Measurement 207 (2023) 112351

Contents lists available at ScienceDirect

Measurement
journal homepage: [Link]/locate/measurement

Machine learning and IoT-based approach for tool condition monitoring: A

review and future prospects
Minh-Quang Tran a, b, *, Hoang-Phuong Doan c, Viet Q. Vu d, Lien T. Vu e
a
Industry 4.0 Implementation Center, Center for Cyber-physical System Innovation, National Taiwan University of Science and Technology, 10607 Taipei, Taiwan
b
Department of Mechanical Engineering, Thai Nguyen University of Technology, 3/2 Street, Tich Luong Ward, Thai Nguyen, Viet Nam
c
International Advanced Technology Program, National Taiwan University of Science and Technology, 10607 Taipei, Taiwan
d
Faculty of International Training, Thai Nguyen University of Technology, 3/2 Street, Tich Luong Ward, 250000 Thai Nguyen, Viet Nam
e
Faculty of Mechanical Engineering and Mechatronics, Phenikaa University, Ha Dong, Hanoi 12116, Viet Nam

A R T I C L E I N F O A B S T R A C T

Keywords: In the “Industry 4.0” era, autonomous and self-adaptive industrial machining attracts significant attention in
Cutting signal processing professional manufacturing. This trend originates from the rising demands for machining quantity and part
Tool condition monitoring quality, as well as the need for optimization of the overall operational cost. Under such circumstances, tool
Industry 4.0
condition monitoring (TCM) systems contribute to the mentioned goals by monitoring the cutting tool operation
Machine learning
states, which detect signs of tool failures such as cracks, chippings, fractures, and estimate the remaining tool life
IoT
Smart manufacturing for the ultimate goal of optimizing the performance of machining processes. This article presents the latest
advancements in each stage of TCM systems, namely fusion sensors methods, modern data acquisition systems
(DAQ), virtual machining, and lightweight TCM models, wherein artificial intelligence (AI) and Internet of
Things (IoT) technologies demonstrate promising operational efficiency and accuracy while showing their po­
tential in practical applications. Moreover, emphasis is paid to the current blank spots and limitations of
implementing machine learning, deep learning, and IoT technologies in tool-embedded TCM, namely big data
handling, generalization of Machine Learning models, as well as cloud computing latency, which suggests cor­
responding solutions to enable the practicality of mass implementation of Industrial Internet of Things (IIoT) for
TCM systems, such as cloud migration, custom TCM network, and shared knowledge databases between multiple
factories. In addition to a comprehensive review of state-of-the-art innovations for TCM systems, the future
prospects of intelligent TCM systems are also presented for smart manufacturing.

1. Introduction is vitally necessary since tool wear heavily affects workpiece quality,
production efficiency, machine stability, and durability of the cutting
Metal cutting is one of the core industries that supports the sus­ process. Damage to the cutting tool is often cited as the primary cause for
tainable development of the global manufacturing sector. Since the unexpected stops to the production workflow. As per research, around
introduction of the concept of the Industry 4.0 revolution in Germany in 20 % of machine downtime is due to tool wear [2]. Changing tools too
2011, the global market has witnessed a rapid surge in the demand for late when the tool wear is already excessive causes low workpiece
automation and digitalization of the existing production infrastructure quality and the risk of an unexpected shutdown. However, doing it
and machinery to meet the goal of smart manufacturing. With the recent prematurely results in tool waste and an additional increase in the total
developments in sensor technologies, embedded systems, machine processing cost. Previous studies found that cutting tools are usually
learning, and cloud computing, the process of instantly obtaining, and used only within a 50–80 % range of their effective life [3]; therefore,
processing real-time data, and achieving real-time monitoring is immediate detection of tool damage and replacement is critical and
becoming much more streamlined [1]. necessary for obtaining high-quality parts and high-efficiency processes.
A CNC machine is a crucial piece of equipment used in the modern The main purpose of a TCM system is to gather real-time cutting data
manufacturing industry. In CNC machining, the cutting tool’s integrity from sensors equipped with the machine tool and to analyze the

* Corresponding author.
E-mail address: [Link]@[Link] (M.-Q. Tran).

[Link]
Received 5 June 2022; Received in revised form 18 November 2022; Accepted 12 December 2022
Available online 16 December 2022
0263-2241/© 2022 Elsevier Ltd. All rights reserved.
M.-Q. Tran et al. Measurement 207 (2023) 112351

measured signals using signal processing approaches to monitor the main concerns for multi-sensor systems. A flexible multi-sensor
cutting conditions and detect tool wear. A powerful TCM system enables approach for tool wear monitoring systems using vibration, current,
enhancing productivity while guaranteeing product quality and there­ and force signals have been developed, in which the total power ob­
fore improves machining efficiency [4]. tained from the cutting is consistent with the tool wear curve [18].
With that said, tool wear monitoring has remained one of the biggest The tool condition system can be generally achieved following the
challenges in metal cutting due to the presence of rich non-linear fea­ procedure as shown in Fig. 2. This typically includes five steps: sensor
tures in most cutting processes, which make it challenging to directly integration, signal collection, signal processing, model prediction, and
measure the tool wear. Direct measurement of tool wear includes tool wear decision-making. The objective of the TCM system is to define
gathering visual data from optical sensors and probes like cameras [5]. the correctness of the operating states of the cutting processes. A TCM
These methods have limitations in terms of cost-efficiency, flexibility, system can be online, offline, or synced in real-time over the internet. A
and installation. Most systems in use prefer indirect measurement local TCM system involves interfering with the machining system and
methods, in which sensors are used to capture signals such as cutting analyzing the tool’s well-being state with specific checking equipment,
force, vibration signature, acoustic emission (AE), tool temperature, etc. for example, a digital microscope [19]. In web-based TCM systems, the
[6]. In this indirect approach, most previous studies used dynamometer machining system will not be intruded upon, yet the parameters will be
and AE sensors to determine the degree of tool wear. However, under estimated and compared with the instrument’s well-being state at spe­
different operating conditions, the sensor’s measurement can be costly cific time intervals. In an effective real-time TCM system, the propensity
and the results are often inconsistent [7]. It is also tricky to monitor tool of cutting tool wear or tool failure can be predicted and prevented, or
conditions in real time under manufacturing environments. In several detected and resolved automatically to prevent the outage situations of
literature studies, the conducted analysis of cutting signals is usually the cutting tool. In that concept, various cost-effective sensors such as an
measured by developed sensors that are mostly not applicable in the accelerometer, microphone, and current sensors, are integrated into the
industry due to their high cost [8]. Smart manufacturing is associated CNC machine center. The architecture of Message Queue Telemetry
with the Internet of Things (IoT) technologies, which show promising Transport (MQTT), a very lightweight IoT protocol, links the sensor
enhancement to manufacturing performance [9]. The cost-effectiveness node and the cloud server to identify machine’s the real-time status,
of IoT systems and the recent trends in machine learning-enhanced wherein the CONTACT Elements for IoT [20] is adopted to demonstrate
analytics toward open-source operating systems opens the door for the processed information graphically on the IoT dashboard.
many appropriate applications in manufacturing environments [10]. In recent years, the expanding implementation of Industry 4.0 and
Thus, a reliable tool condition monitoring system integrated with data emerging technologies enabled novel breakthroughs in the development
analytics and IoT using cutting signals is expected to be more efficient of intelligent machining systems [21]. Various machine learning [22]
for smart CNC machines than conventional cutting chatter monitoring and IoT-based [23] methods have demonstrated promising performance
systems, as shown in Fig. 1 [11]. [24], which have been constantly developed and employed in TCM,
Signal processing techniques play an important role in TCM systems wherein a large amount of input information using multi-sensor signals
as they investigate and establish the connection between tool wear and have cooperated to the tool condition through general machine learning
the input data, namely, vibration, sound, and current signals [7]. Fig. 1 [25] and deep learning techniques [26]. In addition, the recent trend of
illustrates an effective TCM system using the sound signal introduced by cloud migration has triggered the development of more powerful yet
Liu et al. [12]. The analysis of signals using signal processing techniques lightweight TCM systems to achieve online smart factory operation in
such as fast Fourier transform (FFT) [13], Wavelet packet decomposition real-time. For example, in 2021, Wang et al. proposed and prototyped an
(WPD) [14], ensemble empirical mode decomposition (EEMD), and event-driven tool condition monitoring (EDTCM), in which the system is
Hilbert Huang transform [15] are widely used to analyze cutting signals triggered “just in time” after the workpiece starts machining [27]. Deo
and determine the resonant frequency of each signal. Frequency spec­ et al. investigated the White-Box support vector machine (SVM) and its
trum analysis takes in advance for analysis of cutting signals which corresponding swarm-based optimization for TCM. The performance of
contain more non-stationary characteristics since it can display how five meta-heuristic algorithms was tested and compared [28]. In 2022,
much of the signal exists within a given frequency band regarding a Venkatesh et al. proposed and tested various transfer learning models
range of frequencies [16], while time-domain analysis can only display for TCM, the effect of hyper-parameters was compared and the best
the changes in a signal over a span of time. A tool condition monitoring performer was suggested in [29]. Bajaj et al. confirm the feasibility of
system can be improved by applying developed multi-sensor fusion Bayesian discriminant analysis for classification and monitoring in TCM
techniques since more information and patterns related to the cutting using a custom-made DAQ system and fine-tuned hyper-parameter from
tool wear are collected [17]. However, the selection of the most Bayesian optimization search, resulting in an accuracy of 93.3 %,
important features and variables correlated to tool wear is one of the making it industry ready [30]. Stuhr et al. suggested a minimal and

Fig. 1. An IoT platform for online tool condition monitoring [11].

2
M.-Q. Tran et al. Measurement 207 (2023) 112351

Fig. 2. Schematic diagram of tool condition monitoring system based on signal processing.

flexible TCM System, where the algorithms take advantage of repetitive connection of factories scattered across the world. Such a factory
machining operations in mass production and utilized similarity analysis network requires a stable cloud service provider, with suitable database
between the collected signals and data of known conditions [31]. You capacity and capable cloud computing power as well as strong optimi­
et al. developed lightweight image processing and neural network based zation of the TCM system.
on multiple activation functions for TCM, which addressed issues The realization of such a vision demands recent and future advances
regarding image quality, model parameters, dataset scale, cloud in emerging technology for TCM. Therefore, this article includes po­
migration, and embedded devices [32]. tential research that explores the aforementioned requirements of smart
Within 2021 and 2022, there have been various review articles that factory networks, namely sensor fusion methods, high-efficiency ma­
cover the implementation of emerging technology in Tool Condition chine learning models, virtual machining, TCM light-weighting, low-
Monitoring (TCM) and Machine Prognostics Health Management latency transmission for real-time monitoring, cloud migration strategy,
(PHM), which reflect the current transformation of smart factories, and integrated cyber-physical systems (CPS), while covering in-depth
reveal the trends, challenges, and evaluate novel attempts on enhancing theories and detailed analysis on every stage of industrial TCM in each
TCM System. Patil et al. reviewed generic methods based on Deep section. In particular, section 2 discuss the hardware and mechanisms
Learning for milling operations in a versatile environment to construct a for data collecting, section 3 explain the methods in the data processing
general replicable TCM framework [33]. Schwendemann et al. con­ stage, section 4 explores and compares different machine learning
ducted a specific survey on TCM for the bearings in grinding machines models and optimizers, and section 5 shows the IoT framework and
[34]. Pimenov et al. reviewed and compared the direct and indirect network infrastructure for industrial TCM. Lastly, section 6 sums up the
sensing methods for TCM in milling, and the advantages, limitations, current challenges and opens up future prospects and suggested topics,
and prospects of such measurement methods were discussed [35]. followed by section 7 which concludes the main goal of the article.
Ahmad et al. made investigation into energy harvesting techniques for
wireless sensor nodes (WSN), which directly addresses the sustainability 2. Sensor signal and data collection
problem of wireless sensors’ batteries [36]. Mohamed et al.‘s article paid
attention to the development of wirelessly embedded sensors, along Historically, experienced human operators are responsible for
with an extensive review of dimensionality reduction techniques, which monitoring the tool condition during machining processes. The signifi­
prepare for the implementation of IIOT TCM systems [37]. Butler et al. cant improvement and modernization of the TCM system underwent
outline modern techniques and methods found in recent research on different stages of evolution, from human-sensing, and direct measure­
feed drive TCM and PHM, they are categorized into sensor-less, sensor- ment, to fully sensor-based TCM [41]. The current most common sensor-
based, signal-based, and machine learning-based, with additional details based methods used for tool condition monitoring systems are indirect
on the mechanical components of feed drives [38]. Serin et al. sum­ tool wear sensing methods since they overcome the limitations of direct
marized the mechanism of TCM and presented the theory behind recent tool-wear approaches such as cost efficiency, inflexibility, and hard
deep-learning techniques [39]. Kudelina et al. discussed the significance installation [42]. For the indirect tool wear sensing method, the tool
of machine data, and the utilization of physics and mathematical models conditions are indirectly captured through sensor signals such as cutting
of industrial machines for AI model training [22]. Nasir et al. reviewed force, vibration signature, acoustic emission (AE), tool temperature, etc.
and introduced the components of intelligent monitoring; different as­ [41]. For indirect methods, most previous studies adopted the dyna­
pects of available machine learning and deep learning models for TCM mometer and AE sensors to identify the degree of tool wear. The sensor’s
were also presented and compared. Moreover, the author forecasted the signals are then handled by Data Acquisition System (DAQ), which
opportunities for data-driven smart manufacturing, including feature measures and analyzes electrical and physical quantities such as voltage,
engineering automation, high-dimensional data processing, sensor- current, acceleration, orientation, acoustics emission, etc. A DAQ system
fusion integration, and hybrid intelligence [40]. includes sensors, electrical circuitry, analog to digital converter, and
After the latest literature reviews and original research are summa­ processing software. Currently, high costs and inconsistent results of
rized and compared, it can be seen that recent literature has not fully signals collected from different operation conditions are the shortcom­
explored the vision of inter-connected smart factories networks. The ings of the sensor approaches. Typical sensors and data collectors used in
main advantage of such a network is the ability to share common digital tool condition monitoring are categorized by data sensing techniques in
knowledge of tools and machining processes on a cloud database, which Fig. 3.
is a reference for local computers to perform prediction and detection by Several sensors have been utilized to collect the cutting signals for
following standardized and frequently updated TCM models. Such a tool wear monitoring systems in previous studies. A strong relationship
smart factory network can boost system performance, enables mass between force signal and tool wear was found in previous research. The
implementation of TCM, and minimizes the need for local computing magnitudes of the static cutting forces in the frequency domain could be
hardware and custom installations. Machines in the network have access correlated to tool wear [43]. Particularly, the force in the cutting feed
to knowledge of tool faults that have not occurred at the local factory but direction [44] is the most sensitive to tool wear [45]. Vibration signature
have been recorded in another. Once the features are extracted and analysis is known as an efficient method for the identification and
learned, the database can perform automated cleanup of miscellaneous monitoring of tool wear. The change of accelerator spectrum revealed
data to optimize storage, reduce stress, and enable stable sustainability clearly in magnitude corresponding to levels of tool wear reported in
on the whole network. The shared factories’ scale ranges from multiple Ref. [46]. The use of AE sensors is known as one of the efficient ways of
plants within a local manufacturing area or extends to a large global monitoring tool wear. Because AE signals are generated directly from

3
M.-Q. Tran et al. Measurement 207 (2023) 112351

Fig. 3. Common sensing and data collection techniques in tool condition monitoring.

the cutting zone, they are very sensitive to changes in the machining tool wear levels [49]. Recently, the audible sound signal technique has
operations such as machining instability and tool wear. Moreover, the been utilized for tool wear monitoring. This method shows many ad­
higher frequency range of acoustic emission is known as an advantage of vantages and potential for practical applications. By investigating the
using AE signals to monitor tool conditions as compared to other signals correlation between audible sound signals and corresponding tool con­
[47]. For AE signals, using root mean square (RMS) is a simple and ditions from good to failure states, tool wear can be classified into
effective way to monitor tool wear. The RMS voltage could be utilized to different wear classes [24]. The correlation between the sound signal
observe the increase in frank wear [48]. Xiqing et al. investigated RMS and the cutting tool wear under various cutting conditions was inves­
values of AE signals and found a relationship between RMS values and tigated in [50]. It was found that the maximum values of the wavelet

Fig. 4. Multi-sensor installation on a milling machine [14].

4
M.-Q. Tran et al. Measurement 207 (2023) 112351

packet decomposition coefficients have the most important correlation utilized to preserve data continuity [57].
with the flank wear [12]. Furthermore, the data are collected nonstop at a progressing size
More information and wear patterns related to the cutting tool wear with a large number of dimensions. This demands large capacity storage
could be revealed and achieved using multi-sensor signals. This can and computing powers, which brings up a great problem when
enhance the tool condition monitoring system with developed multi- extracting understandable information for decision-making [58].
sensor fusion techniques [51]. However, the feature selection, which Therefore, feature representation is pre-determined from the signal
collects the most significant features and variables related to tool wear, processing stage, which can be understood as a lightweight description
is one of the biggest challenges for multi-sensor signals. A fusion of of most variables [59]. The combination of these newly made features
cutting force, vibration, and AE signals was studied by Zhou et al., [52]. stacks up the dimensionality [60] and includes possible irrelevant or
A flexible multi-sensor approach for tool wear monitoring systems using unnecessary features [61]. Such a problem can still be handled by
vibration, current, and force signals have been developed, in which the applying dimensionality reduction that finely selects more significant
total power obtained from the cutting is consistent with the tool wear features for improving the efficiency of the classification models.
curve [18]. Kandilli used multi-sensor neural network fusion for the
prediction of tool wear in drilling and milling. An example of multi-
3.2. Feature extraction
sensor installation on a milling machine, including an accelerometer,
microphone, and current sensor, is shown in Fig. 4.
Feature extraction plays a critical role in TCM systems, as it con­
The above-collected signals, however, only provide recorded data
tributes the most to the success of the classification models [62]. These
from the past and present, which leaves future faults to be less pre­
features include physical and statistical, which represent the charac­
dictable. To increase the prediction accuracy, the application of the
teristics of the input data that are normally defined in signal processing
Virtual Machining (VM) simulation method is proposed [53]. The
and later optimized in dimensionality reduction [62]. The pattern of the
commercial VM is capable of generating machining data virtually based
tool wear hidden into cutting signals can be extracted through feature
on the input parameters of machining conditions and additional un­
representation in both time responses [63], frequency, and time­
certainties. As the parameters and uncertainties are adjusted, VM
–frequency domains [64]. A comparison of feature extractions in
simulation generates data on faulty and healthy machining processes.
different domains of features is described in Table 1. It is common and
The collected data from the simulation can be used to train the model to
simple to analyze time as a feature domain. It illustrates the signal
detect faults that have not occurred in the past. The diversity of the
amplitude corresponding to a certain time. Time-domain features can be
randomly generated data is expected to be controlled by the VM pro­
statistical values, such as the average, root mean square (RMS),
gram, dimensionality reduction, and feature extraction process. By
maximum or minimum, and peak-to-peak amplitude [47]. Whereas,
reasonably adjusting the level of fitting, the classification results can be
frequency domains are made by transforming time responses onto the
more generalized.
frequency domains to extract the significant components. Various
Finally, in case of extremely high noise, versatile environment, and
transform methods such as the FFT [65], FFT improved versions [66],
cross-domain machining operation, where the TCM System is prone to
and discrete Fourier transform (DFT) [67] were widely shown in recent
unknown faults, specific data collection methods can be taken into ac­
papers. Important features can be the peak frequency and amplitude,
count. The suggested shared machining database method can be
crest factor, and other statistical values. Moreover, time–frequency
considered as a possible answer, in which raw signals from the past and
demonstration [68] of the cutting signals can be generated from the
present are shared to a cloud database between multiple factories. Such
continuous wavelet transform [12], discrete wavelet transform [69],
shared data should be labeled with meta-data, namely date, time ma­
empirical mode decomposition algorithms [70], and short time Fourier
chine type, tool type, spindle speed, feed rate, cutting depth, etc., to be
transform (STFT) [71]. Extractable features can be the wavelet coeffi­
accessible and understandable by other models. This database is a
cient average energy, as well as their statistical values [72].
reference to establish more standard operation criteria, which allows a
model from one factory to detect faults that have not occurred in its local
factory but have been recorded in another. This method also enables the 3.3. Dimensionality reduction
database for known tool faults to be collected, updated, and shared more
regularly and at a larger volume. Due to the heavy variation in cutting, materials, conditions, signals,
and features, the performance of sensors [73], as well as the degradation
3. Signal and data processing for TCM of the tool’s health, the values of constructed features change in many
possibilities [74]. This leads to the lack of certainty and limited gener­
3.1. Signal processing alization in TCM systems since the tool condition descriptive features are
unclearly assigned. It is necessary to eliminate unwanted or redundant
Pre-processing and processing are crucial for the effectiveness of a features while maintaining other information that is correlated to the
data-driven TCM system, they are initialized by a sensor conditioner for degrees of tool wear [75], to minimize the computational time of the
data digitization, corresponding to sensor characteristics and the inter­ models [76]. To handle these problems, feature normalization can be
ference from mechanical, electrical, or ambient triggering factors [19].
In general, the low-level output signal of the sensors is first amplified to Table 1
increase the signal-to-noise ratio and reduce the undesired interference. Comparison of feature extractions.
The sampling rate should be taken at 5 to 10 times the highest frequency Features Advantages Disadvantages
of interest for the acceptable presentation of the tool wear [54]. In case Time-domain Display different signals Excessive noise
the cutting signals contain noise or need to remove the undesired parts, feature immediately and faster
filtering techniques are utilized to filter the raw signals while keeping processing
the relationship between sensor signals and process variables [55]. The Frequency Specify desired frequencies Not easy to identify the
domain feature (ex: resonant cutting relevant frequency band
desired cutting signals can be obtained based on high-pass, low-pass, or
frequencies)
band-pass to remove the noise or undesired signal frequency compo­ Suitable for steady-state
nents. Moreover, segments of the signal can be generated when the systems
cutter engages with the workpiece to collect information about the tool Time-frequency Suitable for non-steady-state Does not have a standard
condition [56]. Segmentation can be applied to every tool rotation to domain feature systems procedure to select important
features
receive repetitive segments, wherein an overlapping time frame is

5
M.-Q. Tran et al. Measurement 207 (2023) 112351

applied to filter out features that are more connected to the tool’s health the tool wear is to be monitored progressively [8]. Effective models that
using statistics from the candidate features or empirical formulas [77]. have been applied to TCM systems can be named, such as the artificial
Other approaches have been done including using ANOVA and f-test to neural networks (ANN) model [87], support vector machine [88],
eliminate features that are more related to cutting states and less related Bayesian neural networks [89], K-Nearest Neighbor (KNN) model [90],
to the tool condition [57]. Input data with high dimensions is chal­ Decision Tree [91], fuzzy logic [92], and Gaussian process regression
lenging for TCM that are built from features from different domains, (GPR) [93].
which reduces the classifier’s accuracy where irrelevant or noisy ele­ In particular, several tree-based algorithms such as decision trees,
ments are taken in [78]. Moreover, a large feature model turns out to be tree ensembles (bagged, boosted trees, random forest), and logistic
exclusively subject to the data quality, which leads to overfitting and model tree models are proposed to classify the failure-free and various
faulty results on any new datasets [79]. Therefore, before applying failure states of the cutting tool [94]. Bagging is used when the variation
dimensional reduction, input data are remapped from a high- of a decision tree needs to be minimized since the base algorithm is
dimensional to a lower-dimensional space, using features selection unstable while Boosting technique sets up several consecutive trees that
and extraction, to filter out the most unique and dominating ones from solve for the error of the previous tree so that the final total error is
the initial feature set. In the local machine model training, only highly minimized. Khairnar et al. [95] and Khade et al. [96] demonstrated the
relatable features should be recognized and included to train an accurate use of a Bagged and Boosted tree in different tool wear fault diagnosis
and power-efficient online TCM system [80]. problems. Patange et al. have conducted a large comparison between 5
In multi-sensor TCM [81], dimensionality reduction techniques can different tree-based algorithms, which results in the Best First Tree al­
achieve many advantages when being applied to the dataset: (1) gorithm achieving the superior accuracy of up to 90 % and a marginally
significantly optimizing the epoch time and the data space; (2) by higher model constructing time of 8.6 s [97]. Pantage et al. proved the
removing noisy and irrelevant features, AI classifiers can be more effi­ classification of tool conditions in turning achieved the best results with
cient and accurate; (3) by evaluating and visualizing patterns, the white-box Random Forest classifiers. Compared to black-box modeling
classification problem can be illustrated more informationally [82]. such as deep learning models, the white-box tree-based models for TCM
Dimensional reduction is divided into two methods known as feature systems allow significantly better transparency in model interpretation
selection and feature transformation. and examination. This is crucial when certain features of the data need
The objective of utilizing dataset feature selection is to figure out the to be examined to determine the cause of the occurred failures [98].
most important features for monitoring tool wear. It also helps to opti­ In the case of classification tasks of tool faults without labels (blind
mize the computational power and enhance the precision of the model data), a Semi-Supervised Learning (SSL) method is suggested with a
while retaining useful information. Generally, feature selection sorts the smaller set of labeled data. Peikarim et. al. proposed the use of the SSL
features referring to how sensitive they are to the tool’s health state and method and a pre-labeled dataset to classify a bigger dataset and
selects the most significant ones. In addition to feature selection, feature enhance the overall performance of the classification task [99]. The
transformations are techniques based on some optimization criteria to author states that it is crucial to ensure a clear distribution of the labeled
apply feature transformation algebraically on the input feature set, dataset before SSL can be implemented effectively on blind data.
resulting in a lower dimensional representation, which reduces the Additionally, The SSL method has also been implemented by Patange et.
demanded computing power. The process outputs a dataset of newly al., where the dataset is first separated into labeled and unlabeled. The
constructed features that represent the initial input traits while keeping labeled dataset is trained by traditional learning models, while the latter
the important description, contrasted to feature selection. However, undergoes a feature extraction process that forms data clusters with
those original traits in the transformed features cannot be explicated, similar properties before the trained model assigns them a class or label
and the information about how each original feature contributes to the [97].
transformation cannot be kept [83]. Although numerous feature trans­ Meta-heuristic optimizers, such as contemporary Swarm-based al­
formation algorithms have been developed, selecting a suitable algo­ gorithms, mimic the clustering of plants and animals in nature and
rithm depends on the quantity, characteristics, and quality of the cutting showcase superior performance in classification tasks. Tejas et al. have
dataset. Since these algorithms generate features without any under­ carried out multiple performance tests of different swarm-based algo­
standable information, recovery techniques such as the Fisher discrim­ rithms, which concluded that the Harris Hawks Optimization exhibit the
inant analysis can be applied to keep the initial information of the new best performance at a slightly above-average epoch time [28]. A sche­
features. This later helps to interpret the data during system perfor­ matic diagram of the proposed tool condition monitoring system with
mance tuning [84]. machine learning classifiers is described in Fig. 5. These algorithms
The current techniques in feature extraction and dimensionality commonly take in manually selected features and have their corre­
reduction are highly dependent on the advancement of data mining in sponding strengths and weaknesses, which were discussed in details in
general machine learning. This stage of a TCM system faces the same [100].
challenges as the data mining, data exploration, and data processing Recently, artificial neural network models are among the most
stage of classic machine learning problems, namely the discriminating commonly used machine learning approaches, as they are frequently
ability and interpreting ability of each extracted feature, as well as the applied in building prediction models for detecting stages of tool wear
prevention, detection, and recovery of lost data during extraction and progression [43]. The idea of neural networks takes inspiration from the
selection process [85]. These challenges can either become a new ma­ natural structures of the brain neurons, as they learn the patterns from
chine learning problem or a valuable research topic in data science that given datasets and make the decision for unseen datasets. In general, a
aims to further enhance the efficiency of big data handling. neural network model consists of inter-connected groups of multilayer
neurons that connect the features to the corresponding results. The
4. Machine learning and deep learning for TCM system neural network is trained by updating the value of the weights and bias,
which are the connection strength between each node, to build a map of
4.1. Machine learning for TCM how likely each node results in the next node in the network [46].
Commonly, a neural network is applied to analyze uncertainty, and
The goal of all Machine Learning models for TCM is to categorize tool unclear relationships and reveal hidden patterns in big data. Neural
conditions, estimate remaining life, and predict future failure of the networks have been commonly applied in TCM thanks to their robust­
machining tool, which helps improve the machining efficiency and ness and adaptivity [101]; however, they come with numerous draw­
reduce the workpiece damage [86]. The decision-making stage is nor­ backs such as slower epoch time, local minima, and additional work
mally handled by classifier-based Machine Learning, especially when needed for tuning multiple weights and biases [102]. The data-driven

6
M.-Q. Tran et al. Measurement 207 (2023) 112351

a tricky paradox: if it can recognize misclassification, it wouldn’t have

misclassified it. Therefore, the suggested methodology to reduce
misclassification is to sample a larger dataset, improve the machine
learning model accuracy, and assign an experienced human supervisor
to many machines. In particular, once a TCM system falls into the failure
zone, the experienced human operator is expected to notice faulty data
on the dashboard and take action to resolve the problem.

4.2. Deep learning models for TCM

With the recent jump in computational power, deep learning is un­

doubtedly an approachable and efficient way to solve classification.
Without the need of choosing and assigning the features one by one,
deep learning automatically maps the relationship features from the
input signals to the suitable output condition. Several sensors collect the
cutting signals for the tool wear monitoring systems in previous studies.
A strong relationship between force signal and tool wear was found in
previous research, the magnitudes of the static cutting forces in the
frequency domain could be correlated to tool wear. In particular, a deep
learning network model has multiple network layers that can learn the
pattern and generate representative features automatically within the
original dataset [110]. It can outperform the conventional machine
learning approaches that required deep expertise and knowledge for
analyzing signals to design and select the important features within the
dataset [39], as illustrated in Fig. 6.

4.3. Transfer learning models for TCM

In simple terms, transfer learning, which is a subset of deep learning,

Fig. 5. Flowchart of the tool condition monitoring topology with machine happens when a previously trained model is reused as the reference
learning classifiers. baseline for another model on a related task, where the learning transfer
is accomplished by transferring features, and either updating or trans­
models were also tested to enhance the robustness and accuracy of the ferring the weights [111]. For instance, a model trained on the first
system [103]. The conclusion decision was selected from a vote between machining process is used as an optimization for a second related pro­
various classification outputs. A review of various state-of-arts of ma­ cess. This method allows rapid enhancement for the training of the
chine learning approaches for decision-making in TCM is summarized in second task and establishes a relationship between the two tasks.
Table 2. To cut down the training resource requirement [112], the transfer
The probability of misclassification, also known as False Negative learning approach’s effectiveness has been recently tested and proven
and False Positive, shares the same idea as the level of accuracy in a [113]. Venkatesh et al. executed a novel attempt to examine the per­
machine learning model. Such possibilities increase proportionally with formances of pre-trained deep learning neural networks for TCM, such
the complexity of the machining process and the uncertainty of the as Alex Net, Google Net, and Res-Net 50, which were trained on massive
working environment. When such problems arise, the model itself faces established datasets [114]. The pre-training process provides a base
model, whose knowledge is then applied to related problems. The results

Table 2
Reviews of machine learning-based prediction models for tool monitoring.
Authors Cutting signal Features Processing methods Prediction Model Model
Accuracy

Salgado et al. Vibration Frequency Singular spectrum analysis Multilayer Neural Network RMSE: 15.11
[104]
Kilundu et al. Vibration Frequency Singular spectrum analysis Neural Network 67.4 % accuracy
[105]
Miao et al. [106] Vibration Frequency Convolutional neural Convolutional neural network 99.92 %
network accuracy
Segreto et al. Force, Acoustic Emission, Frequency Linear predictive analysis Neural Network 98.9 % accuracy
[107] Vibration
Seemuang et al. Sound Time-Frequency STFT Tested spindle noise at various feed –
[50] rates
Liu et al. [12] Sound Time-Frequency WPD ANN Model 8.59 % error
Salgado et. al. Motor Current, Sound Signals Time-Frequency Singular spectrum analysis LS-SVM 4.94–8.72 %
[104] error
Tran et al. [71] Cutting Force Signals Time-Frequency Continuous Wavelet CNN 99.67 %
transform accuracy
Kothuru et al. [24] Sound Frequency FFT SVM 95.92 %
accuracy
Yao et al. [108] Vibrations Time, Frequency, Time- FFT NN based on fuzzy rules 0.0003 % MSE
Frequency
Lu et al. [109] Sound Frequency FFT Hidden Markov Model 91.8 % accuracy

7
M.-Q. Tran et al. Measurement 207 (2023) 112351

Fig. 6. Traditional machine learning along with deep learning model for TCM [39].

show that all pre-trained neural networks displayed high classification accuracy. TCM models that are created with a targeted tool can also be
accuracy but with different batch sizes. While developing TCM, instead re-used to monitor unlearned related tools to optimize the training
of training a model from scratch, the transfer learning method can be resources.
applied readily with advantages in training time and prediction Although current attempts at building a TCM system have utilized

Fig. 7. A four-layer industrial IoT architecture [116].

8
M.-Q. Tran et al. Measurement 207 (2023) 112351

most of the recent and sophisticated emerging technologies, a general, and connected smart factories, since current local computers have been
stable, and comprehensive real-time TCM has not been established and facing challenges in executing calculations and decision-making, as the
is commonly recognized [37]. Current TCM systems are custom- minimum demand for data latency is within milliseconds.
designed to work in controlled conditions and well-understood envi­
ronments, where the uncertainty is low, and the specific tool and 5.2. Possible industrial IoT application
machining processes are already well-defined. This problem triggers the
demand for a more general but systematic approach, with the ability to The purpose of this section is to introduce an example of an intelli­
interchange one system for different processes and conditions, which gent IoT infrastructure, which integrates deep learning algorithms to
requires the machine learning models to be trained on more diverse tool analyze the machining procedure to protect tool integrity and tool life.
conditions and feature sets while managing to retain the models’ size. An online TCM architecture for CNC machines is described in Fig. 8,
Such research avenues have the potential to enable the practicality of wherein the CNN milling center is equipped with multiple sensors and a
TCM systems for general processes across multiple smart factories. Raspberry Pi or an Intel Edison board to capture and ingest the tool’s
health data in real-time. The Intel Edison platform enables data inges­
5. IoT architecture and possible industrial applications tion, which imports real-time data stream into an intermediate database,
additional evaluation of industrial settings, and pre-testing of the service
5.1. IoT architecture for tool condition monitoring using simulation techniques. The measured signals are delivered
through an amplifier and modulator, before being processed by the
In recent years, the quick development of IoT technologies put for­ LabVIEW program and uploaded to a Microsoft Structured Query Lan­
ward numerous possibilities and opportunities for developing industrial guage Server (MS-SQL). The sensor-collected data will be the input of
applications, particularly for building smart factories and smart the machine learning and deep learning model, specially created for
manufacturing. Fig. 7 presents a general four-layer industrial IoT ar­ real-time monitoring of the tool condition, as shown in Fig. 8.
chitecture, which consists of a perception layer with sensing devices; a The input data of the multi-sensor signals are sent through the hid­
network layer for transmitting the data [115]; a processing layer for den layers, which train the model to predict the tool wear during the
storing, collecting, and analyzing the data; and application modules in cutting process. Then, “CONTACT Elements for IoT” [20], a web-based
the last layer [116]. Various researchers and research teams had made platform for business applications using a Digital Twin, is used to
novel attempts to the implementation of the IoT-based online TCM clearly illustrate the analyzed information and control the tool condi­
system. Each of these research projects has different approaches. tion. This platform is combined with standard MQTT protocol to
Moreover, machine learning, deep learning, and IoT architecture can demonstrate more meaningful information onto a graphical dashboard
also facilitate several industrial processes such as monitoring machining to support further signal processing tasks. Further problems such as
stability [11], fault diagnosis of induction motors [117], monitoring of cyber-attack or sensor malfunctions due to environmental temperature,
the gas-insulated switchgear [118], cybersecurity for factory, and en­ humidity, or noisy signals can also be handled by the system. Then, the
ergy management for smart buildings [119]. IoT platform can notify a warning sign on the dashboard and automat­
The recent survey reveals that many ideas for IoT-based systems for ically switch the system to the backup broker to reliably and safely
machining process monitoring were developed. Raja et al. discussed the continue the processes.
evolution of modern DAQ compared to conventional DAQ, including the The raw data is pre-processed using the aforementioned signal pro­
integration of IoT protocols, increased number of data channels, and cess techniques. Then, the significant features are generated from the
additional industrial network capability. Moreover, modern IoT-enabled pre-processed signals such as statistical indexes in the time–frequency
DAQ systems are utilized to transfer data in organized data streams at a domain. Those features are used to recognize the current cutting con­
faster speed between sensors, IoT devices, gateways, servers, and users dition. In the proposal architecture, the sensor node can transmit data to
in real-time [120]. Teti et al. accumulated accomplishments in the field the SQL server, in which a database system is implemented for data
of cutting-edge testing machining conditions and introduced four management. The transmission takes place using a wireless connection.
essential key elements of the TCM system and presented a design of a The communication between the sensor node and the cloud server is
cloud platform for observing cutting tools in turning processes [121]. implemented using MQTT. Which is a message protocol utilized to
Tapoglou et al. proposed a cloud-based platform for the selection of an transfer telemetry data, especially in areas with limited network ca­
ideal machining state [122]. Chen et al. proposed an energy proficiency pacity [129].
TCM framework for an IoT-embedded factory, in which an IoT-based The information can be visualized through a graphical dashboard
energy monitoring system is executed on machine shops. The profi­ after being processed, using CONTACT Elements (IoT Version) with
ciency of the proposed framework is directly applicable in modern fac­ standard MQTT protocols, as shown in Fig. 9. The difficulty will be how
tories [123]. Saif et al. designed a computer vision-based IoT framework to monitor the tool wear accurately, secure the system, and make the
for distinguishing and investigating the flaws in the machining system decisions and actions to the machine’s controller in real-time, especially
[124]. An online computing method to forecast cutting tool conditions when the system is affected by cyberattacks or the inefficient reading of
while efficiently handling the information from collected big data is the sensors due to environmental temperature, humidity, and noise
proposed in [125]. Wu et al. introduced a fuzzy computing method for signals [114].
data-driven machine learning and online monitoring in cyber-physical Setiawan et al. likewise proposed a cutting tool condition monitoring
manufacturing, in which numerous remote sensors have been installed system to support the trend of Industry 4.0. The TCMS comprises a three-
on CNC machines to gather large amounts of constant data [126]. A sensor fusion with a UI to display the ongoing cutting tool condition
detailed cloud computing migration strategy has been proposed by Peng [130]. The sensors incorporate a temperature sensor, an accelerometer,
et al., to further reduce transmission latency from the cloud [127]. Cloud and an electrical power sensor. The obtained information was compared
computing solution, despite being more powerful, faces other latency with the flank wear and hole wear level that is estimated on the cutting
due to network transmission and require a migration strategy, such as devices. The model shows that the temperatures, vibration, and power
using an adaptive mobile terminal that communicates with the cloud utilization data can be gathered by the sensors and sent through the web
and obtains migration decisions for separated sub-tasks. You et al. to the monitoring framework. The early prototypes of this TCMS have
proposed a lightweight network model for tool condition monitoring been created and tested with a few training models. The motivation
based using multiple activation functions, data augmentation and cloud- behind this preparation is to decide whether the equipment can send and
edge collaboration, which reduce the size of the system but still retain store information in a dataset. The information is then processed [129]
the data quality [128]. These strategies bridge the gap to fully integrated to reveal the connection between the cutting tool uncertainty and

9
M.-Q. Tran et al. Measurement 207 (2023) 112351

Fig. 8. A suggested intelligent Industrial IoT system for monitoring smart CNC machines.

Fig. 9. Implementation of IoT platform for online tool condition monitoring [11].

temperature variety, vibration strength, as well as power usage [130]. approach, which consists of three layers that link cyberspace and
One more insightful condition monitoring system was presented by Cao physical space. In the first layer, the sensors gather the machine, tool,
et al. in [131]. Which, they aimed to develop a method that does not and workpiece operating data. Additional information is also obtained
only involve the capability of error detection or failure prediction but from the manufacturing environment and combined with the machine
also obtain the data, which can characterize the monitoring manufacturers’ experience. After that, the machining dataset is stored
manufacturing processes. and analyzed in the intermediate layer, namely the cyber-physical
Apart from CONTACT Elements and licensed platforms, an IoT Sys­ interface. In this stage, various techniques like big data algorithms
tem can also be constructed using only open-source software and and machine learning models are utilized to process the collected data to
hardware. Bajaj et. al. proposed a self-developed and replicable DAQ understand the cutting processes and to learn from manufacturers’ ex­
system that utilizes the ADXL335 accelerometer sensor, Arduino Mega, periences. The third layer, namely the cyberspace layer, is in charge of
Microsoft Excel, and Parallax-DAQ IoT platform [132]. The Parallax- decision-making regarding machine failure prediction and maintenance.
DAQ is a fully open-source platform that can transfer data from the This layer also provides the machine failure models which are generated
Arduino digital output over the air to Microsoft Excel. This setup can from the mining of sensor data. Ontologies are the base to construct the
constantly receive data from up to 26 microcontrollers and saves it into knowledge of machine health, and ontological reasoning approaches are
the Excel datasheet. It offers researchers a fast and straightforward setup used to estimate machine damage, deterioration of machines, and future
using commonly available hardware and software that offers a cus­ maintenance demands. In the end, the model results are transferred
tomizable adaptation to real-time conditions. downwards to the physical space, thus it can help the manufacturer to
Fig. 10 illustrates an intelligent TCM system using a cyber-physical make the right decision promptly.

10
M.-Q. Tran et al. Measurement 207 (2023) 112351

Fig. 10. Schematic view of intelligent condition monitoring system [131].

Recent research on cyber-physical systems and tool condition which have simplified installation processes and systematic manage­
monitoring are shaping the future of smart manufacturing. However, ment, using custom-designed network infrastructure for TCM, with the
current IoT architecture and network infrastructure for TCM are facing scale ranging from regional to global networks. More commonly, the
higher requirements in terms of connectivity, reliability, and systematic researchers can focus on how to improve the handling and transferring
frameworks, which results from the increasing demand for more so­ capability of diverse and high-dimensional data, explore different ma­
phisticated machine learning, a more robust data pipeline, lower latency chine learning models for TCM, and construct more versatile and
cloud computing, and spreading connectivity across multiple factories. general-purpose TCMs that can interchange one system for different
These trends suggest further research to either develop a specifically re- workspaces and conditions. Moreover, building such a large network of
designed network for TCM implementation or new solutions to versatile monitoring systems raises the demand for a solid cyber-security
strengthen the available infrastructure. defender for TCM. Additionally, knowledge of TCM models based on the
mechanics of material and material science is highly necessary. Algo­
6. Challenges and future prospects rithms that utilize the theoretical relationship between the data of force,
torque, fracture, creep, and the tool material as well as geometry would
Different solutions to each stage of Tool Condition Monitoring have be a strong scientific approach for TCM data analysis. Lastly, researchers
been proposed by top-of-the-field researchers to create more industry- can pay additional attention to the economics aspect of the TCM system,
ready TCM systems. Regardless of the huge leap in recent research, which defines the practicality for mass implementation and realization
several aspects of modern TCM still face obstacles that require further of smart factories.
intensive investigation. These challenges include the integration of
sensors into standard industrial machines, the loss of data during data 7. Conclusion
processing, the efficiency versus the accuracy and the generalization of
Machine Learning models, the possibility of online and offline infra­ Tool wear is a common and inevitable phenomenon during many
structure for shared databases, cloud computing latency, the vulnera­ machining operations. It raises numerous challenges towards the pro­
bility of TCM against cyber-attack, as well as the rising expense for TCM ductivity and quality of automated machining processes in different
system development. aspects, which obstruct the potential for smart and inter-connected in­
The aforementioned challenges open up various corresponding dustrial factories. Therefore, the development of an intelligent online
research possibilities. Future studies are suggested to investigate the tool condition monitoring system based on the Industrial Internet of
feasibility of shared machining databases between multiple factories, Things for smart factories is promised to be a key contributor to

11
M.-Q. Tran et al. Measurement 207 (2023) 112351

improving the performance of modern manufacturing. This article re­ [11] M.Q. Tran, et al., Reliable deep learning and IoT-based monitoring system for
secure computer numerical control machines against cyber-attacks with
views the recent novel state-of-the-art innovations for intelligence tool
experimental verification, IEEE Access 10 (2022) 23186–23197, [Link]
condition monitoring systems integrated with machine learning and IoT 10.1109/ACCESS.2022.3153471.
technologies, with a view to form a network of smart factories. The [12] M.-K. Liu, Y.-H. Tseng, and M.-Q. Tran. Tool wear monitoring and prediction
current online TCM system is constructed by connecting the cloud IoT based on sound signal. In: The International Journal of Advanced Manufacturing
Technology 103(9);2019:3361-3373, doi: 10.1007/s00170-019-03686-2.
platform for tool condition monitoring to an industrial machine with [13] M.-K. Liu, Q. M. Tran, Y.-W. Qui, C.-H. Chung, Chatter detection in milling
integrated sensor fusion. The sensors are employed to collect and send process based on time-frequency analysis, 2017. [Online]. Available: [Link]
the signals of the machining process to the IoT platform, where data org/10.1115/MSEC2017-2712.
[14] M.-Q. Tran, M.-K. Liu, M. Elsisi, Effective multi-sensor data fusion for chatter
analysis, signal processing, and machine learning algorithms are utilized detection in milling process, ISA Trans. (2021), [Link]
to perform tool condition monitoring and machining control. The ulti­ isatra.2021.07.005.
mate goal of this research topic is to establish a network of smart, dy­ [15] C. Chung, M.-Q. Tran, M.-K. Liu, Estimation of process damping coefficient using
dynamic cutting force model, Int. J. Precis. Eng. Manuf. 21 (4) (2020) 623–632,
namic, and responsive factories that have the full ability to self-supervise [Link]
and perform decision-making on machines and processes, while [16] V. Srinivasan, C. Eswaran, N. Sriraam, Artificial neural network based epileptic
communicating through an intelligent TCM network with cutting-edge detection using time-domain and frequency-domain features, J. Med. Syst. 29(6);
2005:647-660. doi: 10.1007/s10916-005-6133-1.
human–machine corporations. [17] Y. Zhou, W. Xue, A multisensor fusion method for tool condition monitoring in
No conflicts of interest have been identified in connection with this milling, Sensors 18(11);2018:3866 [Online]. Available: [Link]
publication. The authorship of the manuscript has been approved by all 1424-8220/18/11/3866.
[18] X.Y. Zhang, X. Lu, S. Wang, W. Wang, W.D. Li, A multi-sensor based online tool
authors, and no other persons meet authorship requirements but are not
condition monitoring system for milling process, Procedia CIRP 72 (2018)
listed. There has been no disagreement among the authors as to the 1136–1141, [Link]
order of the listed authors. [19] A. Mohamed, M. Hassan, R. M’Saoubi, H. Attia, Tool condition monitoring for
high-performance machining systems – A review. Sensors 22(6) (2022) 2206.
[Online]. Available: [Link]
[20] IoT Platform for Digital Business Models|CONTACT Software. [Link]
Declaration of Competing Interest
[Link]/en/products/iot-platform-for-digital-business-models/?
fbclid=IwAR0oYDd4qHpCd0BEZaGrLHEAQGYoQ2BhBmDzbF35-
The authors declare that they have no known competing financial cyM6QrNHAkziWDC8yo (accessed November 22, 2021).
interests or personal relationships that could have appeared to influence [21] J. Chen, et al., Toward intelligent machine tool, Engineering 5 (4) (2019)
679–690, [Link]
the work reported in this paper. [22] K. Kudelina, T. Vaimann, B. Asad, A. Rassõlkin, A. Kallaste, G. Demidova, Trends
and challenges in intelligent condition monitoring of electrical machines using
Data availability machine learning, Appl. Sci. 11 (6) (2021) 2761.
[23] M. Schwenzer, K. Miura, T. Bergs, Machine Learning for tool wear classification in
milling based on force and current sensors, in: IOP Conference Series: Materials
No data was used for the research described in the article. Science and Engineering, vol. 520, p. 012009, 2019/04/17 2019, doi: 10.1088/
1757-899x/520/1/012009.
[24] A. Kothuru, S.P. Nooka, R. Liu, Application of audible sound signals for tool wear
Acknowledgment monitoring using machine learning techniques in end milling, Int. J. Adv. Manuf.
Technol. 95 (9) (2018) 3797–3808, [Link]
1.
This research was financially supported by the Ministry of Science [25] M. Li, M. Burzo, Tool wear monitoring using machine learning, in: 2021 IEEE
and Technology in Taiwan under Grant no. MOST 110-2222-E-011-002- Canadian Conference on Electrical and Computer Engineering (CCECE), 12-17
Sept. 2021 2021, pp. 1-5, doi: 10.1109/CCECE53047.2021.9569060.
and the “Center for Cyber-physical System Innovation” from The
[26] A. Gouarir, G. Martínez-Arellano, G. Terrazas, P. Benardos, S. Ratchev, In-process
Featured Areas Research Center Program within the framework of the tool wear prediction system based on machine learning techniques and force
Higher Education Sprout Project by the Ministry of Education (MOE) in analysis, Procedia CIRP 77 (2018) 501–504, [Link]
procir.2018.08.253.
Taiwan.
[27] Y.H. Wang, L.Y. Zheng, Y.W. Wang, Event-driven tool condition monitoring
methodology considering tool life prediction based on industrial internet,“ (in
References English), J. Manuf. Syst. 58 (2021) 205–222, [Link]
jmsy.2020.11.019.
[28] T.Y. Deo, A. D. Patange, S.S. Pardeshi, R. Jegadeeshwaran, A.N. Khairnar, H. S.
[1] L.S. Dalenogare, G.B. Benitez, N.F. Ayala, A.G. Frank, The expected contribution
Khade, A white-box SVM framework and its swarm-based optimization for
of Industry 4.0 technologies for industrial performance, Int. J. Prod. Econ. 204
supervision of toothed milling cutter through characterization of spindle
(2018) 383–394, [Link]
vibrations, arXiv preprint arXiv:2112.08421, 2021.
[2] M. Kious, A. Ouahabi, M. Boudraa, R. Serra, A. Cheknane, Detection process
[29] S. Naveen Venkatesh, et al., Transfer learning-based condition monitoring of
approach of tool wear in high speed milling, Measurement 43 (10) (2010)
single point cutting tool, Comput. Intell. Neurosci. (2022) 2022.
1439–1446, [Link]
[30] N.S. Bajaj, A.D. Patange, R. Jegadeeshwaran, K.A. Kulkarni, R.S. Ghatpande, A.M.
[3] X. Hongjian, Y. Kechong, Y. Rong, The shape characteristic detection of tool
Kapadnis, A Bayesian optimized discriminant analysis model for condition
breakage in milling operations, Int J Mach Tool Manu 37 (11) (1997) 1651–1660,
monitoring of face milling cutter using vibration datasets, J. Nondestruct.
[Link]
Evaluat. Diagnost. Prognost. Eng. Syst. 5(2) (2022).
[4] N. Ambhore, D. Kamble, S. Chinchanikar, V. Wayal, Tool condition monitoring
[31] B. Stuhr, R. Liu, A flexible similarity-based algorithm for tool condition
system: a review, Mater. Today:. Proc. 2 (4) (2015) 3419–3428, [Link]
monitoring, J. Manufact. Sci. Eng. 144(3) (2022).
10.1016/[Link].2015.07.317.
[32] Z. You, H. Gao, S. Li, L. Guo, Y. Liu, J. Li, Multiple activation functions and data
[5] A.A. Thakre, A.V. Lad, K. Mala, Measurements of tool wear parameters using
augmentation based light weight network for in-situ tool condition monitoring,
machine vision system, Model. Simulat. Eng. 2019 (2019) 1876489, [Link]
IEEE Trans. Ind. Electron. (2022).
org/10.1155/2019/1876489.
[33] S. Patil, S. Pardeshi, A. Patange, R. Jegadeeshwaran, Deep learning algorithms for
[6] M. Kuntoğlu et al., A review of indirect tool condition monitoring systems and
tool condition monitoring in milling: a review, in J. Phys.: Conf. Ser. 2021 1969,
decision-making methods in turning: critical analysis and trends, Sensors, 21(1)
no. 1: IOP Publishing, p. 012039.
(2021) 108 [Online]. Available: [Link]
[34] S. Schwendemann, Z. Amjad, A. Sikora, A survey of machine-learning techniques
[7] T. Mohanraj, S. Shankar, R. Rajasekar, N.R. Sakthivel, A. Pramanik, Tool
for condition monitoring and predictive maintenance of bearings in grinding
condition monitoring techniques in milling process — a review, J. Mater. Res.
machines, Comput. Ind. 125 (2021).
Technol. 9 (1) (2020) 1032–1042, [Link]
[35] D.Y. Pimenov, M.K. Gupta, L.R. da Silva, M. Kiran, N. Khanna, G.M. Krolczyk,
[8] C. Nath, Integrated tool condition monitoring systems and their applications: a
Application of measurement systems in tool condition monitoring of milling: a
comprehensive review, Procedia Manuf. 48 (2020) 852–863, [Link]
review of measurement science approach, Measurement (2022).
10.1016/[Link].2020.05.123.
[36] I. Ahmad, L.M. Hee, A.M. Abdelrhman, S.A. Imam, M. Leong, Scopes, challenges
[9] H. Kim, W.-K. Jung, I.-G. Choi, and S.-H. Ahn, A low-cost vision-based monitoring
and approaches of energy harvesting for wireless sensor nodes in machine
of computer numerical control (CNC) machine tools for small and medium-sized
condition monitoring systems: a review, Measurement 183 (2021).
enterprises (SMEs), Sensors 19(20);2019:4506 [Online]. Available: [Link]
[37] A. Mohamed, M. Hassan, R. M’Saoubi, H. Attia, Tool condition monitoring for
[Link]/1424-8220/19/20/4506.
high-performance machining systems—a review, Sensors 22 (6) (2022) 2206.
[10] K. Zhu, G. Li, Y. Zhang, Big data oriented smart tool condition monitoring system,
IEEE Trans. Ind. Inf. 16 (2020) 4007–4016.

12
M.-Q. Tran et al. Measurement 207 (2023) 112351

[38] Q. Butler, Y. Ziada, D. Stephenson, S. Andrew Gadsden, Condition monitoring of [65] S. Paul, S. Mukhopadhyay, S. Mandal, Quantification of tool wear in micro-
machine tool feed drives: a review, J. Manuf. Sci. Eng. 144 (10) (2022). milling using Fast Fourier transform of accelerometer data, in: 2019 3rd
[39] G. Serin, B. Sener, A.M. Ozbayoglu, H.O. Unver, Review of tool condition International Conference on Electronics, Materials Engineering & Nano-
monitoring in machining and opportunities for deep learning, Int. J. Adv. Manuf. Technology (IEMENTech), 29-31 Aug. 2019 2019, pp. 1-3, doi: 10.1109/
Technol. 109 (3) (2020) 953–974, [Link] IEMENTech48150.2019.8981301.
w. [66] M.Q. Tran, M.K. Liu, Chatter identification in end milling process based on
[40] V. Nasir, F. Sassani, A review on deep learning in machining and tool monitoring: cutting force signal processing, IOP Conf. Series: Mater. Sci. Eng. 654 (1) (2019),
methods, opportunities, and challenges, Int. J. Adv. Manuf. Technol. 115 (9) [Link]
(2021) 2683–2709. [67] C. o.-K. Chen, Y.-C. Li, Intelligent real-time monitoring of Computer Numerical
[41] M.S. Bhuiyan, I. Choudhury, review of sensor applications in tool condition Control tool wear based on a fractional-order chaotic self-synchronization system,
monitoring in machining, Comprehensive materials processing 13 (2014) 539- J. Low Freq. Noise Vibrat. Active Control 38(3-4) (2019) 1555-1566 2018, doi:
569, doi: 10.1016/B978-0-08-096532-1.01330-3. 10.1177/1461348418790503.
[42] H. Chen, Investigation of the Methods for Tool Wear On-Line Monitoring during [68] P.-y. Li, Y.-w. Fang, Y. Wang, M. Yang, Q. Yuan, Y. Li, Time-frequency analysis for
the Cutting Process, in: Computer and Computing Technologies in Agriculture IV, cutting tools wear characteristics, in: 2006 International conference on machine
Berlin, Heidelberg, D. Li, Y. Liu, and Y. Chen, Eds., 2011// 2011: Springer Berlin learning and cybernetics, 2006, pp. 3308-3312.
Heidelberg, pp. 215-220. [69] M.Q. Tran, M.K. Liu, Q.V. Tran, T.K. Nguyen, Effective fault diagnosis based on
[43] D.E. Dimla, P.M. Lister, On-line metal cutting tool condition monitoring: I: force wavelet and convolutional attention neural network for induction motors, IEEE
and vibration analyses, Int J Mach Tool Manu 40 (5) (2000) 739–768, https:// Trans. Instrum. Meas. 71 (2022) 1–13, [Link]
[Link]/10.1016/S0890-6955(99)00084-X. TIM.2021.3139706.
[44] L. Xiaoli, L. Han-Xiong, G. Xin-Ping, R. Du, Fuzzy estimation of feed-cutting force [70] J.C. Jáuregui, J.R. Reséndiz, S. Thenozhi, T. Szalay, J.Á.M. Takács, Frequency and
from current measurement-a case study on intelligent tool wear condition Time-frequency analysis of cutting force and vibration signals for tool condition
monitoring, IEEE Trans. Syst. Man Cybernet. C (Appl. Rev.) 34 (4) (2004) monitoring, IEEE Access 6 (2018) 6400-6410, doi: 10.1109/
506–512, [Link] ACCESS.2018.2797003.
[45] Z.W. Zhong, J. Zhou, Y.N. Win, Correlation analysis of cutting force and acoustic [71] M.-Q. Tran, M.-K. Liu, Q.-V. Tran, Milling chatter detection using scalogram and
emission signals for tool condition monitoring, in: 2013 9th Asian Control deep convolutional neural network, Int. J. Adv. Manuf. Technol. 107 (3) (2020)
Conference (ASCC), 23-26 June 2013 2013, pp. 1-6, doi: 10.1109/ 1505–1516, [Link]
ASCC.2013.6606333. [72] N. Fang, P.S. Pai, S. Mosquea, Effect of tool edge wear on the cutting forces and
[46] W. Rmili, R. Serra, A. Ouahabi, C. Gontier, K. Mecheri, Tool wear monitoring in vibrations in high-speed finish machining of Inconel 718: an experimental study
turning process using vibration measurement, 2006. and wavelet transform analysis, Int. J. Adv. Manuf. Technol. 52 (1) (2011) 65–77,
[47] X. Chen, B. Li, Acoustic emission method for tool condition monitoring based on [Link]
wavelet analysis, Int. J. Adv. Manuf. Technol. 33 (9) (2007) 968–976, [Link] [73] M.S.H. Bhuiyan, I.A. Choudhury, 13.22 - Review of Sensor Applications in Tool
org/10.1007/s00170-006-0523-5. Condition Monitoring in Machining, in: S. Hashmi, G.F. Batalha, C.J. Van Tyne,
[48] P.S. Pai, P.K.R. Rao, Acoustic emission analysis for tool wear monitoring in face B. Yilbas (Eds.), Comprehensive Materials Processing, Elsevier, Oxford, 2014,
milling, Int. J. Prod. Res. 40 (5) (2002) 1081–1093, [Link] pp. 539–569.
00207540110107534. [74] B. Zhang, C. Katinas, Y.C. Shin, Robust tool wear monitoring using systematic
[49] M. Xiqing, X. Chuangwen, Tool wear monitoring of acoustic emission signals from feature selection in turning processes with consideration of uncertainties,
milling processes, in: 2009 First International Workshop on Education J. Manuf. Sci. Eng. 140 (8) (2018) pp, [Link]
Technology and Computer Science, 7-8 March 2009 2009, vol. 1, pp. 431-435, [75] K. Jemielniak, T. Urbański, J. Kossakowska, S. Bombiński, Tool condition
doi: 10.1109/ETCS.2009.105. monitoring based on numerous signal features, Int. J. Adv. Manuf. Technol. 59 (1)
[50] N. Seemuang, T. McLeay, T. Slatter, Using spindle noise to monitor tool wear in a (2012) 73–81, [Link]
turning process, Int. J. Adv. Manuf. Technol. 86 (9) (2016) 2781–2790, https:// [76] C. Zhang, X. Yao, J. Zhang, H. Jin, Tool condition monitoring and remaining
[Link]/10.1007/s00170-015-8303-8. useful life prognostic based on a wireless sensor in dry milling operations (in eng),
[51] C. Aliustaoglu, H.M. Ertunc, H. Ocak, Tool wear condition monitoring using a Sensors (Basel) 16 (6) (2016) 795, [Link]
sensor fusion model based on fuzzy inference system, Mech. Syst. Sig. Process. 23 [77] J.L. Ferrando Chacón, T. Fernández de Barrena, A. García, M. Sáez de Buruaga,
(2) (2009) 539–546, [Link] X. Badiola, J. Vicente, A novel machine learning-based methodology for tool wear
[52] Y. Zhou, W. Xue, A multisensor fusion method for tool condition monitoring in prediction using acoustic emission signals [Online]. Available: Sensors 21 (17)
milling (in eng), Sensors (Basel) 18 (11) (2018) 3866, [Link] (2021) 5984 [Link]
s18113866. [78] K. Goebel, W. Yan, Feature selection for tool wear diagnosis using soft computing
[53] Y.Q. Xichun Luo, Multiscale Modeling of Hybrid Machining Processes (Hybrid techniques, 2000. [Online]. Available: [Link]
Machining), Academic Press, Elsevier, 2018. 1803.
[54] Y. Qin, D. Wang, Y. Yang, Integrated cutting force measurement system based on [79] X. Ying, An overview of overfitting and its solutions, J. Phys. Conf. Ser. 1168
MEMS sensor for monitoring milling process, Microsyst. Technol. 26 (6) (2020) (2019), [Link]
2095–2104, [Link] [80] J.V. Abellan-Nebot, F. Romero Subirón, A review of machining monitoring
[55] M. Rizal, J.A. Ghani, M.Z. Nuawi, C.H.C. Haron, An embedded multi-sensor systems based on artificial intelligence process models, Int. J. Adv. Manuf.
system on the rotating dynamometer for real-time condition monitoring in Technol. 47 (1) (2010) 237–257, [Link]
milling, Int. J. Adv. Manuf. Technol. 95 (1) (2018) 811–823, [Link] [81] J. Yuan, L. Liu, Z. Yang, Y. Zhang, Tool wear condition monitoring by combining
10.1007/s00170-017-1251-8. variational mode decomposition and ensemble learning (in eng), Sensors (Basel,
[56] S. Bombiński, K. Błażejak, M. Nejman, K. Jemielniak, Sensor signal segmentation Switzerland) 20 (21) (2020) 6113, [Link]
for tool condition monitoring, Procedia CIRP 46 (2016) 155–160, [Link] [82] C.H. Lauro, L.C. Brandão, D. Baldo, R.A. Reis, J.P. Davim, Monitoring and
org/10.1016/[Link].2016.03.203. processing signal applied in machining processes – a review, Measurement 58
[57] M. Hassan, A. Sadek, M.H. Attia, V. Thomson, A novel generalized approach for (2014) 73–86, [Link]
real-time tool condition monitoring, J. Manuf. Sci. Eng. 140 (2) (2017) pp, [83] Y. Yang, et al., A novel tool (single-flute) condition monitoring method for end
[Link] milling process based on intelligent processing of milling force data by machine
[58] U. Sivarajah, M.M. Kamal, Z. Irani, V. Weerakkody, Critical analysis of Big Data learning algorithms, Int. J. Precis. Eng. Manuf. 21 (11) (2020) 2159–2171,
challenges and analytical methods, J. Bus. Res. 70 (2017) 263–286, [Link] [Link]
org/10.1016/[Link].2016.08.001. [84] Z. Xie, J. Li, Y. Lu, Feature selection and a method to improve the performance of
[59] I. Guyon, S. Gunn, M. Nikravesh, L.A. Zadeh, Feature Extraction: Foundations and tool condition monitoring, Int. J. Adv. Manuf. Technol. 100 (9) (2019)
Applications, Springer, Berlin Heidelberg, 2008. 3197–3206, [Link]
[60] M.-Q. Tran, M. Elsisi, M.-K. Liu, Effective feature selection with fuzzy entropy and [85] W. Jia, M. Sun, J. Lian, S. Hou, Feature dimensionality reduction: a review,
similarity classifier for chatter vibration diagnosis, Measurement 184 (2021), Complex Intell. Syst. (2022) 1–31.
[Link] [86] M. Hu, W. Ming, Q. An, M. Chen, Tool wear monitoring in milling of titanium
[61] M.-Q. Tran, Y.-C. Li, C.-Y. Lan, M.-K. Liu, Wind farm fault detection by monitoring alloy Ti–6Al–4 V under MQL conditions based on a new tool wear categorization
wind speed in the wake region [Online]. Available: Energies 13 (24) (2020) 6559 method, Int. J. Adv. Manuf. Technol. 104 (9) (2019) 4117–4128, [Link]
[Link] 10.1007/s00170-019-04125-y.
[62] R. Zebari, A. Abdulazeez, D. Zeebaree, D. Zebari, J. Saeed, A comprehensive [87] M. Wiciak-Pikula, A. Felusiak, and P. Twardowski, “Artificial Neural Network
review of dimensionality reduction techniques for feature selection and feature models for tool wear prediction during Aluminium Matrix Composite milling, in
extraction, J. Appl. Sci. Technol. Trends 1 (2) (2020) 56–70, [Link] 2020 IEEE 7th International Workshop on Metrology for AeroSpace
10.38094/jastt1224. (MetroAeroSpace), 22-24 June 2020 2020, pp. 255-259, doi: 10.1109/
[63] S. Jianming, L. Yongxiang, W. Gong, Z. Mengying, Milling tool wear monitoring MetroAeroSpace48742.2020.9160064.
through time-frequency analysis of sensory signals, in: 2016 IEEE International [88] G.F. Wang, Y.W. Yang, Y.C. Zhang, Q.L. Xie, Vibration sensor based tool condition
Conference on Prognostics and Health Management (ICPHM), 20-22 June 2016 monitoring using ν support vector machine and locality preserving projection,
2016, pp. 1-5, doi: 10.1109/ICPHM.2016.7542826. Sens. Actuat. A 209 (2014) 24–32, [Link]
[64] M. Du, P. Wang, J. Wang, Z. Cheng, S. Wang, Intelligent turning tool monitoring [89] J. Dong, K.V.R. Subrahmanyam, Y.S. Wong, G.S. Hong, A.R. Mohanty, Bayesian-
with neural network adaptive learning, Complexity 2019 (2019) 8431784, inference-based neural networks for tool wear estimation, Int. J. Adv. Manuf.
[Link] Technol. 30 (9) (2006) 797–807, [Link]

13
M.-Q. Tran et al. Measurement 207 (2023) 112351

[90] S. Nießner, M. Liewald, M. Weyrich, A methodology to quantify tool wear effects [112] W.D. Li, Y.C. Liang, Deep transfer learning based diagnosis for machining process
in a shear cutting process based on an automatic feature extraction, IFAC- lifecycle, Procedia CIRP 90 (2020) 642–647, [Link]
PapersOnLine 55 (2) (2022) 540–547, [Link] procir.2020.02.048.
ifacol.2022.04.250. [113] M. Marei, S.E. Zaatari, W. Li, Transfer learning enabled convolutional neural
[91] G. Li, Y. Wang, J. He, Q. Hao, H. Yang, J. Wei, Tool wear state recognition based networks for estimating health state of cutting tools, Rob. Comput. Integr. Manuf.
on gradient boosting decision tree and hybrid classification RBM, Int. J. Adv. 71 (2021), [Link]
Manuf. Technol. 110 (1) (2020) 511–522, [Link] [114] M. Elsisi, M.Q. Tran, K. Mahmoud, D.-E.A. Mansour, M. Lehtonen, M.M.
05890-x. F. Darwish, Effective IoT-based deep learning platform for online fault diagnosis
[92] J.C. Chen, V. Susanto, Fuzzy logic based in-process tool-wear monitoring system of power transformers against cyberattacks and data uncertainties, Measurement
in face milling operations, Int. J. Adv. Manuf. Technol. 21 (3) (2003) 186–192, 190 (2022), [Link]
[Link] [115] H.N. Dai, Z. Zheng, Y. Zhang, Blockchain for internet of things: a survey, IEEE
[93] D. Kong, Y. Chen, N. Li, Gaussian process regression for tool wear prediction, Internet Things J. 6 (5) (2019) 8076–8094, [Link]
Mech. Syst. Sig. Process. 104 (2018) 556–574, [Link] JIOT.2019.2920987.
ymssp.2017.11.021. [116] N. Abosata, S. Al-Rubaye, G. Inalhan, C. Emmanouilidis, Internet of things for
[94] A.D. Patange, R. Jegadeeshwaran, N.S. Bajaj, A.N. Khairnar, N.A. Gavade, system integrity: a comprehensive survey on security, attacks and
Application of machine learning for tool condition monitoring in turning, Sound countermeasures for industrial applications, Sensors, vol. 21, no. 11, p. 3654,
Vibrat 56 (2) (2022) pp, [Link] 2021. [Online]. Available: [Link]
[95] A. Khairnar, A. Patange, S. Pardeshi, R. Jegadeeshwaran, Supervision of carbide [117] M.Q. Tran, M. Elsisi, K. Mahmoud, M.K. Liu, M. Lehtonen, M.M.F. Darwish,
tool condition by training of vibration-based statistical model using boosted trees Experimental setup for online fault diagnosis of induction machines via promising
ensemble, Int. J. Perform. Eng. 17 (2) (2021). IoT and machine learning: towards industry 4.0 empowerment, IEEE Access 9
[96] H. Khade, A. Patange, S. Pardeshi, R. Jegadeeshwaran, Design of bagged tree (2021) 115429–115441, [Link]
ensemble for carbide coated inserts fault diagnosis, Mater. Today:. Proc. 46 [118] M. Elsisi, M.Q. Tran, K. Mahmoud, D.E.A. Mansour, M. Lehtonen, M.M.
(2021) 1283–1289. F. Darwish, Towards secured online monitoring for digitalized GIS against cyber-
[97] A.D. Patange, R. Jegadeeshwaran, A machine learning approach for vibration- attacks based on IoT and machine learning, IEEE Access 9 (2021) 78415–78427,
based multipoint tool insert health prediction on vertical machining centre [Link]
(VMC), Measurement 173 (2021). [119] M. Elsisi, M.-Q. Tran, K. Mahmoud, M. Lehtonen, M. M. F. Darwish, Deep
[98] A.D. Patange, R. Jegadeeshwaran, Application of bayesian family classifiers for Learning-Based Industry 4.0 and Internet of Things towards Effective Energy
cutting tool inserts health monitoring on CNC milling, Int. J. Prognost. Health Management for Smart Buildings, Sensors, vol. 21, no. 4, p. 1038, 2021. [Online].
Manage. 11 (2) (2020) pp. Available: [Link]
[99] M. Peikari, S. Salama, S. Nofech-Mozes, A.L. Martel, A cluster-then-label semi- [120] H.A. Raja, T. Vaimann, A. Rassõlkin, A. Kallaste, A. Belahcen, in: IEEE, 2021,
supervised learning approach for pathology image classification, Sci. Rep. 8 (1) pp. 737–742.
(2018) 1–13. [121] R. Teti, K. Jemielniak, G. O’Donnell, D. Dornfeld, Advanced monitoring of
[100] Y. Zhou, W. Xue, Review of tool condition monitoring methods in milling machining operations, CIRP Ann. 59 (2) (2010) 717–739, [Link]
processes, Int. J. Adv. Manuf. Technol. 96 (5) (2018) 2509–2523, [Link] 10.1016/[Link].2010.05.010.
10.1007/s00170-018-1768-5. [122] N. Tapoglou, J. Mehnen, A. Vlachou, M. Doukas, N. Milas, and D. Mourtzis,
[101] K. Venkatesh, M. Zhou, R.J. Caudill, Design of artificial neural networks for tool Cloud-based platform for optimal machining parameter selection based on
wear monitoring, J. Intell. Manuf. 8 (3) (1997) 215–226, [Link] function blocks and real-time monitoring, J. Manufact. Sci. Eng. 137(4) (2015),
10.1023/A:1018573224739. doi: 10.1115/1.4029806.
[102] J. Lee, H.J. Choi, J. Nam, S.B. Jo, M. Kim, S.W. Lee, Development and analysis of [123] X. Chen, C. Li, Y. Tang, Q. Xiao, An Internet of Things based energy efficiency
an online tool condition monitoring and diagnosis system for a milling process monitoring and management system for machining workshop, J. Clean. Prod. 199
and its real-time implementation, J. Mech. Sci. Technol. 31 (12) (2017) (2018,) 957–968, [Link]
5695–5703, [Link] [124] Y. Saif, et al., Development of a smart system based on STEP-NC for machine
[103] E. Kannatey-Asibu, J. Yum, T.H. Kim, Monitoring tool wear using classifier fusion, vision inspection with IoT environmental, Int. J. Adv. Manuf. Technol. 118 (11)
Mech. Syst. Sig. Process. 85 (2017) 651–661, [Link] (2022) 4055–4072, [Link]
ymssp.2016.08.035. [125] H. Li, X. Zhang, S. Tao, A cloud computing-based approach for efficient
[104] D.R. Salgado, F.J. Alonso, Tool wear detection in turning operations using processing of massive machine tool diagnosis data, J. Circ. Syst. Comput. (2021)
singular spectrum analysis, J. Mater. Process. Technol. 171 (3) (2006) 451–458, 2150297.
[Link] [126] D. Wu, et al., A fog computing-based framework for process monitoring and
[105] B. Kilundu, P. Dehombreux, X. Chiementin, Tool wear monitoring by machine prognosis in cyber-manufacturing, J. Manuf. Syst. 43 (2017) 25–34, [Link]
learning techniques and singular spectrum analysis, Mech. Syst. Sig. Process. 25 org/10.1016/[Link].2017.02.011.
(1) (2011) 400–415, [Link] [127] C. Peng, H. Du, T.W. Liao, A research on the cutting database system based on
[106] M. He, D. He, A new hybrid deep signal processing approach for bearing fault machining features and TOPSIS, Rob. Comput. Integr. Manuf. 43 (2017) 96–104.
diagnosis using vibration signals, Neurocomputing 396 (2020) 542–555, https:// [128] Z. You, H. Gao, S. Li, L. Guo, Y. Liu, J. Li, Multiple Activation functions and data
[Link]/10.1016/[Link].2018.12.088. augmentation-based lightweight network for in situ tool condition monitoring,
[107] T. Segreto, A. Simeone, R. Teti, Multiple sensor monitoring in nickel alloy turning IEEE Trans. Ind. Electron. 69(12) (2022).
for tool wear assessment via sensor fusion, Procedia CIRP 12 (2013) 85–90, [129] E. Uhlmann, A. Laghmouchi, C. Geisert, E. Hohwieler, Smart wireless sensor
[Link] network and configuration of algorithms for condition monitoring applications, J.
[108] C. Zhang, X. Yao, J. Zhang, H. Jin, Tool condition monitoring and remaining Mach. Eng. 2017.
useful life prognostic based on a wireless sensor in dry milling operations, Sensors [130] A. Setiawan, D. Angela, B. Irawan, Modelling of cutting tool condition monitoring
16(6) (2016) 795 [Online]. Available: [Link] system (CTCMS) to support industry 4.0, MATEC Web Conf., vol. 204, p. 02012,
795. 2018. [Online]. Available: [Link]
[109] M.-C. Lu, B.-S. Wan, Study of high-frequency sound signals for tool wear [131] Q. Cao, F. Giustozzi, C. Zanni-Merk, F. de Bertrand de Beuvron, C. Reich, Smart
monitoring in micromilling, Int. J. Adv. Manuf. Technol. 66 (9) (2013) Condition monitoring for industry 4.0 manufacturing processes: an ontology-
1785–1792, [Link] based approach, Cybernet. Syst. 50(2) (2019) 82-96, doi: 10.1080/
[110] X. Zhang, C. Han, M. Luo, D. Zhang, Tool wear monitoring for complex part 01969722.2019.1565118.
milling based on deep learning, Appl. Sci. 10(19) (2020), doi: 10.3390/ [132] A.D.P. Naman, S. Bajaj, R. Jegadeeshwaran, Kaushal A. Kulkarni, Rohan S.
app10196916. Ghatpande, Atharva M. Kapadnis, A Bayesian optimized discriminant analysis
[111] J. Dou, S. Jiao, C. Xu, F. Luo, L. Tang, X. Xu, Unsupervised online prediction of model for condition monitoring of face milling cutter using vibration datasets,
tool wear values using force model coefficients in milling, Int. J. Adv. Manufact. ASME J Nondestructive Evaluation, 2022.
Technol. 109 (3) (2020) 1153–1166, [Link]
05684-1.

14