification Spur Gear Training Data- 640
base
ent and non-stationary better for target optimization. This
Wear Tooth Surface Modification Spur Gear Test Database 160
model is utilized for the detection of gear defects, enabling
Tooth Surface Modification Helical Gear Training Database 640
the rapid identification of gear wear when it occurs. This
Tooth Surface Modification Helical Gear Test Database 160
study focuses on defect detection related to uneven trans-
Wear Tooth Surface Modification Helical Gear Training 640
mission wear and gear surface abrasion generated during
Database
gear transmission.
Wear Tooth Surface Modification Helical Gear Test Database 160
In this research, two gear conditions were considered:
undamaged gears and wear gears. The schematic diagram of
the monitoring system is shown in Fig. 5. The accelerometer
will be placed above the bearing in this experiment to collect 4 Results
vibration data during gear transmission. We used the z-axis
of the accelerometer for measurement, and its main moni- This research conducts inspections on spur gears and heli-
toring direction is perpendicular to the transmission surface cal gears after gear tooth surface conditioning. The detected
of the gear. We used the time-accelerometer (t-a) curve. If signals are transformed into time-domain, frequency-
the gear is defective or unevenly worn, the defect can cause domain, and DWT signals. Subsequently, these signals are
vibrations at other frequencies such that the expected start- processed through deep learning using a DNN and undergo
ing speed frequency of 10.41 Hz is not reached. Vibration spectral analysis to establish an effective defect monitoring
data was sampled for 800 s at a rate of 1024 Hz. Surface- system. The DNN accuracy is illustrated in Fig. 6.
conditioned spur gears and helical gears were tested sepa- From Fig. 6, it can be observed that the data collected
rately for their undamaged and wear conditions. For each from profile-modified spur gears and profile-modified
gear condition, 800 sets of vibration data were collected. helical gears in the time domain, frequency domain, and
In this experiment, the collected data set was divided into a discrete wavelet transform (DWT) were subjected to
test set and a training set. The classification method is to ran- deep learning training for 10, 30, 50, 80, and 100 epochs,
domly classify the 1000 pieces of data collected in a ratio of respectively. High accuracy was achieved in the frequency
8:2. Since this study is mainly aimed at detecting the defects domain, and the results are the same as [27]. Accuracy in
of longitudinal module variation tooth surface modification. the time domain and DWT was not satisfactory. This indi-
Therefore, I chose fixed data partitioning for evaluation. The cates that there is no significant difference in the gears in
training and testing data are detailed in Table 2. the time domain and DWT. Other methods are needed for

Fig. 5  Gear inspection system

schematic diagram
1582 The International Journal of Advanced Manufacturing Technology (2025) 136:1577–1585

Fig. 6  a Tooth surface modification spur gear accuracy and epochs. b Tooth surface modification helical gear accuracy and epochs

detection. Spectrum analysis of the profile-modified spur exhibit wear, the amplitude increases compared to non wear
gears is shown in Fig. 7. gears. The most obvious amplitude appears at times 1 to 1.5 s.
From Fig. 7, it can be observed that the amplitude of This increase in amplitude suggests that there is an accompa-
surface-conditioned spur gears falls within the range of − 20 nying increase in gear vibrations during transmission. When
to − 60. In contrast, the amplitude of wear surface-condi- the gear amplitude transitions from − 60 to − 70, it indicates
tioned spur gears falls within the range of − 20 to − 60. When the presence of gear wear, suggesting that replacement is rec-
gears exhibit wear, the amplitude increases compared to ommended. The spectral analysis of surface-conditioned heli-
non wear gears. This increase in amplitude suggests that cal gears is presented in Fig. 8. From Figs. 7 and 8, it can be
there is an accompanying increase in gear vibrations during concluded that when the gears exhibit wear, the amplitude is
transmission. When the gear amplitude transitions are lower significantly higher than that of undamaged gears. This indi-
than − 60, it indicates the presence of gear wear, suggesting cates that gear wear leads to an increase in vibration ampli-
that replacement is recommended. The spectral analysis of tude, thereby affecting gear operation. The coefficient plot for
surface-conditioned helical gears is presented in Fig. 8. surface-conditioned spur gears is shown in Fig. 9.
From Fig. 8, it can be observed that the amplitude of the From Fig. 9, it can be observed that the amplitude of
surface-conditioned spur gears falls within the range of − 20 undamaged surface-conditioned spur gears falls within
to − 60. In contrast, the amplitude of wear surface-conditioned the range of − 0.04 to − 0.03. In contrast, the amplitude of
spur gears falls within the range of − 20 to − 80. When gears wear surface-conditioned spur gears falls within the range

Fig. 7  Modification spur gear spectral analysis. a Undamaged tooth surface. b Wear tooth surface
The International Journal of Advanced Manufacturing Technology (2025) 136:1577–1585 1583

Fig. 8  Modification helical gear spectral analysis. a Undamaged tooth surface. b Wear tooth surface

Fig. 9  Modification spur gear coefficient chart. a Undamaged tooth surface. b Wear tooth surface

of − 0.03 to − 0.04. When the gears are in operation and the helical gears is significantly smaller than that of undam-
amplitude falls within the range of − 0.03 to − 0.04, it indi- aged surface-conditioned spur gears. From Figs. 9 and 10,
cates the presence of gear wear, and replacement is recom- it can be concluded that the amplitude of undamaged sur-
mended. The coefficient plot for surface-conditioned helical face-finished helical gears is significantly lower than that of
gears is shown in Fig. 10. undamaged surface-finished spur gears. This indicates that
From Fig. 10, it can be observed that the amplitude of helical gears have better operational stability than spur gears
undamaged surface-conditioned spur gears falls within in an undamaged state.
the range of − 0.01 to 0.015. In contrast, the amplitude of
wear surface-conditioned spur gears falls within the range
of − 0.04 to 0.05. When the gears are in operation, there is 5 Conclusion
a noticeable increase in amplitude. If the amplitude falls
within the range of − 0.03 to 0.04, it indicates the presence This research introduces a monitoring system for surface-
of gear wear, and replacement is recommended. Compara- conditioned gears, focusing on gear tooth surface condi-
tively, the amplitude of undamaged surface-conditioned tioning. Time-domain, frequency-domain, and DWT data
1584 The International Journal of Advanced Manufacturing Technology (2025) 136:1577–1585

Fig. 10  Modification helical gear coefficient chart. a Undamaged tooth surface. b Wear tooth surface

are subjected to DNN deep learning. Various methods are pared to undamaged helical gears, when the coeffi-
analyzed for their accuracy in gear monitoring to ensure cient falls within the range of − 0.03 to 0.04. This
the system’s reliability. Based on the comprehensive anal- indicates gear wear, recommending replacement.
ysis results, the following conclusions can be drawn: V. In the future, the focus will continue to be on vibra-
tion analysis for monitoring gear health. This can be
I. In this research, the data collected from the time extended to include bevel gears and gears of various
domain, frequency domain, and DWT are processed modules to expand the number of monitoring system
through DNN deep learning. DNN is effective in cases.
detecting gear wear.
II. From the training and accuracy of DNN, it is
observed that the accuracy in the time domain and
DWT is not satisfactory, and there is no improvement
Author contribution Not applicable.
with the increase of epochs. In the frequency domain,
very good accuracy can be obtained, but there is also Data availability The data and material that support the findings of
no improvement with the increase of epochs. this study are available from the corresponding author, Chia-Hung Lai,
III. From the spectrum analysis of spur gears and helical upon reasonable request.
gears, it was observed that worn spur gears exhib- Code availability The code that supports the findings of this study is
ited larger amplitudes compared to undamaged spur available from the corresponding author, Chia-Hung Lai, upon reason-
gears. When the amplitude drops below − 60, it indi- able request.
cates gear wear, suggesting replacement. Similarly,
worn helical gears exhibited larger amplitudes com- Declarations
pared to undamaged helical gears, with the maximum Ethics approval This research did not contain any studies involving
amplitude occurring between 1 and 1.5 s. When the animal or human participants, nor did it take place on any private or
amplitude changes from − 60 to − 70, it indicates gear protected areas.
wear, recommending replacement.
Consent to participate Additional informed consent was obtained from
IV. From the DWT analysis of spur gears and helical all individual participants for whom identifying information is included
gears, it was found that when the coefficient of the in this article.
spur gear falls within the range of − 0.03 to 0.04
during transmission, it indicates gear wear, suggest- Consent for publication Additional informed consent was obtained
from all individual participants for whom identifying information is
ing replacement. In comparison, worn helical gears included in this article.
exhibited a significant increase in amplitude com-
The International Journal of Advanced Manufacturing Technology (2025) 136:1577–1585 1585

Competing interests This research did not contain any studies involv- 15. Samani S F, Molaie M, Pellicano F (2019) Nonlinear vibration
ing animal or human participants, nor did it take place on any private of the spiral bevel gear with a novel tooth surface modification
or protected areas. method. Meccanica 54(7):1071–1081
16. Eritenel T, Parker RG (2013) Nonlinear vibration of gears with
tooth surface modifications. J Vib Acoust 135(5):051005
17. Mohammed OD, Rantatalo M (2016) Dynamic response and time-
References frequency analysis for gear tooth crack detection. Mech Syst Sig-
nal Process 66:612–624
18. Dai P, Wang J, Wu L, Yan S, Wang F, Niu L (2022). Defect diag-
1. Li J, Wang S, Yang J, Zhang H, Zhao H (2023) A digital twin-based
nosis of gear-shaft bearing system based on the OWF-TSCNN
state monitoring method of gear test bench. Appl Sci 13(5):3291
composed of wavelet time-frequency map and FFT Spectrum 1.
2. Hu C, Smith WA, Randall RB, Peng Z (2016) Development of a
Shock and Vibration, 2022.
gear vibration indicator and its application in gear wear monitor-
19. Wu JD, Hsu CC, Wu GZ (2009) Fault gear identification and
ing. Mech Syst Signal Process 76:319–336
classification using discrete wavelet transform and adaptive neuro-
3. Zhang R, Gu F, Mansaf H, Wang T, Ball AD (2017) Gear wear
fuzzy inference. Expert Syst Appl 36(3):6244–6255
monitoring by modulation signal bispectrum based on motor cur-
20. Sharma S, Tiwari SK, Singh S (2019) Diagnosis of gear tooth fault
rent signal analysis. Mech Syst Signal Process 94:202–213
in a bevel gearbox using discrete wavelet transform and autore-
4. Cao P, Zhang S, Tang J (2018) Preprocessing-free gear fault
gressive modeling. Life Cycle Reliability and Safety Engineering
diagnosis using small datasets with deep convolutional neural
8:21–32
network-based transfer learning. Ieee Access 6:26241–26253
21. Suresh S, Naidu VPS (2020) Gearbox health condition monitoring
5. Wang D, Miao Q, Zhou Q, Zhou G (2015) An intelligent prog-
using DWT features. In Proceedings of the 6th National Sympo-
nostic system for gear performance degradation assessment and
sium on Rotor Dynamics: NSRD 2019 (pp. 361–374). Singapore:
remaining useful life estimation. J Vib Acoust 137(2):021004
Springer Singapore
6. Xue S, Howard I (2018) Torsional vibration signal analysis as
22. Li F, Pang X, Yang Z (2019) Motor current signal analysis using
a diagnostic tool for planetary gear fault detection. Mech Syst
deep neural networks for planetary gear fault diagnosis. Measure-
Signal Process 100:706–728
ment 145:45–54
7. Feng K, Smith WA, Randall RB, Wu H, Peng Z (2022) Vibration-
23. Liu Z, Tan C, Liu Y, Li H, Cui B, Zhang X (2023) A study of a
based monitoring and prediction of surface profile change and
domain-adaptive LSTM-DNN-based method for remaining useful
pitting density in a spur gear wear process. Mech Syst Signal
life prediction of planetary gearbox. Processes 11(7):2002
Process 165:108319
24. Wang J, Jiang X, Li S, Xin Y (2017) A novel feature representa-
8. Feng K, Smith WA, Peng Z (2021) Use of an improved vibration-
tion method based on deep neural networks for gear fault diagno-
based updating methodology for gear wear prediction. Eng Fail
sis. In 2017 Prognostics and system health management confer-
Anal 120:105066
ence (PHM-Harbin) (pp. 1–6). IEEE
9. Feng K, Ji JC, Ni Q, Li Y, Mao W, Liu L (2023) A novel vibration-
25. Li B, Liu T, Liao J, Feng C, Yao L, Zhang J (2023) Non-invasive
based prognostic scheme for gear health management in surface
milling force monitoring through spindle vibration with LSTM
wear progression of the intelligent manufacturing system. Wear
and DNN in CNC machine tools. Measurement 210:112554
522:204697
26. El Fouki M, Aknin N, El Kadiri K (2019) Multidimensional
10. Mao Y, Tong J, Chin ZY, Borghesani P, Peng Z (2023) Transmis-
approach based on deep learning to improve the prediction per-
sion-error-and vibration-based condition monitoring of gear wear
formance of DNN models. International Journal of Emerging
with contaminated lubricant. Wear 523:204760
Technologies in Learning 14(2)
11. Liu Y, Ma D (2022) Surface modification and tooth contact anal-
27. Lai CH, Wu TE (2023) Multicriteria gear monitoring system
ysis of variable hyperbolic circular-arc-tooth-trace cylindrical
based on deep neural networks. Sensors Mater 35
gears. Mechanical Sciences 13(2):909–920
12. Lin CJ, Chu WL, Wang CC, Chen CK, Chen IT (2020) Diagnosis
Publisher's Note Springer Nature remains neutral with regard to
of ball-bearing faults using support vector machine based on the
jurisdictional claims in published maps and institutional affiliations.
artificial fish-swarm algorithm. Journal of Low Frequency Noise,
Vibration and Active Control 39(4):954–967
Springer Nature or its licensor (e.g. a society or other partner) holds
13. Wang CC, Kuo PH, Chen GY (2022) Machine learning prediction
exclusive rights to this article under a publishing agreement with the
of turning precision using optimized xgboost model. Appl Sci
author(s) or other rightsholder(s); author self-archiving of the accepted
12(15):7739
manuscript version of this article is solely governed by the terms of
14. Yuan B, Liu G, Yue Y, Liu L, Shen Y (2021) A novel tooth sur-
such publishing agreement and applicable law.
face modification methodology for wide-faced double-helical gear
pairs. Mech Mach Theory 160:104299