You are on page 1of 10

Nonlinear Engineering 2021; 10: 404–413

Zhe Mi, Tiangang Wang, Zan Sun, and Rajeev Kumar*

Vibration signal diagnosis and analysis of rotating


machine by utilizing cloud computing
https://doi.org/10.1515/nleng-2021-0032
Received Jul 6, 2021; accepted Sep 17, 2021.
1 Introduction
Abstract: Vibration signal diagnosis and analysis plays The productivity loss, time loss, safety and environment
an important role in the industrial machinery since it en- problems are resulted by the industrial equipment fail-
hances the machinery performance under supervision. ure. The organization’s manufacturing operations are re-
The information regarding the future condition is given quired to be maintained and it is important for availability
by vibration diagnosis techniques which is growing inter- and reliability of product and production facilities [1, 2].
est for the scientific and industrial communities. Informa- The maintenance gained less attention as compared to
tion for failure diagnostic and prediction are provided by the manufacturing problems and it leads to low mainte-
the motor vibration through signal processing. The devel- nance efficiency in industry. The equipment maintenance
opment of mechanical systems fault prognosis and in the is very important and critical aspect as the maintenance
last decades, research is done at a very rapid rate. The ex- activities carried by the traditional reactive maintenance
amination of vibration signal monitoring is done in this only after failure detection. About the periodicity of fail-
paper with the aid of Cyber-Physical Systems (CPS) and ure, the periodic maintenance activities are implied by
Cloud Technology (CT). The machines maintenance strate- the widespread preventive maintenance. Predictive main-
gies are implemented by using the data collected from ma- tenance is required for cost saving and to prevent the fail-
chines which are based on the fault prognosis. The cloud ure of equipment. For various industries, motors are es-
computing platform is presented in this paper which is sential and are used for various applications such as elec-
having three layers and the unlabelled data is received to tric generators, window lifting motor, etc. Sample of motor
generate an interpreted online decision. Feature extrac- is shown in Figure 1. There is an unexpected stop if there
tion of the vibration signal is obtained in terms of range, are failures in the motor and loss of production time and
mean value, root mean square value, and standard devi- money is caused by it. The online condition monitoring is
ation and crest values. The performance of the model is developed by the researchers to avoid the motor failure.
evaluated by utilizing the classical statistical metrics such The fault diagnosis techniques are used for the detection of
as RMSE Mean Absolute Error (MAE) and Mean Absolute faults and to provide the diagnostic information of motors.
Percentage Error (MAPE) of the vibration signal. It is ob- Vibrations signals depend on the motor states and fault is
tained that the proposed technique is 25% and 90% bet- connected with vibration signals [3–6]. For fault diagno-
ter than the Adaptive Neurofuzzy Inference System and the sis, characteristic frequencies and the states of the motor
Single Modeling System respectively in terms of RMSE. The correlation is required.
performance in terms of MAPE, then the proposed tech- The early fault diagnostic methods detect the early
nique outperforms the existing Adaptive Neurofuzzy In- fault state and sometimes the motor is damaged in only 5
ference System and the Single Modeling System by 8 % minutes by short-circuiting. For the permanently damage
and 60% respectively. The presented technique is better of motor, faults like air-gap, and broken rotor take much
than the existing Adaptive Neurofuzzy Inference System time [7, 8]. The analysis of diagnostic signals is done by the
and the Single Modeling techniques by average of 15% and
30 % respectively.

Keywords: Vibration signal; cloud technology; cyber- Zhe Mi, Tiangang Wang, Baoding Vocational and Technical Collage,
physical systems; diagnosis and prognosis; rapid rate; Baoding 071051, China, E-mail: Miezhe1010@outlook.com, wang-
mean absolute error; single modeling system tiangang211@outlook.com
Zan Sun, Baoding University, Baoding 071000, China, E-mail:
sunzan677@outlook.com
*Corresponding Author: Rajeev Kumar, Chitkara University Institute
of Engineering and Technology, Chitkara University, Punjab, India,
E-mail: rajeev_chauhan364@yahoo.co.in

Open Access. © 2021 Zhe Mi et al., published by De Gruyter. This work is licensed under the Creative Commons Attribution
alone 4.0 License.
Zhe Mi, Tiangang Wang, Zan Sun, and Rajeev Kumar, Vibration signal diagnosis and analysis | 405

condition monitoring of electric motors. Measurements of c. The cloud computing platform having three layers and
acoustic signal and thermal signal can be done without the unlabelled data are received to interpreted online
touching the motor. decision generation.
d. Feature extraction of the vibration signal is obtained
in terms of range, mean value, root mean square
value, and standard deviation and crest values.

The rest of the paper is organized as follows. Section


2 provides an overview of the exhaustive literature survey
followed by a methodology adopted in section 3. A detailed
discussion of obtained results is in section 4. Finally, con-
cluding remarks are provided in Section 5.
Figure 1: Sample of motor in industries

The fault diagnosis is performed by the model-based 2 Literature review


methods which are relied on analytical redundancy and
analytical models expect the process behavior. The math- Many researchers have worked on the motor vibration di-
ematical model is exploited by analytical models of the agnosis based on cloud computing in the previous years.
monitored machine [9]. The residual generation methods The electric motors are monitored in real-time; the au-
based on this explicit model are utilized to obtain signals thor details the testing of an Industrial Internet of Things
which are indicative of fault presence in the machine. The (IIoT) system [14]. For the capability comparison, the fre-
vibration signals i.e., the residuals evaluation; the fault de- quency domain analysis is carried out in the gateway. The
tection and isolation are achieved as shown in Figure 2. edge and fog computing is advantageous which is carried
by the complement to cloud computing. In an industrial
dairy plant, testing of the prototype is done in the labo-
ratory. The author presents the diagnosing failures in ro-
tary machines method by utilizing the Machine Learning
techniques [15]. For fault diagnosis of the rotational un-
Figure 2: Basic block diagram of fault diagnosis balance, author proposed a support vector machine algo-
rithm in the rotor. For vibration analysis, the most popular
classification methods is support vector machines (SVMs).
The linear system models, process models and behav- The vibration-based fault diagnosis techniques are pre-
ioral models have been examined and included in vari- sented by the author in this paper which is based on the
ety of models [10–13]. The faults diagnosing of mechani- motor [16]. The signal processing methods and the vibra-
cal system components such as bearings rotors are applied tion signals are utilized by the proposed method. The three
for model-based methods. A model is built accurately by states of healthy CM, CM with broken tooth on sprocket,
model-based approaches effectively. CM with broken rotor coil recognition efficiency are ana-
Research gaps: Information for failure diagnostic and lyzed by the authors. The prediction through signal pro-
prediction are provided by the motor vibration through sig- cessing and the information for failure diagnostic are pro-
nal processing. The fault alarm systems examination and vided by the motor vibration [17]. The IoT-based monitor-
monitoring are not effective in terms of mean value, root ing system is presented by the author in this paper for
mean square value, and standard deviation. recording the vibration of the induction motor. The vibra-
Contribution: tion data from the motor is collected by the 27mm piezo-
electric. The 3 scenarios experiments are conducted for
a. The examination and monitoring of alarm systems, the vibration data collection from different positions. The
with the aid of Cloud Technology (CT). recorded vibration is sent by the each node vibration to the
b. Implementation based on the machines fault progno- cloud with an average delay of 1 sec.
sis by utilizing the machines’ data collection. In the last few decades, fault prognosis have been re-
searched and developed in mechanical systems at a very
rapid rate [18]. The approaches-based fault diagnosis and
406 | Zhe Mi, Tiangang Wang, Zan Sun, and Rajeev Kumar, Vibration signal diagnosis and analysis

prognosis are investigated in this paper by the authors and


predictive maintenance knowledge is obtained. For fault
3 Methodology
prognosis, the applications of the data mining methods
The motor monitoring status is detected by the several
are illustrated in the case study. The author in this paper
types of data collection methods. There are different types
presents a remote observing system dependent on Inter-
of signals from which the vibration signal is the common
net of Things (IoT) for protected and economic data com-
fault detection technique. The vibration signal is analyzed
munication [19]. The parameters like Temperature, vibra-
and illustrated here in this section.
tions, speed, motor current and voltage of induction ma-
chine are screened by the module of transducers and sen-
sors. The Automatic ON/OFF for faulty conditions is addi-
3.1 Analysis of vibration signal features
tionally presented by the system which shows the informa-
tion on LCD show and data is passed on. The model gives
The feature extraction is the vibration signal’s first step.
information to WIFI IoT applications to make the system
The rotating system failure status is diagnosed by these
quick and easy to understand. Due to the potential advan-
forms [23–25]. The diagnosis technique’s accuracy is im-
tages, the considerable subjects of system are the Machine
proved by extracting the hidden data. The times domains
fault prognostic techniques [20]. There is rapid growth
feature calculation from the vibration signal areas given in
from the past few years in the machine fault diagnosis and
equations.
prognosis. The wide range of statistical approaches is cov-
ered by the publications to model-based approaches. Also, Range (R) = max(y i ) − min(y i ) (1)
the challenges and the opportunities are also discussed
in machine prognosis field. Author details the consider- PN
able condition-based maintenance system i.e., faults prog- i=1 yi
Mean Value (M) = (2)
nostic techniques of machine [21]. The decreasing main- N
tenance costs and increasing machine availability, poten-
s
tial advantages are gained by the maintenance strategies. P 2
(y i − y” )
There is rapid development of the machine fault diagno- Standard Deviation (D) = (3)
(N − 1)
sis and prognosis research. The recent published tech-
niques in diagnosis and prognosis of rotating machinery v
are summarized and classified by synthesizing and pro-
u
u1 XN

viding the information of these researches. In the field Root Mean Square (RMS) = t (y i − y” )2 (4)
N
i=1
of machine prognosis, various challenges and the oppor-
tunities are discussed by the authors for conducting ad-
vance research. The machine vision, feature extraction, max (y i )
Crest Factor = (5)
and support vector machine (SVM) are utilized and de- RMS
tailed by the authors in this study for the vibration mon-
itoring system (VMS) to composition [22]. A vision-based PSNR = 10 * log10 (2n − 1)2 /MSE (6)
data-driven methodology is presented by the author for
industrial robot health assessment. The experimental evi-
Structural Similarity Index (SSIM) =
dence of the methodology is provided on the system which
comprises the five binary squared fiducial markers and two (2µx µy + c1) (2σx, y + c2)
(7)
µ2x + µ2Y + C1 σ2X + σ2Y + C2
 
monocular cameras. The deviation of the end-effect or is
accurately tracked by the fiducial marker system. The tra- µ x and µ y = average of x and y.
jectory deflection is monitored by utilizing the weights at- σ2 x and σ2 y = variance of x and y.
tached to the end-effect or [23]. The supervised learning re- σ x,y = Covariance of x and y.
gression models are trained by tracing the trajectory infor-
mation. The presented protocol is robust, rigorous and re-
liable and identifies the mechanical element that produces 3.2 The clustering analysis
the non-kinematic errors. The presented VMS technique’s
experiment setup is straightforward and the surveillances The vibration signal information is illustrated and the
are cooperated in industrial environments. nodes are found between the classes by feature utilization
Zhe Mi, Tiangang Wang, Zan Sun, and Rajeev Kumar, Vibration signal diagnosis and analysis | 407

in equation.
( )
1 if u ≥ α
b(u) = (8)
0 if u < α

3.5 Pattern generation


The formulation of a mixed-integer and linear program-
ming based on pattern generation. The discriminating
powers between the different classes are controlled by the
approach [31]. By the pattern generation algorithm, the
generated number of patterns has no limit to create more
Figure 3: Clustering analysis patterns finding possibilities.

after the vibration signals analysis. Basic clustering analy- 3.6 Unlabelled remote diagnosis system
sis presentation is shown in Figure 3. The homogenous ob-
architecture
jects are characterized by the clustering analysis deal with
the complex data-sets. In maintenance innovation, Cloud technologies (CT) are
The best clusters are obtained by the K-mean cluster- becoming the trend for the machines maintenance and
ing as it is commonly utilized algorithm [26–28]. The sums industrial process. CT is cost effective and has merits of
of squared distances are minimized by the clustering pro- unlimited storage space and security. The machines are
cess. Some steps are utilized by this technique. required to be continuously monitored for the achieve-
ment of near-zero breaks down [32–34]. The architecture of
1. The randomness of K centroid is the initial based. cyber-physical system maintenance is detailed in Figure 3.
2. To the nearest centroid, each point of the dataset is al- 1) Sensor Layer: The multisource unlabelled data is pro-
located. vided by the sensors consisted in the sensor layer. The
3. Each data point is recomputed to the cluster of cen- wireless communication capability of the sensor enables
troid cluster for intra-cluster reduction. the data transfer for the decision making layer [35]. The
machining process is reported by the intelligent sensors
and shares with the process controller.
3.3 Logical Analysis of Data (LAD) 2) Connection layer: The file storage distribution and par-
allel computing is consisted in this layer. The physical and
The pattern recognition based phenomena is allowed by the cyber side connection occur by CT utilization [36, 37].
data-driven LAD technique, training and testing are the 3) Decision-making layer: The unlabelled data is then
two stages which are applied. The data is utilized as train- transferred by the analyzer software and then sent back
ing data for patterns extraction [29]. The training data is to the maintenance team.
well balanced to build the diagnosis model for healthy and The internet-based services which are cloud services
failure classes. offer all the things that are provided by the computer sys-
tem. There is an overwhelming data flow due to the data
collation increment and CT has the merits of virtualized re-
3.4 Data binarization sources and data service integrated with scalable storage
of data. The best performance is provided by the virtual-
By using a binarization technique, the transformation of ized machine with low cost. Figure 4 shows the unlabelled
raw data into binary data in which each numerical feature cyber-physical system’s proposed architecture.
is translated into the set of binary attributes. The data set The sensor in the model takes a monitored machine
in ascending order are utilized by the binarization tech- signal. The online remote diagnosis center converts the
nique [30]. Between every two values, the cut point is in- collected unlabelled data into the useful data and then the
serted and then the average is calculated, value as shown status of the machine is determined. An alarm and correc-
tive action are sent by the system to prepare the required
element. The system accuracy is improved by the cloud.
408 | Zhe Mi, Tiangang Wang, Zan Sun, and Rajeev Kumar, Vibration signal diagnosis and analysis

Figure 6: Mean values (M) of the different samples

Figure 4: Cyber-physical system

Figure 7: Standard Deviation (SD) of the different samples

4 Results and discussion


4.1 Analysis of the system

To test and analyze the system, there is a development of


cyber-physical maintenance system of rolling element fail-
ure simulation model to test the system. The samples of the
datasets are publically available and can access online. A
snapshot of the vibration signal is collected by the card un-
Figure 8: Root Mean Square (RMS) of the different samples
til a failure occurred. Table 1 illustrates feature extraction
using MATLAB Software.
The vibration signal feature extraction is tabulated in
above table in terms of range, mean, standard deviation,
root mean square and the crest values. For better visual-
ization and analysis these obtained values are graphically
represented in Figures 5-9 for different samples.

Figure 9: Crest Factor (CF) of the different samples

tics patterns by using the new data after clustering as ex-


hibited in Table 2. The negative patterns in the condition
of health status are discovered by positive patterns as gen-
erated by LAD.
Figure 5: Range values (M) of the different samples
The quality of the healthy and unhealthy class is pre-
sented graphically in Figure 10.
A healthy or unhealthy condition is lead by the ma- The data set file sent the vibration signal to the cloud
chine operation status after detecting by the characteris- file in the Drop-box application via internet. Two centers
Zhe Mi, Tiangang Wang, Zan Sun, and Rajeev Kumar, Vibration signal diagnosis and analysis | 409

Table 1: Feature extraction of samples

Standard
Samples Range Mean Root Mean Square Crest Factor
Deviation
1 0.834 -0.002 0.071 0.074 6.120
2 0.729 -0.019 0.073 0.079 7.621
3 0.183 -0.021 0.078 0.032 6.865
4 0.729 -0.003 0.038 0.054 6.678
5 0.319 -0.042 0.021 0.210 7.562
6 1.872 -0.038 0.359 0.261 8.572
7 3.210 -0.041 0.048 0.311 5.563
8 1.628 -0.218 0.029 0.201 6.672
9 0.873 -0.331 0.071 0.022 7.352
10 2.577 -0.214 0.063 0.182 6.563

Table 2: Results of K-mean clustering

Sample Class Quantity


1 Healthy 750
2 Unhealthy 205

provement of the offline governor patterns generation ac-


curacy. The stored historical data is used for the automatic
software update.

4.2 Statistical analysis of the signal


vibration in system

The performances of the models are assessed by utilizing


the classical statistical metrics such as RMSE Mean Ab-
Figure 10: Quantity of healthy and unhealthy class
solute Error (MAE) and Mean Absolute Percentage Error
(MAPE). These are the basic and mostly used performance
metrics which are accepted by the research community.
are included in it, the physical center and the cyber cen- These are used for the models development is analyzed by
ter. The sensors, actuators and the maintenance team of the RMSE. This parameter is the amplitude sensitive. The
the monitoring machine are included in it. The data from forecast evaluation is done by the MAE and it is insensitive
the physical center is received by the cyber center which to the outliers. The mean deviation of each sample deter-
is then converted into the useful information for proper mination is done by the MAPE and amplitude normalizes
monitoring and controlling of physical world. The local the samples. It also helps to unify the scale and can be used
maintenance team is allowed by the system to see a live to errors of signals comparison with different levels of am-
view of the running machine interaction. Additionally, the plitudes. Four scenarios have been considered for analy-
cyber-physical system code is built by the MATLAB. The sis, that is, the healthy condition (HC), a bearing failure
signals are sent and received by the gateway and cyber- (BF) condition, a half-broken rotor bar (HBRB) condition,
physical center generates the corrective action via the in- and full-broken rotor bar (FBRB). These different parame-
ternet. High scalability and lower costs are the benefits of ters obtained for four different scenarios by the proposed
using this gateway. Analysis of the received data and the method as shown in Table 3. It is also represented graphi-
generation of suitable decision, every certain time is run cally in Figure 11 for better visualization and analysis.
by the built governing code. The positive pattern is identi- From the Figure 10, it is clear that the all these param-
fied and an alarm report is automatically sent by the sys- eters are high for full broken rotor case. The RMSE, MAPE
tem including the decision and the corrective action. The and MAE are less for the healthy rotor. These parameters
gateway carrier sent the generated decision to the machine go worse with the condition of rotor. So these parameters
controller. The Drop Box file stores the data for the im- are beneficial to check the condition of rotor. The quality
410 | Zhe Mi, Tiangang Wang, Zan Sun, and Rajeev Kumar, Vibration signal diagnosis and analysis

Table 3: Different parameters obtained for four different scenarios by the proposed method

Different Scenarios Proposed Method


RMSE MAPE (%) MAE
Healthy condition 0.010 1.55 0.112
Bearing failure 0.016 2.18 0.120
Half-broken rotor 0.018 2.70 0.129
Full-broken rotor bar 0.021 3.35 0.134

Figure 11: Parameters obtained for four different scenarios Figure 12: Percentage improvement of the proposed technique over
existing techniques

parameters like SNR and SSIM are also checked to check


the performance of the technique as shown in Table 4. Neurofuzzy Inference System and the Single Modeling Sys-
Higher the value of the PSNR, higher will be the qual- tem by 8% and 60%, respectively. The presented technique
ity, the healthy condition rotor have the high PSNR value is better than the existing Adaptive Neurofuzzy Inference
as compared to the bearing failure, half-broken rotor and System and the Single Modeling techniques by average of
the full-broken rotor bar. The value of SSIM is in between 15% and 30%, respectively.
the 0 and 1. If the value approaches to 1 shows the high Applications: Development of fault diagnosis and prog-
quality. nosis strategies that achieved extensive utility in a wide
range of application domains in recent years. Typically,
there are two major classes: Model based and data-driven.
4.3 Comparison with other techniques A model-based technique depends on dynamic system
model accuracy. It uses actual technique for generation
The results obtained by the presented technique for the of the difference between the two outputs, indicative of a
four different scenarios are also compared with the exist- fault condition.
ing techniques for the validation purpose. Comparison of The anticipated fault condition is addressed by the
the presented technique is done on the basis of RMSE, data-driven techniques where a fault model collects the
MAPE and MAE as tabulated in Table 5. constructs like neural networks which must be trained first
From the table, it is clear that the minimum error is with prototype fault patterns.
obtained by the proposed technique as compared to other
existing techniques. The percentage improvement of the
proposed technique is also calculated and shown in Table 5 Conclusion
6.
The graphical representation is also shown in Figure The diagnosis of machine performance is done by build-
12 for better analysis and visualization. It is obtained that ing the Maintenance-CPS platform and unlabelled data is
the proposed technique is around 25% and 90% better explored. The expert’s expensive onsite visit, reduction
than the Adaptive Neurofuzzy Inference System and the of training cost and the non-zero breakdown is achieved
Single Modeling System respectively in terms of RMSE. by the continuous monitoring; these are the merits of the
If one see the performance in terms of MAPE, then proposed system. The proposed layers of cyber-physical
the proposed technique outperforms the existing Adaptive
Zhe Mi, Tiangang Wang, Zan Sun, and Rajeev Kumar, Vibration signal diagnosis and analysis | 411

Table 4: PSNR and SSIM obtained for four different scenarios by the proposed method

Different Scenarios Proposed Method


PSNR (dB) SSIM
Healthy condition 56 0.92
Bearing failure 49 0.88
Half-broken rotor 40 0.79
Full-broken rotor bar 34 0.71

Table 5: Comparison of the presented technique over existing techniques

Adaptive Neurofuzzy Single Modeling


Different Scenarios Proposed Method
Inference System [38] System [38]
MAPE MAPE MAPE
RMSE MAE RMSE MAE RMSE MAE
(%) (%) (%)
Healthy condition 0.013 1.69 0.135 0.203 3.45 0.129 0.010 1.55 0.112
Bearing failure 0.020 2.27 0.125 0.275 6.77 0.265 0.016 2.18 0.120
Half-broken rotor 0.022 2.82 0.133 0.478 6.92 0.184 0.018 2.70 0.129
Full-broken rotor bar 0.025 3.44 0.139 0.394 8.93 0.192 0.021 3.35 0.134

Table 6: Percentage improvement of the proposed technique over existing techniques

Different Scenarios Percentage Improvement of the Proposed Technique


In terms of RMSE over In terms of MAPE over In terms of MAE over
Adaptive Adaptive
Single Single Adaptive Neuro Single
Neuro-fuzzy Neuro-fuzzy
Modeling Modeling fuzzy Inference Modeling
Inference Inference
System System System System
System System

Healthy condition 30.00 % 95.07% 9.03% 55.07% 20.54% 13.18%


Bearing failure 25.00% 94.18% 4.13% 67.80% 4.17% 54.72%
Half-broken rotor 22.22% 96.23% 4.44% 60.98% 3.10% 29.89%
Full-broken rotor bar 19.05% 94.67% 2.69% 62.49% 3.73% 30.21%

system are connected to the Cloud technologies. The Funding information: The authors state no funding
different layers of the system are sensor, analysis, and involved.
decision making. Based on an interpretable method deal-
ing with data, decision making and analysis is done. The Author contributions: All authors have accepted re-
rolling element bearing are used to test the presented sys- sponsibility for the entire content of this manuscript and
tem. Finally, the cloud technologies and the future fault approved its submission.
prognosis systems maintenance field are discussed. The
performances of the models are evaluated by utilizing the Conflict of interest: The authors state no conflict of inter-
classical statistical metrics such as RMSE Mean Absolute est.
Error (MAE) and Mean Absolute Percentage Error (MAPE).
It is obtained that the proposed technique is around 25%
and 90% better than the Adaptive Neurofuzzy Inference
System and the Single Modeling System respectively
References
in terms of RMSE. Higher the PSNR, higher will be the [1] Wu D, Liu S, Zhang L, Terpenny J, Gao RX, Kurfess T, et al. A
quality, the healthy condition rotor have the high PSNR fog computing-based framework for process monitoring and
value as compared to the bearing failure, half-broken rotor prognosis in cyber-manufacturing. J Manuf Syst. 2017;43:25–
and the full-broken rotor bar. The future work will focus 34.
on the development of an automatic anomaly detection in [2] Qian G, Lu S, Pan D, Tang H, Liu Y, Wang Q. Edge computing:
A promising framework for real-time fault diagnosis and dy-
the gateway.
namic control of rotating machines using multi-sensor data.
IEEE Sens J. 2019;19(11):4211–20.
[3] Poongodi M, Sharma A, Vijayakumar V, Bhardwaj V, Sharma
AP, Iqbal R, et al. Prediction of the price of Ethereum
412 | Zhe Mi, Tiangang Wang, Zan Sun, and Rajeev Kumar, Vibration signal diagnosis and analysis

blockchain cryptocurrency in an industrial finance system. [18] Li Z, Wang Y, Wang KS. Intelligent predictive maintenance for
Comput Electr Eng. 2020;81:106527. fault diagnosis and prognosis in machine centers: Industry 4.0
[4] Lu S, Zhou P, Wang X, Liu Y, Liu F, Zhao J. Condition monitoring scenario. Adv Manuf. 2017;5(4):377–87.
and fault diagnosis of motor bearings using undersampled [19] Shyamala D, Swathi D, Prasanna JL, Ajitha A. IoT platform for
vibration signals from a wireless sensor network. J Sound condition monitoring of industrial motors. 2017 2nd
Vibrat. 2018;414:81–96. International Conference on Communication and Electronics
[5] Sharma A, Kumar R. Service-level agreement—energy co- Systems (ICCES); 2017 Oct. 19-20; Coimbatore, India. IEE;
operative quickest ambulance routing for critical healthcare 2017 p.
services. Arab J Sci Eng. 2019;44(4):3831–48. 260-265. https://doi.org/10.1109/CESYS.2017.8321278.
[6] Al Sunny SN, Liu X, Shahriar MR. Remote monitoring and [20] Van Tu ng T, Yang BS. Machine fault diagnosis and prognosis:
online testing of machine tools for fault diagnosis and the state of the art. Int J Fluid Mach Syst. 2009;2(1):61–71.
maintenance using MTComm in a cyber-physical [21] Firmansah A, Mufti N, Affandi AN, Zaeni IA. Self-powered
manufacturing cloud. 2018 IEEE 11th International Conference IoT Based Vibration Monitoring of Induction Motor for Diag-
on Cloud Computing (CLOUD); 2018 Jul 2-7; San Francisco nostic and Prediction Failure IOP Conf Series Mater Sci Eng.
USA. 2018 p. 532-539. 2019;588(1):012016.
IEEE. https://doi.org/10.1109/CLOUD.2018.00074. [22] Chang CY, Chang EC, Huang CW. In Situ Diagnosis of Industrial
[7] Deng W, Zhang S, Zhao H, Yang X. A novel fault diagnosis Motors by Using Vision-Based Smart Sensing Technology.
method based on integrating empirical wavelet transform and Sensors. 2019;19(24):5340.
fuzzy entropy for motor bearing. IEEE Access. 2018;6:35042– [23] Izagirre U, Andonegui I, Eciolaza L, Zurutuza U. Towards
56. manufacturing robotics accuracy degradation assessment:
[8] Medoued A, Mordjaoui M, Soufi Y, Sayad D. Induction ma- A vision-based data-driven implementation. Robot Comput-
chine bearing fault diagnosis based on the axial vibration Integr Manuf. 2021;67:102029.
analytic signal. Int J Hydrogen Energy. 2016;41(29):12688– [24] Ilonen J, Kamarainen JK, Lindh T, Ahola J, Kalviainen H, Par-
95. tanen J. Diagnosis tool for motor condition monitoring. IEEE
[9] Huo Z, Mukherjee M, Shu L, Chen Y, Zhou Z. Cloud-based Trans Ind Appl. 2005;41(4):963–71.
data-intensive framework towards fault diagnosis in large- [25] Sharma A, Kumar R. A framework for pre-computated multi-
scale petrochemical plants. 2016 International Wireless Com- constrained quickest QoS path algorithm. J Telecommun Elec-
munications and Mobile Computing Conference (IWCMC). 2016 tron Comput Eng. 2017;9(3-6):73–7.
Sept 5-9. Paphos, Cyprus; 2016 p. 1080-1085. IEEE. [26] Lu S, Zhou P, Wang X, Liu Y, Liu F, Zhao J. Condition monitoring
[10] Li B, Chow MY, Tipsuwan Y, Hung JC. Neural-network-based and fault diagnosis of motor bearings using undersampled
motor rolling bearing fault diagnosis. IEEE Trans Ind Electron. vibration signals from a wireless sensor network. J Sound
2000;47(5):1060–9. Vibrat. 2018;414:81–96.
[11] Shao S, Sun W, Wang P, Gao RX, Yan R. (2016, August). Learn- [27] Wang J, Fu P, Zhang L, Gao RX, Zhao R. Multilevel information
ing features from vibration signals for induction motor fault fusion for induction motor fault diagnosis. IEEE/ASME Trans
diagnosis. 2016 International Symposium on Flexible Automa- Mechatron. 2019;24(5):2139–50.
tion (ISFA); 2016 Aug 1-3; Cleveland, USA. IEE; 2016 p. 71-76. [28] Shifat TA, Hur JW. An effective stator fault diagnosis frame-
[12] Iorgulescu M, Beloiu R. (2012, October). Study of DC motor work of BLDC motor based on vibration and current signals.
diagnosis based on the vibration spectrum and current analy- IEEE Access. 2020;8:106968–81.
sis. 2012 International Conference on Applied and Theoretical [29] Alwodai A, Gu F, Ball AD. A comparison of different techniques
Electricity (ICATE); 2012 Oct 25-27; Craiova, Romania. IEE; for induction motor rotor fault diagnosis. J Phys Conf Ser. 2012
2012 p. 1-4. May;364(1):012066.
[13] Chen HX, Chua PS, Lim GH. Adaptive wavelet transform [30] Liu HC, Lin QL, Ren ML. Fault diagnosis and cause analysis us-
for vibration signal modeling and application in fault diag- ing fuzzy evidential reasoning approach and dynamic adaptive
nosis of water hydraulic motor. Mech Syst Signal Process. fuzzy Petri nets. Comput Ind Eng. 2013;66(4):899–908.
2006;20(8):2022–45. [31] Gritli Y, Di Tommaso AO, Miceli R, Rossi C, Filippetti F. Diagno-
[14] Magadán L, Suárez FJ, Granda JC, García DF. Low-cost real- sis of mechanical unbalance for double cage induction motor
time monitoring of electric motors for the Industry 4.0. Proce- load in time-varying conditions based on motor vibration sig-
dia Manuf. 2020;42:393–8. nature analysis. 2013 International Conference on Renewable
[15] Pinheiro AA, Brandao IM, Da Costa C. Vibration Analysis in Energy Research and Applications (ICRERA); 2013 Oct 20-23;
Turbomachines Using Machine Learning Techniques. Eu- Madrid, Spain. IEE; 2013 p. 1157-1162.
ropean Journal of Engineering and Technology Research. [32] Finley WR, Howdowanec MM, Holter WG. Diagnosing motor vi-
2019;4(2):12–6. bration problems. Conference Record of 2000 Annual Pulp and
[16] Sakthivel NR, Sugumaran V, Babudevasenapati S. Vibration Paper Industry Technical Conference (Cat. No. 00CH37111);
based fault diagnosis of monoblock centrifugal pump using 2000 Jun 19-23; Atlanta, USA. IEE; 2000 p. 165-180.
decision tree. Expert Syst Appl. 2010;37(6):4040–9. [33] Iorgulescu M, Beloiu R, Popescu MO. (2010, May). Vibration
[17] Firmansah A, Mufti N, Affandi AN, Zaeni IA. Self-powered IoT monitoring for diagnosis of electrical equipment’s faults. 2010
Based Vibration Monitoring of Induction Motor for Diagnostic 12th International Conference on Optimization of Electrical and
and Prediction Failure. IOP Conf Series Mater Sci Eng. 2019 Electronic Equipment; 2010 May 20-22; Brasov, Romania. IEE;
Aug;588(1):012016. 2010 p. 493-499.
IEEE. https://doi.org/10.1109/OPTIM.2010.5510332.
Zhe Mi, Tiangang Wang, Zan Sun, and Rajeev Kumar, Vibration signal diagnosis and analysis | 413

[34] Gangsa r P, Tiwari R. Comparative investigation of vibra-


tion and current monitoring for prediction of mechanical
and electrical faults in induction motor based on multiclass-
support vector machine algorithms. Mech Syst Signal Process.
2017;94:464–81.
[35] Sharma A, Kumar R. Computation of the reliable and quick-
est data path for healthcare services by using service-
level agreements and energy constraints. Arab J Sci Eng.
2019;44(11):9087–104.
[36] Gupta V, Kaur BP, Jangra S. An efficient method for fault toler-
ance in cloud environment using encryption and classification.
Soft Comput. 2019;23(24):13591–602.
[37] Kannan R, Manohar SS, Kumaran MS. IoT-Based Condition
Monitoring and Fault Detection for Induction Motor. Pro-
ceedings of 2nd International Conference on Communication,
Computing and Networking; 2018 Sept 8; Chandigarh, India.
Springer, Singapore; 2018 p. 205-215.
[38] Zurita-Millán D, Delgado-Prieto M, Saucedo-Dorantes JJ,
Cariño-Corrales JA, Osornio-Rios RA, Ortega-Redondo JA, et al.
Vibration signal forecasting on rotating machinery by means
of signal decomposition and neurofuzzy modeling. Shock Vib.
2016;2016.

You might also like