Failure Log Analytics for Reducing Electrical Machine Downtime using
Deep Learning
SUMIT KUMAR1, RAKESH RANJAN2, BHUPATI3, HARISH DUTT SHARMA4,
YOGESH MISRA5
1Department of Artificial Intelligence & Machine Learning,
COER University,
Roorkee,
INDIA
2Department of Information Technology,
ABES Engineering College,
Ghaziabad, Uttar Pradesh-201009,
INDIA
3Department of IoT,
K. L. Deemed to be University,
Vaddeswaram, Guntur-522302,
INDIA
4USCS Department,
Uttaranchal University,
Dehradun, Uttarakhand,
INDIA
5Department of Electronics & Communication Engineering,
GMR Institute of Technology,
Rajam, Andhra Pradesh,
INDIA
Abstract: - Electrical machine downtime reduces productivity across various operation times that are addressed
using stored data logs in the controller. Analyzing such logs is useful in preventing/ reducing machine
downtime through precise controller options. This article proposes a Downtime Reduction-focused Log
Analytical Model (DR-LAM) for improving the machine operation time by reducing operation failures. In this
model, deep learning is employed for differentiating the production-less electrical cycles in correlation with the
previous output. This differentiation is conditional using run-time failures and failed operation cycles.
Therefore the logs are analyzed based on the above differentiations for precise problem identification. The
training for the deep learning network is provided using previous differentiated cycle logs improving the
detection ratio.
Key-Words: - Deep Learning, Downtime, Electrical Machines, Log Analysis, operation time, orun-time failure,
peration cycle.
Received: September 11, 2022. Revised: October 14, 2023. Accepted: November 6, 2023. Published: December 7, 2023.
1 Introduction
Electrical machine downtime detection is a process
that detects the process that is stopped due to
unplanned events in a machine. The machine
downtime detection detects the exact downtime
cause and reasons by identifying the cell of the
machine, [1], [2]. The predictive maintenance
(PdM) tool-based detection method is used for
machine downtime detection. PdM is used as an
early machine downtime fault detection which
reduces the latency in performing tasks. The PdM
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.46
Sumit Kumar, Rakesh Ranjan,
Bhupati, Harish Dutt Sharma, Yogesh Misra
E-ISSN: 2224-3402
444
Volume 20, 2023
method uses features that contain faults in electrical
machines, [3], [4].
An electrical machine failure log analysis is a
process that provides visible information for further
processes. The log analysis uses pattern recognition
and classification methods to identify the important
logs in electrical machines, [5]. The exact failures
that occur in electrical machines are detected based
on the information that is collected from machine
storage. The accurate failure log search ratio is
identified which reduces the complexity level in
electrical machines, [6], [7].
Log data analytics is a process that reviews the
logs and identifies the threats which are presented in
the log, [8]. A Log data analytics-based method is
used to reduce the overall electrical machine
downtime. Log data analytics ensures compliance
and reviews the issues to solve the problems, [9].
The log data analytics-based method protects the
downtime level of machines which improves the
accuracy level in performing tasks, [10].
2 Related Works
The study, [11], proposed a machine learning (ML)
evaluation method for maintenance records in
photovoltaic (PV) inverters. The ML method
analyzes the datasets and produces leverages to the
evaluation process. The failure frequencies and
circuits are also analyzed to reduce the
computational cost ratio of inverters. The actual
inverter-related records are identified which
minimizes the energy consumption in the
computation process. The proposed method reduces
the failure ratio in maintenance which enhances the
performance range of PV inverters.
The study, [12], introduced a smart condition-
monitoring strategy using wireless accelerometer
sensor modules. The main aim of the strategy is to
maintain the records and operations that are
presented in smart devices. Data-driven capabilities
and the Internet of Things (IoT) are used in
condition monitoring systems. The data-driven
capabilities improve the cost-effective range of
heavy machinery. The introduced strategy increases
the feasibility level of the devices.
The study, [13], designed a principal component
analysis (PCA) approach for fault detection. The
PCA approach uses an analysis method that
analyzes the exact health conditions and scenarios of
the patients. The analyzed data produce optimal
information for issues and faults detection in the
condition monitoring system. Both vibrational and
electrical signatures are used here to predict the
issues in the systems. The designed PCA approach
improves the accuracy of the fault detection process.
The study, [14], proposed a new analysis
methodology for failure prediction in automotive
industries. Electrical terminals are used to detect the
root and cause of the problems in industries. The
actual failure mechanisms, characteristics, and
features are detected based on the analyzed data.
The proposed methodology improves the energy
efficiency range of the systems. The proposed
method provides quantifiable services and functions
to automotive industries.
The study, [15], introduced a support vector
machine (SVM) based failure diagnosis method for
PV generators. The actual goal of the method is to
detect the failures which are occurred in the
generators. The introduced method identifies the
issues based on operational and functional data. The
normal and abnormal failures are classified based on
the severity of the issues. The introduced method
increases the accuracy of failure diagnosis which
enhances the performance level of PV generators.
The study, [16], developed a multi-class SVM
classifier, particle swarm optimization (PSO)
algorithm-based fault diagnosis method for
electrical machines. The developed method also
uses a gravity search algorithm (GSA) to diagnose
the faults in machines. The SVM classifier is mainly
used here to classify the types of faults based on the
condition of the machines. The developed method
reduces the faults which improves the performance
range of electrical machines.
The study, [17], proposed a new combined
failure severity analysis method for rotating
machines. The proposed method is used as a
machine failure prediction that predicts the exact
issues in rotating machines. A supervisory system is
implemented to monitor the machine and identify
inappropriate issues in the machines. The
supervisory system sends an alert message to alert
the devices via smart signals. The proposed method
enhances the feasibility and significance range of
the machines.
3 Proposed Analytical Model
The proposed DR-LA model is designed to improve
the electrical machine operation time for precise
problem identification. The stored data logs analysis
in the controller in different operation times is
addressed for reducing electrical machine
downtime. The objective of this model is to analyze
such logs to prevent downtime using precise
controller options in different time intervals.
Addressed electrical machine downtime reduces
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.46
Sumit Kumar, Rakesh Ranjan,
Bhupati, Harish Dutt Sharma, Yogesh Misra
E-ISSN: 2224-3402
445
Volume 20, 2023
productivity in different time intervals using stored
data logs for reducing the adverse impact occurrence
in electrical machines.
Fig. 1: Proposed Model Illustration
In this model, deep learning is used for
differentiating run-time failure and operation failure
detected at the time of processing the electrical
machine. This learning is employed for
differentiating the production-less electrical cycles
in correlation with the previous output for downtime
failure classification. The electrical machine
operations are observed across various networks and
time intervals for analyzing such logs through the
proposed model and deep learning paradigm for
preventing machine downtime. The proposed model
is illustrated in Figure 1.
The failure differentiation is pursued to improve
the machine operation time for processing the stored
data logs in the controller. Based on the above
differentiation, precise problem detection is
achieved. These run-time and operation failures are
differentiated to ensure support for the electrical
machine users. In this proposed model, the training
is provided for the deep learning network using
previous differentiated cycle logs for better
operation and improving detection ratio. The
operation time is computed to identify and
differentiate the run time and operation failure from
the controller at different time intervals. In this
article, the stored data logs are analyzed for training
the deep learning network using DR-LLAM to
reduce operation time and failures in the controller.
4 Stored Data Log Analysis
The data log analysis is pursued by
differentiating run-time failures and operation
failures through a deep-learning network for
reducing operation time. The input data logs can be
stored and retrieved for further processing. In this
scenario, the operation cycles can adapt their nature
based on the services. Therefore, the first data log
analysis is represented as
󰇛󰇜󰇛󰇜 (1)
where,


(2)
The above equation (1) and (2) validates the
overall electrical machine operations based on data
logs analysis using the proposed model irrespective
of the machine downtime  in the controller
is identified for rectifying the problems. Where,
used to denote the number of production-less
electrical cycles in the network. These production-
less electrical cycles are addressed in the controller
leading to run-time failures and operation failures.
The variables  and  indicates run-time
and operation time are computed at the time of log
analysis for improving the machine operation time
by reducing operation failures. In this analysis, the
machine run-time and operation time differ.
Therefore, the downtime failure classification
should be optimized to perform log analysis. The
proposed model verifies the available resources and
their services rely on the above differentiation. The
downtime failure classification  based on log
analysis output is computed as follows:
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.46
Sumit Kumar, Rakesh Ranjan,
Bhupati, Harish Dutt Sharma, Yogesh Misra
E-ISSN: 2224-3402
446
Volume 20, 2023
 
 (3)
Fig. 2: Log Analysis Process
In equation (3), the different operations
performed by the electrical machines reduce human
work and ease to completion of the work. If the
variable  represents the number of active
operation cycles in electrical machines. Based on
the log analysis, the controllers are regulated by the
machines for processing multiple operations in that
network. The downtime failure classification is
pursued using deep learning and correlates with
previous output for reducing operation failures.
Hence, this proposed model differentiates the run-
time failures and operation failures for accurate
problem detection. Using this deep learning, the
time interval addressed from the machine downtime
is balanced for reducing failure occurrence in the
deep learning network. The log analysis process is
illustrated in Figure 2.
The operation interval generates logs based on
machine operations. The run time and downtime
analysis are performed to validate failures. The
classification using is valid under the above data
logs for correlation. In this correlation, the previous
logs are used for verifying the failure. Using this
deep learning, the proposed model is used for
processing multiple operations in less processing
time (Figure 2). Based on deep learning, the
production-less electrical cycles are identified. After
identifying this problem, the learning helps to
differentiate run-time failure and operation failure in
any operation instance. The operation time is
computed based on the service requirements of users
to reduce failures; reliable operations are made by
the electrical machine. Later, the stored data logs are
analyzed priority-wise to reduce production-less
cycles in that network. Therefore, maximum
productivity is achieved with less run time and less
operation time is the successful output. The
correlation of current differentiated cycles with
previous differentiated cycles  for the problem,
identification is expressed as:



(4)
In equation (4), precise problem identification is
achieved from the previous operation output and
downtime failure classification based on minimum
and maximum machine downtime. This condition
differentiates the minimum and maximum failures at
different time intervals. The identified operation
failures from the controller used for differentiating
production-less electrical cycles in correlation with
the previous output. The failure data log analysis for
reducing electrical machine downtime using deep
learning in different time intervals. In this condition,
the run-time and operation failure are identified and
differentiated for precise problem identification.
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.46
Sumit Kumar, Rakesh Ranjan,
Bhupati, Harish Dutt Sharma, Yogesh Misra
E-ISSN: 2224-3402
447
Volume 20, 2023
Hence, the logs are analyzed based on the above
differentiations at any downtime time interval.
Therefore, the final log analysis for the available
controller  󰇛󰇜 is computed as:
 󰇛󰇜󰇛󰇜

 (5)
Fig. 3: Learning Process Illustration
In equation (5), the final log analytics is
performed to reduce operation time and failures in
different time intervals. The electrical machine
downtime is identified for improving machine
operation time based on the differentiation for time
intervals. This proposed model is used to ensure the
seamless support of the electrical machine is
improved. Therefore, precise problem identification
and production-less cycle differentiation are
performed sequentially to improve operation time
by reducing failure occurrence. The learning process
is illustrated in Figure 3.
The  is analyzed for all in detecting.
The output is classified using deep learning as
or. The  represents a classification
failure and  denotes the classification. This
second classification is validated recurrently for
new (refer to Figure 3). The minimum and
maximum possibility of run-time failure and
operation failure are differentiated for accurate
problem identification. Based on the occurrence of
the failure and downtime, the training provided for
the learning process is provided using previous
differentiated cycle logs for reducing operation time.
Now, the previous differentiated cycle logs are
represented as:

󰇟󰇛󰇜󰇠 (6)

󰇣 󰇛󰇜
 󰇛󰇜

󰇤 (7)
As per equations (6) and (7), the downtime
failure classification is pursued for reducing
operation failures in different time intervals. The
logs are sequentially analyzed based on
differentiation output. This proposed model and
deep learning are used to improve the operation of
electrical machines. Based on the data provided in
[18], the log analysis is performed for and  as
in Figure 4. The variation in is used for the
analysis.
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.46
Sumit Kumar, Rakesh Ranjan,
Bhupati, Harish Dutt Sharma, Yogesh Misra
E-ISSN: 2224-3402
448
Volume 20, 2023
Fig. 4: and Analysis
The  varies accordingly for multiple operation
hours of the electrical machine. The lag analysis
for and  are validated for the event's
accuracy. The event of downtime/ failure is
identified throughout the operation hours for
improving the  without. Therefore
consecutive  variations are handled using
successful  preventing maximum downtime.
5 Performance Assessment
The performance assessment is validated using the
metrics detection ratio, analysis rate, operation
failure, and analysis time. The operation time is
considered from 1 to 18 hours, with the existing
methods H-GSAPSO, [16], and DF-PCA, [13].
6 Detection Ratio
Fig. 5: Detection Ratio
The high electrical machine downtime is reduced
through deep learning and the proposed model based
on log analytics. Using the proposed model, deep
learning is employed for differentiating the
production-less electrical cycles for problem
detection. The proposed model is used for
differentiating the run-time failures and failed
operation cycles in different time intervals to
achieve high downtime detection as presented in
Figure 5.
7 Analysis Rate
Fig. 6: Analysis Rate
This proposed DR-LAM achieves a high analysis
rate for computing the stored data based on previous
operation output and downtime failure classification
for reducing the operation failure occurrence (Refer
to Figure 6). The production-less electrical cycles in
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.46
Sumit Kumar, Rakesh Ranjan,
Bhupati, Harish Dutt Sharma, Yogesh Misra
E-ISSN: 2224-3402
449
Volume 20, 2023
an electrical machine are reduced using deep
learning. Therefore, regardless of consistent
operation is achieving a high analysis rate.
8 Operation Failure
Fig. 7: Operation Failure
This proposed model achieves high operation
failure compared to the other factors as represented
in Figure 7. The operation failures are identified to
ensure the support of electrical machine processing.
Therefore, the proposed model improving the
detection ratio using previous differentiated cycle
logs for training the deep learning is provided to
achieve less operation failure.
9 Analysis Time
Fig. 8: Analysis Time
This proposed model used to achieve high data
log analysis time based on the above differentiation
is pursued by precise problem identification. The
downtime failure classification is verified with
previous operation output for differentiating the
production-less operation cycle outputs in less
failures and analysis time is represented in Figure 8.
If the operation failure is high in this analysis, the
downtime is reduced. Deep learning is used to
reduce operation failures and reduces analysis time.
From the discussion above the comparative analysis
summary is presented in Table 1. The analysis of the
proposed system is compared with the help of
performance metrics to find the detection ratio,
analysis rate, failure, and analysis time. The
proposed DR-LAM detects the highest detection
ratio of 64.15% when compared to other approaches
H-GSAPSO and DF-PCA. The analysis rate of DR-
LAM achieves a higher rate of 91%; the proposed
model DR-LAM reduces the failure rate and also
reduces the analysis time when compared to other
approaches.
Table 1. Comparative Analysis Summary
Metrics
H-
GSAPSO
DF-
PCA
Detection Ratio
41.4
53.4
Analysis Rate
(Logs/Hr)
45
72
Failure (%)
17.1
12.4
Analysis Time (s)
16.2
11.6
The proposed model is found to improve the
detection ratio and analysis rate by 8.38% and
11.9% respectively. This model reduces the failure
and analysis time by 7.87% and 8.12% respectively.
10 Conclusion
This article introduced a downtime reduction-
focused log analytical model for recommending
better machine operation times. The machine
operation times are validated for the different failure
logs analyzed over the different operation intervals.
The failures between successive operation intervals
are reduced by validating less or no-output intervals.
Based on the active operation cycles the log is
correlated for providing multiple recommendations
for mitigating the maximum downtime. The log
analysis is repeated until optimal recommendations
are provided for different cycles under run and
operation times. From the comparative analysis, it is
seen that the proposed model reduces failure by
7.87% whereas it increases the detection ratio by
8.38% for different operation cycles. The proposed
work is planned to integrate self-analytical operation
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.46
Sumit Kumar, Rakesh Ranjan,
Bhupati, Harish Dutt Sharma, Yogesh Misra
E-ISSN: 2224-3402
450
Volume 20, 2023
modules for prediction-based operation control. This
is required to prevent machine detentions in timer-
less work allocations.
The algorithm described in this research is a
method towards a reliable and versatile method for
electrical machine failure prediction, which may be
applied to a variety of faults. This research suggests
an innovative Deep-learning method for electrical
machine preventive maintenance. As a novel idea,
this one may require further development and
investigation. This method reduces electrical
downtime by using the current characteristic
variances that result from failure log analysis in
electric devices. Future studies will also involve
applying this technique to more complex defects and
larger data samples.
References:
[1] Sarma, N., Tuohy, P. M., Özgönenel, O., &
Djurović, S. (2023). Early life failure modes
and downtime analysis of onshore type-III
wind turbines in Turkey. Electric Power
Systems Research, vol.216, 108956.
[2] Mohan, T. R., Roselyn, J. P., Uthra, R. A.,
Devaraj, D., & Umachandran, K. (2021).
Intelligent machine learning based total
productive maintenance approach for
achieving zero downtime in industrial
machinery. Computers & Industrial
Engineering, vol.157, 107267.
[3] Desai, V., & Singh, S. (2023). Assessment of
Straight-Line Drawing Machine and
Workforce Utilization Using Time Study,
Work Sampling and Downtime Analysis in
Steel Industry. Transactions of the Indian
National Academy of Engineering, vol.8(1),
pp.99-111.
[4] Bakdi, A., Kristensen, N. B., & Stakkeland,
M. (2022). Multiple instance learning with
random forest for event logs analysis and
predictive maintenance in ship electric
propulsion system. IEEE Transactions on
Industrial Informatics, vol.18(11), pp.7718-
7728.
[5] Vallachira, S., Orkisz, M., Norrlöf, M., &
Butail, S. (2019). Data-driven gearbox failure
detection in industrial robots. IEEE
Transactions on Industrial Informatics,
vol.16(1), pp.193-201.
[6] Milić, S. D., Đurović, Ž., & Stojanović, M. D.
(2023). Data science and machine learning in
the IIoT concepts of power plants.
International Journal of Electrical Power &
Energy Systems, vol.145, 108711.
[7] Chen, X., Xu, W., Liu, Y., & Islam, M. R.
(2019). Bearing corrosion failure diagnosis of
doubly fed induction generator in wind
turbines based on stator current analysis.
IEEE Transactions on Industrial Electronics,
vol.67(5), pp.3419-3430.
[8] He, L., Hao, L., & Qiao, W. (2021). Remote
monitoring and diagnostics of pitch-bearing
defects in an MW-scale wind turbine using
pitch symmetrical-component analysis. IEEE
Transactions on Industry Applications,
vol.57(4), pp.3252-3261.
[9] Wang, H., Li, Y. F., & Ren, J. (2023).
Machine learning for fault diagnosis of high-
speed train traction systems: A review.
Frontiers of Engineering Management,
https://doi.org/10.1007/s42524-023-0256-2.
[10] Araste, Z., Sadighi, A., & Jamimoghaddam,
M. (2022). Fault Diagnosis of a Centrifugal
Pump Using Electrical Signature Analysis and
Support Vector Machine. Journal of Vibration
Engineering & Technologies, vol.11,
pp.2057–2067,
https://doi.org/10.1007/s42417-022-00687-6.
[11] Gunda, T., Hackett, S., Kraus, L., Downs, C.,
Jones, R., McNalley, C., ... & Walker, A.
(2020). A machine learning evaluation of
maintenance records for common failure
modes in PV inverters. IEEE Access, vol.8,
pp.211610-211620.
[12] Aqueveque, P., Radrigan, L., Pastene, F.,
Morales, A. S., & Guerra, E. (2021). Data-
driven condition monitoring of mining mobile
machinery in non-stationary operations using
wireless accelerometer sensor modules. IEEE
Access, vol.9, pp.17365-17381.
[13] Elsamanty, M., Ibrahim, A., & Salman, W. S.
(2023). Principal component analysis
approach for detecting faults in rotary
machines based on vibrational and electrical
fused data. Mechanical Systems and Signal
Processing, vol.200, 110559.
[14] Silva, F. J. G., Morgado, L., Teixeira, A., Sá,
J. C., Ferreira, L. P., & de Almeida, F. (2020).
Analysis and Development of a Failure
Prediction Model for Electrical Terminals
Used in the Automotive Industry. Procedia
Manufacturing, vol.51, pp.207-214.
[15] Cho, K. H., Jo, H. C., Kim, E. S., Park, H. A.,
& Park, J. H. (2020). Failure diagnosis
method of photovoltaic generator using
support vector machine. Journal of Electrical
Engineering & Technology, vol.15, pp.1669-
1680.
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.46
Sumit Kumar, Rakesh Ranjan,
Bhupati, Harish Dutt Sharma, Yogesh Misra
E-ISSN: 2224-3402
451
Volume 20, 2023
[16] Ding, S., Hao, M., Cui, Z., Wang, Y., Hang,
J., & Li, X. (2023). Application of multi-SVM
classifier and hybrid GSAPSO algorithm for
fault diagnosis of electrical machine drive
system. ISA transactions, vol.133, pp.529-
538.
[17] de Martins, D. H. C., Viana, D. P., de
Lima, A. A., Pinto, M. F., Tarrataca, L.,
Lopes e Silva, F., ... & Haddad, D. B. (2021).
Diagnostic and severity analysis of combined
failures composed by imbalance and
misalignment in rotating machines. The
International Journal of Advanced
Manufacturing Technology, vol.114(9-10),
pp.3077-3092.
[18] Erisen, B. (2018). Wind Turbine Scada
Dataset, Kaggle, [Online].
https://www.kaggle.com/datasets/berkerisen/
wind-turbine-scada-dataset (Accessed Date:
November 11, 2023).
Contribution of Individual Authors to the
Creation of a Scientific Article (Ghostwriting
Policy)
The authors equally contributed in the present
research, at all stages from the formulation of the
problem to the final findings and solution.
Sources of Funding for Research Presented in a
Scientific Article or Scientific Article Itself
No funding was received for conducting this study.
Conflict of Interest
The authors have no conflicts of interest to declare.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.en
_US
WSEAS TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS
DOI: 10.37394/23209.2023.20.46
Sumit Kumar, Rakesh Ranjan,
Bhupati, Harish Dutt Sharma, Yogesh Misra
E-ISSN: 2224-3402
452
Volume 20, 2023