Predictive Maintenance of Induction Motor in the Cutting Section of
Paper Industry
KALAVATHIDEVI T, MADHAN MOHAN M, BALUPRTHVIRAJ K N,
SANGAVI B, RAJARAGHAVENDRAA S K, RAJESHKANNA K
Electronics and Instrumentation Engineering
Kongu Engineering College
Perundurai, INDIA
Abstract: - Induction Motors are vital to the paper industry because they power a variety of equipment needed
for pulp processing, drying and cutting among other tasks. Ensuring the consistent manufacture of these motors
and satisfying market demands depends heavily on their dependability. However, the extreme humidity, dust
and fluctuating loads that occur in the paper sector present serious obstacles to the efficiency and durability of
induction motors. To maintain continuous production and satisfy consumer demand, the paper industry
significantly depends on the effective operation of a variety of machinery including induction motors in the
cutting area. Unexpected malfunctions in these vital parts however might result in expensive downtime and lost
output. By using data-driven strategies that anticipate and avoid breakdowns before they happen, predictive
maintenance techniques provide a proactive way to reduce such risks. This report provides an extensive
analysis of the application of predictive maintenance techniques designed especially for induction motors in the
paper industry's cutting division. The predictive maintenance techniques includes Random Forest Tree, Linear
Regression and Support Vector Machines algorithm with these algorithm the Induction Motor’s variations in
temperature, vibration, sound, and speed parameters were collected and trained for predicting the failure.
Among these algorithms Support Vector Machines shows greater advantage in predicting the failure with the
accuracy of 84% whereas Random Forest Tree with 75% and Linear regression with 72%. As well as the real
time data’s were collected and stored in the database. The performance of the Induction Motor were efficiently
improved and monitored under normal and fault condition by Machine Learning techniques.
Key-Words: -Induction Motors, Sensing Data, Downtime, Motor Failure Prediction, Random Forest Tree,
Linear Regression, Support Vector Machines
Received: March 6, 2024. Revised: August 24, 2024. Accepted: September 18, 2024. Published: October 15, 2024.
1. Introduction
In the paper industry, the quality of the
finished product is largely determined by the cutting
section. This section's machinery is powered by
induction motors, however because of their
proneness to wear and tear they can have unplanned
breakdowns and expensive downtime. Predictive
maintenance uses machine learning algorithms to
anticipate possible breakdowns in advance
providing a proactive way to reduce these risks [1].
Machine reliability is crucial during this crucial
stage of production in order to maintain
manufacturing processes and satisfy strict quality
standards. Induction Motors which precisely and
effectively drive the cutting tools are among the
vital parts that power these machines [5].However, a
number of variables, such as environmental
considerations, mechanical loads, and wear and tear
can affect how well Induction Motors operate in the
cutting area. Unexpected failures or breakdowns in
these motors can result in expensive production
downtime, lower output, and lowered product
quality [6]. As a result, maintaining the
competitiveness and profitability of paper
production operations depends critically on the
induction motor's dependability and optimal
performance. Predictive maintenance offers a
proactive solution by utilizing cutting-edge
technologies and data-driven insights to anticipate
and prevent failures before they occur [9].
Predictive maintenance systems can identify
early indications of deterioration or approaching
failures by continuously monitoring motor
performance characteristics like temperature,
vibration, sound, speed and current [4]. Predictive
maintenance also optimizes maintenance schedules,
lowers unscheduled downtime and lowers overall
maintenance costs in addition to improving
equipment reliability [3]. This study is to investigate
the benefits, methods and guiding principles of
predictive maintenance with a focus on induction
motors used in the paper industry's cutting division.
The primary objectives of this research are:
DESIGN, CONSTRUCTION, MAINTENANCE
DOI: 10.37394/232022.2024.4.14
Kalavathidevi T., Madhan Mohan M.,
Baluprthviraj K. N., Sangavi B.,
Rajaraghavendraa S. K., Rajeshkanna K.
E-ISSN: 2732-9984
124
Volume 4, 2024
1.To implement the predictive maintenance for
Induction Motors in the cutting section of the paper
industry is to minimize unplanned downtime.
2.To optimize maintenance costs associated with
Induction Motors in the cutting section.
3.To extend the lifespan of Induction Motors in the
cutting section of the paper industry by identifying
and addressing potential issues early on.
Harsh operating environments, complex dynamics,
data availability issues and integration hurdles with
traditional maintenance practices pose significant
obstacles. Overcoming these challenges is essential
for enhancing equipment reliability, minimizing
downtime and optimizing maintenance costs to
ensure operational efficiency and competitiveness in
the market. This is the problem statement given by
the Seshasayee Paper and Boards Ltd (SPB). To
overcome the existing problem, we came over an
idea of predicting the failures in advance by
collecting the data from the induction motor via
sensors. Implementing predictive maintenance
involves deploying sensor systems to monitor
Induction Motor parameters such as temperature,
vibration, sound, and speed. Data is collected and
analyzed using machine learning algorithms to
detect anomalies indicating potential failures. This
method optimizes maintenance practices in the
paper industry's cutting section enhancing
equipment reliability and efficiency.
2. Literature Survey
An Artificial Intelligence based fault
detection system was proposed by Marichal et al.
This work examined the vibration properties of a
water-based oil separation system. In order to
forecast the early stages of failures, the vibration
signals were first processed in the frequency domain
and then applied to a genetic neuro fuzzy system
[7]. The issue of using power signals and a genetic
algorithm to optimize Support Vector Machines
(SVMs) is addressed in order to create an optimal
classification model for electric motor defect
diagnostics. When fault diagnostics is used an
electric motor's operating status can be quickly
identified, and a response is made to increase the
motor's reliability [2]. A real-time monitoring
system with preventive defect detection alerts were
described because it uses big data processing, hybrid
prediction models, and Internet of Things-based
sensors, this system can efficiently handle and
analyze massive volumes of data. In this case, the
hybrid prediction model combines the more accurate
defect identification of Random Forest classification
with noise-based outlier detection. It demonstrated
how the suggested paradigm enhanced decision-
making and helped avoid unforeseen errors [13].
Analysis is carried for predictive maintenance in
aviation, particularly for line maintenance near the
gate. It proposes a methodology utilizing
prognostics and the extended Kalman filter for
optimizing maintenance of redundant aeronautical
systems, considering multiple wear conditions. The
aim is to enhance aircraft availability and reduce
costs, addressing critical aspects of the aviation
industry's maintenance challenges [14].
The unexpected downtime of diagnostic and
therapeutic imaging systems presents financial
issues for hospitals and Original Equipment
Manufacturers (OEMs) in the cost-sensitive
healthcare business. The suggested methodology
makes use of contemporary connectivity to support
the connection of equipment to a typical monitoring
station, allowing for predictive maintenance and
remote monitoring. This proactive strategy reduces
unscheduled downtime by foreseeing possible faults
and is based on a data-driven, machine learning
framework [10]. The recurrent equipment
breakdowns and unplanned downtime from reliance
on Reactive and Preventive Maintenance at
Company X, this paper proposes an Artificial
Intelligence (AI)-based model for optimizing
current maintenance strategies. Using the Nowlan
and Heap risk analysis matrix, critical equipment-
pumps, storage tanks, valves, and the standby power
supply system was identified at the fuel depot.
Ishikawa diagrams were applied to refine the
Preventive Maintenance strategy [8]. The
integration of vibration sensors into the Internet of
Things (IoT) is gaining prominence, driven by
advancing technology that enhances measurement
accuracy and reduces hardware costs. These sensors,
affixed to core equipment in control and
manufacturing systems like motors and tubes, offer
crucial insights into device operational status. Our
data engine focuses on Remaining Usefulness
Lifetime (RUL) estimation, crucial for cyber-
physical system maintenance, demonstrating on real
manufacturing sites a 1.2x extension in tube lifetime
and a 20% reduction in replacement costs
[3].Process Monitoring and Predictive Maintenance
is evident in manufacturing, aiming to cut
maintenance costs and minimize downtime. This
paper introduces an adaptive Predictive
Maintenance based flexible maintenance scheduling
decision support system, emphasizing opportunity
and risk costs. Validation on a real industrial dataset
related to Ion Beam Etching in semiconductor
manufacturing demonstrates the system's
effectiveness in enhancing maintenance strategies
[12]. An experimental approach for integrating
DESIGN, CONSTRUCTION, MAINTENANCE
DOI: 10.37394/232022.2024.4.14
Kalavathidevi T., Madhan Mohan M.,
Baluprthviraj K. N., Sangavi B.,
Rajaraghavendraa S. K., Rajeshkanna K.
E-ISSN: 2732-9984
125
Volume 4, 2024
Industry 4.0 into a small bottling plant, specifically
focusing on early fault detection in conveyor motors
and generate predictive maintenance schedule.
Using advanced programming functions of a
Siemens S7-1200 PLC, vibration speed data is
monitored through sensors, allowing for efficient
maintenance planning. Additionally, a decentralized
monitoring system facilitates cloud-based reporting
and sends immediate email notifications to
supervisors for generated maintenance schedules,
enhancing the practical implementation of Industry
4.0 in the bottling plant [5].
The significance of condition monitoring
and Predictive Maintenance in preventing economic
losses due to unforeseen motor failures and
improves the system reliability. It introduces a
Machine Learning architecture based on the
Random Forest approach for predictive
maintenance. The system's validation involves a real
industry example, incorporating data collection from
diverse sensors, PLCs, and communication
protocols within the Azure Cloud. [9]. The constant
push for reducing operational and maintenance costs
of Induction Motors (IMs) underscores the
importance of regular system health monitoring.
This paper offers a state-of- the-art review on IM
faults and diagnostic schemes, addressing the
increasing demand for condition monitoring in
industrial applications. Various fault diagnosis
techniques for IMs are explored; highlighting the
potential of non-invasive data acquisition for future
dynamic machine maintenance and failure
prediction [1].The paper introduces the PdM
approach, outlines a PdM scheme for automatic
washing equipment and discusses challenges in
PdM research. It categorizes industrial applications
based on six machine learning and deep learning
algorithms, comparing performance metrics for
each. The analysis delves into the accuracy of these
PdM applications, evaluating algorithm
performance in detail [15]. This paper reviews
recent advancements in predictive maintenance for
motors, highlighting outcomes, ongoing research,
and key contributions by researchers, reflecting the
growing importance of PdM in ensuring the
reliability of electric motors [6]. The analysis
categorizes research based on metric capacity unit
algorithms, machine learning class, machinery,
instrumentation, data acquisition devices, and data
size and type. It highlights key contributions by
researchers and offers insights for further research.
The paper presents a Random Forest model for
predicting machine failures in manufacturing,
demonstrating its superior accuracy and precision
compared to the Decision Tree algorithm [4]. This
paper provides an overview of key approaches to
bearing-fault analysis in grinding machines. It
categorizes these approaches into two main parts:
the first involves classifying bearing faults based on
detection, error position, and severity, while the
second focuses on predicting remaining useful life
to optimize replacement costs and minimize
downtime [11]. One-class support vector machine
(OC-SVM), one of the several boundary-based
techniques and algorithms suggested to address this
issue, is regarded as particularly good and efficient.
However, one major disadvantage of this classifier
group is their excessive sensitivity to the presence of
noise and outliers in the training data [16].
This paper presents an innovative method
for anomaly detection and fault diagnosis featuring
online adaptive learning. The method combines
classification and clustering, demonstrating superior
performance, particularly with limited known fault
types. Experimental validation on ball bearing and
Iris datasets confirms its effectiveness [17]. An
ensemble of hybrid intelligent models is proposed
for induction motor condition monitoring. Motor
Current Signature Analysis (MCSA) is chosen for
its online, non-invasive nature and single input
requirement, ensuring cost-effectiveness. The
proposed hybrid model combines the Fuzzy Min–
Max (FMM) neural network with Random Forest
(RF), forming an ensemble of Classification and
Regression Trees for improved accuracy and
robustness in monitoring [18]. This paper introduces
a pattern recognition system for ongoing induction
motor monitoring. It utilizes visually efficient
invariant features to identify 3-D current state space
patterns, enabling automatic fault detection and
severity assessment. The system handles time-
variant electric currents, focusing on identifying
specific patterns in three-phase stator currents.
Simulation and experimental results confirm the
methodology's effectiveness in continuous
monitoring of complex systems [19]. It explains the
generation of vibration and noise in bearings,
covering measurements in both time and frequency
domains. Signal processing techniques like the high-
frequency resonance method are discussed. Acoustic
measurement methods such as sound pressure,
sound intensity, and acoustic emission are reviewed.
Recent research trends, including the wavelet
transform method and automated data processing,
are also examined [20].
DESIGN, CONSTRUCTION, MAINTENANCE
DOI: 10.37394/232022.2024.4.14
Kalavathidevi T., Madhan Mohan M.,
Baluprthviraj K. N., Sangavi B.,
Rajaraghavendraa S. K., Rajeshkanna K.
E-ISSN: 2732-9984
126
Volume 4, 2024
3. Block Diagram for implementing
Predictive Maintenance of Induction
Motor in the cutting section of paper
industry
Fig. 1: Block Diagram for implementing Predictive
Maintenance of Induction Motor
The block diagram for Predictive Maintenance of
Induction Motor in the cutting section of paper
industry is depicted in Fig. 1. It consists of a
Induction Motor, Sound sensor, Temperature sensor,
Vibration Sensor, IR sensor, Alarm system and
LCD Display. Various sensors are used to monitor
parameters of the induction motor including
vibration, temperature, sound and speed sensors.
These sensors are connected to ESP32 controller
which collects data from the sensor systems and
processes it into a usable format. The collected data
and real time monitoring data where stored in the
database. Then the machine learning models are
used to identify patterns, anomalies and potential
faults in the motor's operation.Finallythe algorithms
predicts the future health condition of the induction
motor based on the current data and historical
trends.
4. Prototype for Predictive
Maintenance of Induction Motor in
the cutting section of paper industry
The hardware system for predictive
maintenance of induction motors in the paper
industry’s cutting section incorporates a variety of
sensors including vibration sensor, temperature
sensor, speed sensor and sound sensor. These
sensors are strategically placed on the motor and
surrounding equipment to capture data related to the
motor’s operating conditions. The hardware system
includes a predictive analytics module that utilizes
machine learning algorithms to analyze sensor data
and predict potential motor failures. This module
continuously learns from historical data to improve
accuracy over time. A user interface which is a
graphical display provides visualizations of motor
health metrics and alerts for detected faults. This
interface enables operators and maintenance
personnel to make informed decisions. A reliable
power supply ensures continuous operation of the
hardware components even during power outages.
Fig.2: Hardware developed for Predictive
Maintenance of Induction Motor
Fig. 2 shows the Prototype of Predictive
Maintenance of Induction Motor in the cutting
section of paper industry. By integrating these
hardware components into a compatible system
predictive maintenance of induction motors in the
paper industry’s cutting section becomes feasible.
The system enables proactive maintenance
practices, reduces downtime and enhances overall
operational efficiency.
5. Percentage Occurrence of Induction
Motor faults
Fig. 3 describes the percentage occurrence of
faults in Induction Motors which can vary
depending on factors such as operating conditions,
maintenance practices and environmental factors.
Bearing wear is one of the most common faults in
induction motors, accounting for a significant
percentage of total faults. The occurrence rate is
observed as 69% of all motor faults. Rotor bar and
end ring defects such as broken bars or high
resistance connections are another prevalent fault in
induction motors. Their occurrence rate is typically
DESIGN, CONSTRUCTION, MAINTENANCE
DOI: 10.37394/232022.2024.4.14
Kalavathidevi T., Madhan Mohan M.,
Baluprthviraj K. N., Sangavi B.,
Rajaraghavendraa S. K., Rajeshkanna K.
E-ISSN: 2732-9984
127
Volume 4, 2024
observed as 7%. Stator winding faults including
short circuits, open circuits and insulation
degradation are also relatively common in induction
motors. Their occurrence rate varies but generally
falls within 21%. Shaft misalignment which can
occur due to improper installation or mechanical
wear is another significant fault in induction motors.
Its occurrence rate is typically within 3%. Hence
from this it is analyzed that the bearing fault is the
major cause for the induction motor failure.
Fig. 3: Percentage occurrence of Induction motor
faults
6. Dataset description and Pre-
processing
6.1 Gathering Data
This study uses actual data from a local paper
firm that supplies hundreds of businesses with paper
products as described in Table 1. This article
focuses on the examination and forecasting of an
Induction motor projection from a paper
manufacturer. With an initial sampling interval of
two hours, the information in result describes the
fundamental parameters of temperature, vibration,
speed, and sound utilized by this paper company to
describe the operation of its induction motor. To aid
in the study of Induction Motors in order to avoid
failure and to direct the company's manufacturing.
The data’s were collected every 240 minutes for 12
days. The time series plot was obtained as shown in
Fig. 4.
Table 1. Database obtained from Sensors
S.No
Temperature
(°C)
Vibration
(m/s2)
Sound
(dB)
x10
Speed
(RPM)
x10
1.
77
4
7
28
2.
49
4
6
86
3.
32
0
5
76
4.
103
5
3
68
5.
90
0
2
143
6.
124
1
5
87
7.
130
2
6
142
8.
99
3
15
92
9.
52
1
8
113
10.
79
1
12
52
6.2 Data Pre-processing
The outlined procedure addresses common
challenges encountered in gathered data analysis,
encompassing the handling of null values, outliers,
time index setup and data normalization.
1. Null and Outlier Management: During data
compilation, efforts were made to identify and
eliminate null values and outliers. Missing data
points were eradicated, and duplicate entries were
removed from consideration.
2. Establishing the Time Index: To facilitate time
series analysis, it was imperative to convert the
data's datetime column into a datetime data type.
Consequently, the DataFrame's index was set to the
datetime type, with retime chosen as the designated
timestamp for data indexing.
3. Normalization Procedure: Normalization was
carried out with the objective of enhancingmodel
training precision and expediting convergence. The
normalization formula, as described in Equation 1,
was employed:
xnorm=x–xmin/xmax–xmin (1)
By systematically addressing these steps, the time
series data is prepared for subsequent analysis or
modeling, ensuring data integrity and suitability for
further investigation.
6.3 Data Visualization and analysis
Data input, time series creation, etc. are all
included in the visual presentation. Excel files
containing the data for 2023 and 2024 must now be
exported to a Python environment in order to be
displayed as a data frame. A programmable loop is
used to import because of the volume of data and
the repeated nature of the process. Since the data is
only recorded once every minute, down sampling
the data is required since the presentation of the data
in minutes is both too large and too little. Matplotlib
was then used to perform a visual analysis, and Fig.
4 displays the average monthly Induction Motor
projection for 2024.
DESIGN, CONSTRUCTION, MAINTENANCE
DOI: 10.37394/232022.2024.4.14
Kalavathidevi T., Madhan Mohan M.,
Baluprthviraj K. N., Sangavi B.,
Rajaraghavendraa S. K., Rajeshkanna K.
E-ISSN: 2732-9984
128
Volume 4, 2024
Fig. 4: Average monthly Induction Motor Prediction
statics for 2024
7. Induction Motor Prediction
7.1 Random Forest Tree
It can classify different fault types and
estimate remaining useful life based on historical
data and sensor measurements. The strategy
employed here is to achieve these objectives is as
follows:
1. Minimizing Individual Error: To ensure low
individual error, trees are expanded to their
maximum depth.
2. Minimizing Residual Correlation: To reduce
residual correlation,
a) Each tree is grown on a bootstrap sample
obtained from the training dataset.
b) A significantly smaller subset m (where
m<<p, p representing the number of covariates) is
specified. At each node of every tree, m covariates
are randomly selected, and the optimal split for that
node is determined based on these selected
covariates.
This approach aims to minimize both individual
error and residual correlation, thereby improving the
model's predictive accuracy and generalization
capabilities.
Entropy is measured to check the impurity in a
dataset. It's often calculated using the formula,
(2)
where H(S) is the entropy of a set S, c is the number
of classes, and pi is the proportion of instances in
class i.
Information gain measures the reduction in entropy
or impurity after splitting a dataset on a particular
attribute. It's calculated using a formula similar to:
) (3)
where IG (D,A) represents the information gain
achieved by splitting dataset D on attribute A, |D| is
the total number of instances in dataset D, |DV| is the
number of instances in dataset D with value V for
attribute A, and H(D) and H(DV) are the entropies of
dataset D and its subsets DV respectively. Fig. 5
shows the block diagram of Random Forest Tree
which is a machine learning algorithm commonly
effective in analysing sensor data to predict motor
failures and schedule maintenance tasks. The
Random Forest Tree assesses motor health by
monitoring temperature (>85°C), vibration (>2.5
m/s²), sound (>70dB), and speed (<700rpm). Any
deviation signals potential failure.
Fig. 5: Block Diagram of Random Forest Tree for
Induction Motor Prediction
7.2 Linear Regression
Linear Regression provides the relationship between
two variables that is a dependent variable and an
independent or explanatory variable. It is used in
prediction, forecasting and error reduction. Fig. 6
depicts the relationship between dependent and
independent variables.
Fig. 6: Relationship between dependent and
independent variables
DESIGN, CONSTRUCTION, MAINTENANCE
DOI: 10.37394/232022.2024.4.14
Kalavathidevi T., Madhan Mohan M.,
Baluprthviraj K. N., Sangavi B.,
Rajaraghavendraa S. K., Rajeshkanna K.
E-ISSN: 2732-9984
129
Volume 4, 2024
In predictive maintenance of induction motors using
linear regression, the one commonly used equation
is the linear regression model itself.
y=β0+β1x1+β2x2+……………+βnxn+ϵ (4)
where, y is the dependent variable (e.g., motor
health indicator), x1, x2,……..,xn are the independent
variables (e.g., motor operating conditions,
environmental factors), β0+ β1+………+ βn are the
coefficients representing the relationship between
the independent variables and the dependent
variable, ϵ represents the error term. In this specific
scenario, the dependent variable would likely be a
binary outcome indicating whether the motor is in a
failure condition or not (e.g., 1 for failure, 0 for non-
failure). The independent variables would be the
parameters that are believed to influence motor
failure, such as temperature, vibration, sound, and
speed. If the predicted probability of failure exceeds
a certain threshold, the motor would be classified as
being in a failure condition. Otherwise, it would be
classified as not being in a failure condition.
7.3 Support Vector Machines
Support Vector Machines (SVMs) are a
popular algorithm used in predictive maintenance
tasks including those related to induction motors.
SVMs are primarily used for classification tasks
which involve predicting a categorical label based
on input features. In the context of predictive
maintenance for induction motors, SVMs can be
used to classify the health status of the motor (e.g.,
normal, faulty, impending failure) based on features
extracted from sensor data or other relevant sources.
The basic idea of SVM is to find the
hyperplane that best separates the data points
belonging to different classes while maximizing the
margin between the classes. In the case of non-
linearly separable data SVM can use a kernel trick
to map the input data into a higher-dimensional
space where it becomes linearly separable. The
features extracted from sensor data exhibit a linear
separation between different health states of the
motor (e.g., normal, faulty) so that the linear SVM
is preferred. For example, if the sensor
measurements directly correlate with the health
status in a roughly linear manner a linear SVM can
provide a simple and effective solution.
The decision function for a linear SVM is proposed
as,
f(x)=sign(w.x+b) (5)
Where:
x is the input feature vector,
w is the weight vector,
b is the bias term,
sign is the sign function indicating the predicted
class.
The weight vector w and bias term b are learned
during the training phase.
The objective function of the linear SVM can be
formulated as a constrained optimization problem:
(6)
subject to the constraints:
yi(w.xi+b)≥1for i = 1,…..,N (7)
where xi are the training samples, yi are their
corresponding labels (+1 or -1 for binary
classification), and N is the number of training
samples.
In the case of SVM for predictive
maintenance of induction motors, the input features
would include various sensor measurements such as
temperature, vibration, current, etc., and the output
would be the health status of the motor (e.g.,
normal, faulty). The SVM algorithm would learn to
classify the health status based on these features.
The SVM model is trained using the pre-processed
data. During training, the model learns to
distinguish between normal and failure conditions
based on the provided features. In this case, the
parameters for the SVM model would need to be set
to appropriately capture the conditions specified,
temperature > 85°C, vibration > 2.5 m/s², sound >
70 dB, and speed < 700 rpm. These parameters
would guide the decision boundary of the SVM to
classify instances accordingly.
Finally, the trained SVM model can be used
to predict the condition of the motor in real-time
based on new measurements of temperature,
vibration, sound, and speed. If the conditions
specified (temperature > 85°C, vibration > 2.5 m/s²,
sound > 70 dB, and speed < 700 rpm) are not met,
the SVM model would predict that the motor is
under a failure condition.
8. Experimental Flowchart using the
algorithms
Fig. 7 represents the flowchart of Predictive
Maintenance of Induction Motor in the cutting
section of paper industry. Algorithm for predicting
the failure of the induction motor is shown below:
Step 1: Data Collection: Gather data from the
induction motor and its environment. This includes
motor operating parameters such as temperature,
vibration data, speed and sound.
DESIGN, CONSTRUCTION, MAINTENANCE
DOI: 10.37394/232022.2024.4.14
Kalavathidevi T., Madhan Mohan M.,
Baluprthviraj K. N., Sangavi B.,
Rajaraghavendraa S. K., Rajeshkanna K.
E-ISSN: 2732-9984
130
Volume 4, 2024
Step 2: Data Preprocessing: Clean the data to
remove noise and handle missing values. Normalize
the data to ensure consistency and compatibility
across different features.
Step 3: Feature Extraction: Extract relevant features
from the preprocessed data.
Step 4: Model development: Suitable machine
learning models such as Linear regression, Support
Vector machine and Random Forest tree models are
chosen for parameter prediction.
Step 5: Train the selected models using historical
data, with features as inputs and the target variable
being either future motor condition or likelihood of
failure.
Step 6: Model Evaluation: Validate models using
separate test datasets to ensure their effectiveness.
Step 7: Deployment and monitoring: Set up real-
time monitoring systems to collect streaming data
from the motor. Apply the trained model to make
predictions or detect anomalies.
By implementing a predictive maintenance
algorithm adapt to the specific needs of the cutting
section of the paper industry, businesses can
minimize downtime, reduce maintenance costs and
optimize the lifespan of critical equipment like
Induction Motors.
Fig. 7: Flowchart of Predictive Maintenance of
Induction Motor
9. Results and Discussion
The trained model by Support Vector
Machines algorithm, the outputs are obtained under
two conditions,
a)Induction Motor operated under normal condition
b)Induction Motor operated under fault condition
Table 2 shows the Conditions for
Temperature, Vibration, Speed and Sound of the
motor to determine whether it is under normal state
or fault state. If the temperature value exceeds 85°C
or Vibration value exceeds 2.5 m/s2 or Sound value
exceeds 7dB or Speed value decreases below 70rpm
the motor will face failure.
Table 2. Test Set Conditions
Parameters
Conditions
Temperature
>85°C
Vibration
>2.5 m/s2
Sound
>7dB x 10
Speed
<70RPM x 10
a) Induction Motor operated under normal
condition
Table 3. Real-time monitored data of normal motor
(Without fault)
Temperat
ure (°C)
Sound
x10
(dB)
Vibrati
on
(m/s2)
Speed
x 10
(RPM
)
Motor’s
Condition
33.23558
0.7820
0
63
NORMAL
22.48289
4.3988
0
69
NORMAL
41.05572
0.6827
1.2736
65
NORMAL
25.41545
5.6696
0.7542
75
NORMAL
31.28055
3.0303
1.2078
77
NORMAL
46.43206
0.4455
0.2234
75
NORMAL
30.79179
4.6920
1.1706
73
NORMAL
37.14565
3.2258
1.5564
78
NORMAL
59.13979
2.3460
2.1804
82
NORMAL
68.91496
1.1730
1.9962
88
NORMAL
Table 3 shows the real-time monitored
parameters of the induction motor this includes
Temperature, Vibration, Speed and Sound of the
motor. Table 3 depicts the real-time monitored data
when a motor is operating under normal conditions,
without any fault, typically includes a range of
parameters and metrics that indicate the health and
performance of the motor. For example, in Table 3
Temperature is noted as 33.23558°C, Sound is noted
as 0.78201 dB, Vibration is observed as 0 m/s2 and
Speed is noted as 63 RPM, all these data were
below the test set conditions (Temperature >85°C,
Sound >7dB, Vibration >2.5 m/s2 and Speed <70
DESIGN, CONSTRUCTION, MAINTENANCE
DOI: 10.37394/232022.2024.4.14
Kalavathidevi T., Madhan Mohan M.,
Baluprthviraj K. N., Sangavi B.,
Rajaraghavendraa S. K., Rajeshkanna K.
E-ISSN: 2732-9984
131
Volume 4, 2024
RPM). From these values it is validated that the
motor is operating without any fault.
These data are crucial for ensuring smooth
operation and early detection of any potential issues.
From the above data the motor failure can be
predicted according to the trained dataset. If any of
the four parameters fails to achieve the normal value
then we can say that the motor is in fault stage
otherwise the motor will operate under normal
condition. Here there will be no failure occurs since
it has no faults in it.
Fig. 8: The Output obtained in ThinkSpeak for Real-
time monitored data when motor is operated under
normal state (Without fault)
From Fig. 8 it is seen that the motor is
operating normally without any fault. When
monitoring a motor's operation under normal
conditions without any faults, the real-time data
displayed on a platform like ThingSpeak provides a
comprehensive view of various parameters. For
example, in Table 3 Temperature is noted as
25.41545°C, Sound is noted as 5.6696 dB, Vibration
is observed as 0.7542 m/s2 and Speed is noted as 75
RPM, all these data were below the test set
conditions (Temperature >85°C, Sound >7dB,
Vibration >2.5 m/s2 and Speed <70 RPM). These
parameters are usually displayed in real-time on a
ThingSpeak dashboard or interface, allowing
operators to monitor the motor's health and
performance remotely. Any deviations from
expected values or sudden changes outside normal
ranges may trigger alerts or notifications, prompting
further investigation or preventive maintenance
actions to ensure uninterrupted operation and
prevent potential faults or failures.
b)Induction Motor operated under fault
condition
Table 4 shows the real-time monitored
parameter’s of the fault induction motor this
includes Temperature, Vibration, Speed and Sound
of the motor.
Table 4. Real-time monitored data of fault motor
(With fault)
Temperat
ure (°C)
Sound
x10
(dB)
Vibrati
on
(m/s2)
Speed
x 10
(RPM
)
Motor’s
Condition
60.5368
5.8125
0
90
NORMAL
86.2889
14.698
2.564
70
FAILURE
81.1521
13.264
2.8736
75
FAILURE
85.5145
14.597
3.7542
85
FAILURE
99.8855
13.135
2.2458
75
FAILURE
96.5366
8.1435
2.2734
60
FAILURE
100.8927
14.321
2.4716
80
FAILURE
104.8445
15.139
1.5784
70
FAILURE
109.1969
12.540
2.9847
65
FAILURE
118.9466
11.710
1.9972
50
FAILURE
Table 4 depicts the real-time monitored data
of fault motor. Real-time monitored data of a faulted
motor provides crucial insights into the abnormal
conditions and issues affecting its operation. From
the above data the motor failure can be predicted
according to the trained dataset. For example, in
Table 4 Temperature is noted as 86.2889°C, Sound
is noted as 14.6982 dB, Vibration is observed as
2.564 m/s2 and Speed is noted as 70 RPM, all these
data were exceeded the test set conditions
(Temperature >85°C, Sound >7dB, Vibration >2.5
m/s2 and Speed <70 RPM). From these values it is
validated that the motor is facing failure.
Fig. 9: The Output obtained for Real-time monitored
data when motor is operated under fault state (With
fault)
Here, the fault occurs due to bearing failure
majorly and shaft failure minorly. If any of the four
parameters fails to achieve the normal value then we
can say that the motor is in fault stage otherwise the
motor will operate under normal condition. Fig. 9
DESIGN, CONSTRUCTION, MAINTENANCE
DOI: 10.37394/232022.2024.4.14
Kalavathidevi T., Madhan Mohan M.,
Baluprthviraj K. N., Sangavi B.,
Rajaraghavendraa S. K., Rajeshkanna K.
E-ISSN: 2732-9984
132
Volume 4, 2024
depicts the Output obtained for Real-time monitored
data when motor is operated under fault state (With
fault). Fig. 9 depicts sudden spikes in the graph
that deviate significantly from the expected patterns.
In Fig. 9 the glitches is due to the bearing failure
and shaft failure. Because of that the Temperature
and Vibration varies drastically to the highest point
where as the Speed and Sound varies to the lowest
point. These anomalies could indicate faults or
irregularities in the motor's operation.
Calculating prediction accuracy of Support
Vector Machine algorithm is another method for
evaluating and comparing classifiers. The values
can be obtained from Table 3 and Table 4.
(8)
Where TP = Number of true positives instances
(TP),
TN = Number of true negatives instances (TN).
The model correctly identifies 12 samples as true
positives (failure of motor) and 9 samples as true
negative (motor under normal state) out of a total of
25 samples, the accuracy is computed as follows:
Accuracy = (12 + 9) / 25 * 100 = 84%
The accuracy comparison of various Machine
Learning algorithms carried out is shown below,
Table 5. Accuracy rate of various algorithms
Model
Random
Forest Tree
Linear
Regression
Support
Vector
Machines
Accuracy
75%
72%
84%
Table 5 depicts the accuracy rate of various
algorithms, when comparing the accuracy of these
algorithms it's essential to consider the specific
characteristics of the dataset such as its size,
dimensionality, linearity and class distribution.
Conducting cross-validation and tuning
hyperparameters can help in obtaining more reliable
accuracy estimates for each algorithm. Additionally,
ensemble methods like Random Forest can
sometimes outperform individual models like Linear
Regression or SVM particularly in complex datasets
with nonlinear relationships.
The performance comparison of various Machine
Learning algorithms is depicted below,
Table 6. Performance Comparison among various
algorithms
Metrics
Random
Forest Tree
Linear
Regression
Support
Vector
Machines
Precision
0.87
0.68
0.92
Recall
0.82
0.75
0.88
F1-Score
0.84
0.71
0.90
Based on the results from Table 6, we can
observe that the Support Vector Machines (SVM)
algorithm outperforms both Random Forest and
Linear Regression in terms of accuracy, precision,
recall, and F1-score. SVM achieves the highest
values for all these metrics, indicating its superiority
in predicting motor failure conditions based on the
specified parameters.
10. Conclusion
In conclusion, Predictive Maintenance of
Induction Motors in the paper industry's cutting
section holds significant promise for improving
operational efficiency, reducing downtime and
minimizing maintenance costs. By leveraging
advanced techniques such as sensor data analysis,
machine learning algorithms like Random Forest,
Linear Regression, Support Vector Machines and
IoT-enabled monitoring systems, predictive
maintenance enables early detection of potential
faults and proactive intervention before failures
occur. Improved Reliability by identifying and
addressing potential issues before they escalate into
major failures, predictive maintenance enhances the
reliability and uptime of induction motors in the
cutting section. Cost Savings by Proactive
maintenance strategies help minimize unplanned
downtime, reduce repair costs and optimize
maintenance schedules, resulting in significant cost
savings for paper industry operations.
Acknowledgement:
We thank Kongu Engineering College for
motivating and encouraging us to do this project in
our academic year.
References:
[1] Choudhary A., Goyal D., Shimi S.L. and Akula
A., Condition monitoring and fault diagnosis of
induction motors, Archives of Computational
Methods in Engineering, Vol.26, No.4, 2019,
pp. 1221-1238.
[2] Gou X., Bian C., Zeng F., Xu Q., Wang W. and
Yang S.,A Data-Driven smart fault diagnosis
method for electric motor, IEEE International
DESIGN, CONSTRUCTION, MAINTENANCE
DOI: 10.37394/232022.2024.4.14
Kalavathidevi T., Madhan Mohan M.,
Baluprthviraj K. N., Sangavi B.,
Rajaraghavendraa S. K., Rajeshkanna K.
E-ISSN: 2732-9984
133
Volume 4, 2024
Conference on Software Quality, Reliability
and Security Companion, 2018, pp. 250-257.
[3] Jung D., Zhang Z. and Winslett M.,Vibration
analysis for IoT enabled predictive
maintenance, International conference on data
engineering, 2022, pp. 1271-1282.
[4] Karrupusamy P., Machine Learning Approach
to Predictive Maintenance in Manufacturing
Industry, Journal of Soft Computing Paradigm,
Vol.02, No.4, 2020, pp. 246-255.
[5] Kiangala K.S. and Wang Z.,Initiating
predictive maintenance for a conveyor motor in
a bottling plant using industry 4.0 concepts,
The International Journal of Advanced
Manufacturing Technology, Vol. 97, 2023,
pp.3251-3271.
[6] Manjare A.A. and Patil B.G.,Condition based
techniques and predictive maintenance for
motor.International Conference on Artificial
Intelligence and Smart Systems, 2021, pp. 807-
813.
[7] Marichal G.N., Avila D., Hernandez A., Padron
I. and Castejon C.,Feature extraction from
indirect monitoring in marine oil separation
systems,Sensors, Vol. 18, No. 9, 2018,
pp.3159.
[8] Mushiri T., Hungwe R. and Mbohwa C.,An
artificial intelligencebased model for
implementation in the petroleum storage
industry to optimize maintenance,
International Conference on Industrial
Engineering and Engineering Management,
2021, pp. 1485-1489.
[9] Paolanti M., Romeo L., Felicetti A., Mancini
A., Frontoni E. and Loncarski J.,Machine
learning approach for predictive maintenance in
industry 4.0., International Conference on
Mechatronic and Embedded Systems and
Applications, 2020, pp. 1-6.
[10] Patil R.B., Patil M.A., Ravi V. and Naik
S.,Predictive modeling for corrective
maintenance of imaging devices from machine
logs, International Conference of the IEEE
Engineering in Medicine and Biology Society,
2019, pp. 1676-1679.
[11] Schwendemann S., Amjad Z. and Sikora A.,A
survey of machine-learning techniques for
condition monitoring and predictive
maintenance of bearings in grinding machines,
Computers in Industry, Vol. 125, 2021,
pp.103380.
[12] Susto G.A., Wan J., Pampuri S., Zanon M.,
Johnston A.B., Hara P.G. and McLoone S.,An
adaptive machine learning decision system for
flexible predictive maintenance,International
Conference on Automation Science and
Engineering, 2019, pp.18–22.
[13] Syafrudin M., Alfian G., Fitriyani N.L. and
Rhee J.,Performance analysis of IoT-based
sensor, big data processing, and machine
learning model for real-time monitoring system
in automotive manufacturing,Sensors, Vol. 18,
No. 9, 2018, pp.2946.
[14] Vianna W.O.L. and Yoneyama T.,Predictive
maintenance optimization for aircraft redundant
systems subjected to multiple wear
profiles,IEEE Systems Journal, Vol. 12, No. 2,
2020, pp.1170-1181.
[15] Weiting Zhang, Dong Yang and Hongchao
Wang,Data-Driven Methods for Predictive
Maintenance of Industrial Equipment,IEEE
Systems Journal, Vol. 13, No. 3, 2022, pp.
2213-2227.
[16] Tian Y, Mirzabagheri M, Bamakan S.M.H,
Wang H. and Qu Q.,Ramp loss one-class
support vector machine; a robust and effective
approach to anomaly detection problems,
Neurocomputing, Vol. 310, 2018, pp.223-235.
[17] Dong L.I., Shulin L.I.U. and Zhang H.,A
method of anomaly detection and fault
diagnosis with online adaptive learning under
small training samples, Pattern Recognition,
Vol. 64, 2017, pp.374-385.
[18] Seera M., Lim C.P., Nahavandi S. and Loo
C.K.,Condition monitoring of induction
motors: A review and an application of an
ensemble of hybrid intelligent models,Expert
Systems with Applications, Vol. 41, No. 10,
2019, pp.4891-4903.
[19] Martins J.F., Pires V.F. and Amaral
T.,Induction motor fault detection and
diagnosis using a current state space pattern
recognition, Pattern Recognition Letters, Vol.
32, No. 2, 2020, pp.321-328.
[20] Tandon N. and Choudhury A.,A review of
vibration and acoustic measurement methods
for the detection of defects in rolling element
bearings, Tribology International, Vol. 32, No.
8, 2017, pp.469-480.
DESIGN, CONSTRUCTION, MAINTENANCE
DOI: 10.37394/232022.2024.4.14
Kalavathidevi T., Madhan Mohan M.,
Baluprthviraj K. N., Sangavi B.,
Rajaraghavendraa S. K., Rajeshkanna K.
E-ISSN: 2732-9984
134
Volume 4, 2024
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
that are relevant to the content of this article.
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
