International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN: 2395-0072

Artificial Intelligence-Based Technique for Fault Detection and

Diagnosis of PMSM Motor
Dr Sandhya Kulkarni 1, Aniket Lakhe 2, Mayuri Rajput3

1 Professor, Department of Electrical Engineering, Government College of Engineering Aurangabad,

Chhatrapati Sambhajinagar, Maharashtra, India.
2 UG Student, Department of Electrical Engineering, Government College of Engineering Aurangabad, Chhatrapati

Sambhajinagar, Maharashtra, India.

3 UG Student, Department of Electrical Engineering, Government College of Engineering Aurangabad, Chhatrapati

Sambhajinagar, Maharashtra, India.

---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - This project presents the detection and 1.INTRODUCTION
diagnosis of stator winding short-circuit faults in
Permanent Magnet Synchronous Motors (PMSM) using Fault Detection and Diagnosis (FDD) is an essential
artificial intelligence techniques. PMSM behavior is modeled condition monitoring technique designed to assess the
and analyzed in MATLAB Simulink to simulate both healthy operational status of electric motors. By enabling early
and faulty operating conditions. Simulation provides detection of faults and distinguishing between fault types,
insights into dynamic behavior and electromechanical FDD helps in proactive decision-making to prevent
interactions under fault conditions without requiring potential hazards and ensure system reliability.
physical experimentation.
Permanent Magnet Synchronous Motors (PMSMs) are
The approach involves simulating PMSM performance under widely used in modern electric vehicles (EVs) due to their
normal and stator winding short-circuit scenarios. Results high efficiency, high torque density, and superior
demonstrate close alignment between simulated and performance characteristics. However, like all electrical
expected behaviors. For fault detection, the K-Nearest machines, PMSMs are vulnerable to various types of faults
Neighbours (KNN) algorithm is applied to classify motor during operation, with stator winding short-circuit faults
conditions based on features extracted from current signals. being among the most common and critical. Early
For fault diagnosis and severity estimation, Decision Tree detection and accurate diagnosis of such faults are
Classifier and Random Forest Regressor models are used. essential to ensure the reliability, safety, and longevity of
EV drive systems.
The analysis shows KNN effectively detects faults, while
Decision Tree and Random Forest models accurately Traditional fault detection methods, such as thermal
diagnose the type and extent of stator faults. The motor monitoring or vibration analysis, are often limited in terms
modeling and fault analysis system, developed in MATLAB of sensitivity and response time. With the advancement of
Simulink, enables detailed study and evaluation of fault computational tools and artificial intelligence (AI)
scenarios. techniques, data-driven approaches have emerged as
powerful alternatives for the early detection and diagnosis
Using machine learning (ML) for PMSM fault detection and of motor faults. AI-based systems can analyze complex
diagnosis offers advantages over traditional methods. ML patterns in motor signals and classify fault types with high
models can automatically learn complex patterns from accuracy.
large datasets, allowing for early and accurate fault
detection. Once trained, these models enable real-time This project focuses specifically on the detection and
monitoring and predictive maintenance, enhancing system diagnosis of stator winding short-circuit faults in PMSMs
reliability and reducing costs. This data-driven approach using AI techniques. The PMSM motor model is developed
significantly improves the efficiency and effectiveness of and simulated in MATLAB Simulink software under both
fault diagnosis in PMSM. healthy and faulty conditions. For fault detection, the K-
Nearest Neighbours (KNN) algorithm is employed, which
Key Words: AI-Based techniques, Machine Learning classifies the motor's operational status based on features
(ML), KNN, Decision Tree Classifier, Random Forest extracted from the stator current signals. For fault
Regressor. diagnosis and severity estimation, Decision Tree Classifier
and Random Forest Regressor models are used to further
analyze and categorize the fault condition.

© 2025, IRJET | Impact Factor value: 8.315 | ISO 9001:2008 Certified Journal | Page 732
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN: 2395-0072

1.1 Problem Statement 2.1 Components used in Simulink Model

Electric Vehicle (EV) motors are key to how well an EV  Battery
performs, how efficient it is, and how safe it remains over
time. However, these motors can develop faults like short Battery is a device that stores chemical energy and
circuits in the stator windings, which, if not caught early, converts it into electrical energy through electrochemical
can cause sudden failures, increase repair costs, and reactions. It is a portable and convenient source of DC
shorten the motor’s life. (Direct Current) power.

Traditional ways of detecting motor faults usually involve In our project simulation, we have used a battery model to
a lot of physical checking and may not catch problems represent the power source for the electric motor system.
early enough. Thanks to advancements in Artificial The battery was configured with a voltage rating of 300
Intelligence (AI) and machine learning, we can now spot volts, which is a typical value for mid-range electric
motor issues faster and more accurately using data-based vehicle applications. This voltage level was chosen to
methods that don't require installing extra sensors[1]. ensure the simulation closely resembles real-world EV
performance scenarios, allowing accurate analysis of
By using motor models built in MATLAB Simulink and current flow, motor response, and overall system
applying AI techniques like K-Nearest Neighbours (KNN), behaviour under various operating conditions.
Decision Trees, and Random Forests, it’s possible to create
a smart system that can detect faults quickly and even  Inverter
judge how serious the problem is. This helps keep EVs
safe, reliable, and efficient. An inverter in an electric vehicle is a key power electronics
device that converts DC (Direct Current) from the battery
Because of its importance to today’s automotive industry into AC (Alternating Current) to drive the AC motor
and the push toward smarter electric vehicles, this topic is (typically a synchronous or induction motor). Since most
very relevant and opens new opportunities for research EV motors operate on 3-phase AC power, the inverter
into AI-based motor monitoring systems[2]. plays a vital role in enabling motor control and vehicle
movement.
1.2 Objective Formation
In our project simulation, we have incorporated an
The goal is to develop a smart system that can quickly inverter block to control the AC synchronous motor. The
detect and diagnose faults in Electric Vehicle (EV) motors inverter receives 300V DC from the simulated battery and
using Artificial Intelligence (AI) techniques. By simulating converts it into 3-phase AC, which is then supplied to the
motor behaviour in MATLAB Simulink and applying motor. This setup enables accurate modeling of motor
machine learning algorithms like K-Nearest Neighbours dynamics, torque-speed characteristics, and overall
(KNN), Decision Trees, and Random Forests, the system system performance under various load conditions.
will identify the type of fault and how severe it is. This will
help prevent motor failures, reduce maintenance costs,  VI Measurement
and improve the safety and reliability of EVs[2].
The Voltage and Current Measurement block in MATLAB
2. IMPLIMENTATION DETAILS Simulink is used to monitor electrical quantities such as
voltage and current in a simulated electrical circuit.

In our project simulation, we have used the VI

Measurement block to accurately monitor the voltage
across the Inverter and the current supplied to the motor.
This helped us analyze the electrical behaviour of the
system under different load and speed conditions,
ensuring realistic simulation results and performance
validation.

 Current and Speed Controller

In electric vehicle (EV) motor control systems, Current and
Speed Controllers are essential components that ensure smooth
and efficient motor performance. Current Controller regulates
the motor’s armature or stator current, which directly affects
Figure-1: System Modeling in Simulink
torque output, while Speed controller maintains the

© 2025, IRJET | Impact Factor value: 8.315 | ISO 9001:2008 Certified Journal | Page 733
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN: 2395-0072

rotational speed of the motor based on driver or system 3. FAULT INJECTION AND SIMULATION
input.

 Three Phase Fault Block

The Three-Phase Fault block in MATLAB Simulink is used
to simulate electrical faults in power systems, such as
short circuits between phases or to ground. It is a critical
tool for analyzing system behavior under fault condition.

 Permanent Magnet Synchronous Motor

Figure-4: Fault Injection and Simulation

3.1 Speed and Torque Characteristics

Figure-2: Structure of PMSM

A PMSM is a type of AC synchronous motor in which the

rotor contains permanent magnets instead of windings as
shown in figure-2.

PMSM have several key advantages that make them a

popular choice for electric vehicles (EVs):

 High Efficiency
 High Power Density
 Precise Control
 High Torque Density
 Reliability and Durability Figure-5: Speed and Torque Characteristics of PMSM

In the present study, a Permanent Magnet Synchronous

Motor (PMSM) is used to simulate the operational
behaviour under both normal and fault conditions as
shown in figure-5. The motor is initially set to operate at a
constant speed of 1500 RPM, which corresponds to the
synchronous speed in a 50 Hz, 4-pole system.

Normal Operating Condition (0 – 5 seconds):

 Speed: Speed increases to 1500 RPM and stays

constant.
 Starting Torque: At startup, the motor generates a
Figure-3: PMSM Motor Efficiency Vs Output Power plot. high torque of 18 N·m to overcome initial inertia.
 Running Torque: After achieving rated speed, the
Figure-3 shows plot between Motor Efficiency and Output torque stabilizes at 15 N·m, which is sufficient to
Power. This motor was selected for its high efficiency, drive the mechanical load in steady-state
torque control accuracy, and suitability for electric vehicle operation.
applications. It allows us to simulate realistic vehicle
behaviour under varying load and speed conditions, The torque remains steady and aligned with the load
supporting detailed analysis of performance and energy demand during this period, indicating normal operation of
consumption[3]. the PMSM without any faults or disturbances.

© 2025, IRJET | Impact Factor value: 8.315 | ISO 9001:2008 Certified Journal | Page 734
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN: 2395-0072

Fault Condition (After 5 seconds):

A fault is introduced at 5 secondsby using Three Phase

Fault block as shown in figure-4.

 Speed Response: Due to the fault, the motor is

unable to sustain its rotation and the speed drops
rapidly to 0 RPM, which we can observe in figure-
5.
 Torque Response: As the motor loses
synchronization and power delivery is disrupted,
the torque also decreases sharply, this is
illustrated in figure-5.

This behaviour indicates a complete shutdown or stalling Figure-6-b.

of the motor as a result of the fault, which reflects a loss of Figure-6(a,b): Current under normal operating condition
control or failure in the drive system. of PMSM.

This behaviour of the PMSM under normal and fault During normal operation, the PMSM motor draws a steady
conditions is essential for validating the fault detection current of 17 A. This current supports continuous
and diagnosis technique being developed in our project. operation at 1500 RPM with a running torque of 15 N.m.
The clear transition in torque and speed offers valuable The waveform in figure-6 is balanced and sinusoidal,
data for identifying the onset of motor faults using AI- indicating healthy motor behaviour[4].
based analysis methods.
Under Fault Condition
3.2 Fault Detection and Diagnosis

 Data Acquisition
Sensors: Sensor is installed on the EV motor to collect data
related to its performance. This sensor can measure
current values of stator winding.

The stator current waveform of a Permanent Magnet

Synchronous Motor (PMSM) provides critical insight into
the motor’s performance and health. In our project, the
motor operates under a normal load for the first 5 seconds
and then experiences a fault, which is clearly reflected in
the current profile[4].

Under Normal Condition Figure-7: Current after stator winding short circuit.

At the 5-second , a stator winding short circuit fault is

introduced. As we can see in figure-7, the stator current
reacts sharply, spiking to extremely high levels between
50 kA and 100 kA. This sudden surge is indicative of a
severe short-circuit fault, where the impedance drops
drastically and the motor draws an abnormally high
current. The waveform becomes distorted, non-sinusoidal,
and highly unstable, often accompanied by harmonic
content and transients. Such high fault currents pose a
significant risk to the motor windings, power electronics,
and overall system stability[4].

 Data Collection
Figure-6-a. The sensors continuously collect data, capturing the
normal and abnormal operating conditions of the motor.

© 2025, IRJET | Impact Factor value: 8.315 | ISO 9001:2008 Certified Journal | Page 735
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN: 2395-0072

To facilitate the development of our AI-based fault  Data Preprocessing

detection system, we implemented a data collection
mechanism using MATLAB Simulink[5].

Figure-10: Importing Libraries and loading dataset.

In this code, we have imported the necessary libraries

Figure-8: Simulation output in MATLAB Workspace required for data analysis, preprocessing, and
visualization. After setting up the environment, we loaded
Key motor parameters such as stator current, speed, and the dataset using the read_csv() function. The dataset,
torque were monitored and logged during simulation. We which appears to be related to diabetes, has been
utilized the ‘To Workspace’ blocks in Simulink to export successfully read into a Data Frame and the first few rows
these signals directly to the MATLAB workspace as shown are displayed using the head() function. As shown in the
in figure-8. This enabled us to capture accurate time- picture, this allows us to get an initial view of the dataset,
domain data under both normal and fault conditions as including its structure and the types of values it contains.
shown in figure-9. The collected data was then used for This step is important for understanding the data before
feature extraction and further processed for training and proceeding to further analysis or model building.
testing machine learning algorithms. This approach
provided a robust dataset necessary for analyzing motor
behaviour and validating the fault detection model.

Figure-9: Data Collection in MATLAB Workspace Figure-11: Checking for missing values.

In this part of the code, we use the df.info() function to

obtain a concise summary of the dataset. This function
provides valuable information such as the total number of
entries, the number of non-null values in each column, the
data types of each feature, and the memory usage of the
DataFrame. It helps in understanding the overall structure
of the dataset and identifying if any columns contain
missing values or incorrect data types. Additionally, by
using df.isnull().sum(), we check for the presence of null
or missing values in each column. This function returns

© 2025, IRJET | Impact Factor value: 8.315 | ISO 9001:2008 Certified Journal | Page 736
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN: 2395-0072

the total number of null values in every column, which is the matrix contains a correlation coefficient ranging from -
crucial for detecting incomplete or inconsistent data that 1 to 1, indicating the strength and direction of the linear
might affect further analysis or model training. Identifying relationship between two variables.
and handling missing data is an important step in the data
preprocessing phase[5]. A value of 1.00 (as seen on the diagonal) indicates perfect
positive correlation of each variable with itself. For
example, Stator_Current and Torque show a strong
positive correlation of 0.93, suggesting that as the torque
increases, the stator current also increases. On the other
hand, Time and Stator_Current have a strong negative
correlation of -0.97, meaning as time progresses, the
stator current tends to decrease. Similarly, Speed and
Torque are negatively correlated at -0.71, implying that an
increase in speed may result in a decrease in torque under
certain conditions.

This type of correlation analysis is crucial in

understanding how motor parameters influence each
other, which can be valuable for fault detection, control
strategies, and system optimization.
Figure-12: Discription of various parameters in dataset
 Model Training and Fault Detection
We have used the df.describe() code to get a descriptive
For model training and fault detection, we employed the
overview of the dataset. This provides a quick summary of
K-Nearest Neighbours (KNN) algorithm using Python. The
key statistical measures for each numerical column, such
dataset used for training was obtained from MATLAB
as the count, mean, standard deviation, minimum and
Simulink simulations of the PMSM motor under both
maximum values, and the quartiles. It helps us understand
normal and fault conditions[6].
the distribution, central tendency, and spread of the data,
which is useful for detecting outliers or unusual patterns. Key feature such as current magnitude were extracted and
In addition to this, by analyzing these statistics, we can labeled accordingly. The KNN model was trained to
also observe potential correlations or relationships classify the motor's operating state based on this feature.
between different variables in the dataset, which can guide During testing, the model demonstrated high accuracy in
further analysis or feature selection in the modeling distinguishing between normal and fault conditions by
process. comparing the input feature to the most similar instances
in the training data. This method proved to be effective
due to KNN's simplicity, interpretability, and ability to
handle non-linear fault patterns present in real-time
motor signals[6].

Figure-13: Correlation between different parameters

Figure-13 represents a correlation matrix visualized as a

heatmap, showing the pairwise correlations between Figure-14: Importing Libraries and dataset loading for
different parameters of a PMSM (Permanent Magnet model training
Synchronous Motor). The variables include Time,
Stator_Current, Voltage, Speed, and Torque. Each cell in

© 2025, IRJET | Impact Factor value: 8.315 | ISO 9001:2008 Certified Journal | Page 737
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN: 2395-0072

By observing the outputs generated from the KNN model,

it is evident that the system is accurately identifying faults
in the PMSM motor based on current data. During testing,
the model was provided with new current inputs
representing both normal and fault conditions. In each
case, the model successfully classified the motor state by
comparing the input features to the nearest patterns in the
training dataset[6].

 Fault Diagnosis
In this project, stator current was used as the sole input
feature to analyze motor faults. Once the model has
Figure-15: User input for prediction classified the data, a decision is made regarding the
severity and location of fault[7].
After training the KNN model using extracted current
features, the system is now capable of detecting motor
conditions based solely on current data. When new real-
time current values are provided whether from simulation
or actual operation the model analyzes the pattern and
magnitude of the input and classifies the motor state as
either normal or faulty. This approach simplifies the
detection process by relying on a single, easily measurable
parameter, while still maintaining reliable fault
identification performance.

Figure-18: Importing libraries and dataset loading for

fault diagnosis

Figure-16: Model output for healthy motor current

Figure-17: Model output for faulty motor current Figure-19: Model training for fault diagnosis

© 2025, IRJET | Impact Factor value: 8.315 | ISO 9001:2008 Certified Journal | Page 738
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN: 2395-0072

 Diagnosed Fault Current:

In the diagnosed scenario, the ML model identified a fault
current of 50 kA, marking the presence of a severe fault
in the system.

 Predicted Fault Location:

The machine learning model successfully predicted the
fault to be located in the stator winding of the PMSM
motor, aligning with the expected fault scenario.

 Predicted Fault Severity:

The fault severity was evaluated and predicted as 100%,
indicating a complete and critical failure in the affected
motor component.

 Location Prediction Accuracy:

The model achieved 100% accuracy in predicting the
Figure-20: Model output indicating accurate fault location fault location, demonstrating the high reliability and
and severity effectiveness of the ML-based diagnostic system.
A Decision Tree Classifier was implemented to identify the
location of the fault within the motor system. By learning 5. CONCLUSION
patterns in the current signal, the model classifies which
Based on the simulation results, the project successfully
section or component is likely affected[8].
demonstrates the effectiveness of machine learning
To assess the severity of the fault, a Random Forest techniques for fault detection and diagnosis in Permanent
Regressor was employed. It predicts a continuous value Magnet Synchronous Motor. The ML model accurately
representing how severe the fault is, based on variations identified abnormal operating conditions, diagnosing a
in the stator current. The ensemble nature of the Random fault current of 50 kA, which significantly exceeded the
Forest improves prediction accuracy and robustness by normal range of 14 A to 17 A. It precisely predicted the
averaging the outputs of multiple decision trees[9]. fault location in the stator winding with 100% accuracy
and assessed the fault severity as 100%, highlighting the
Using just the stator current signal, this approach offers a model’s reliability and sensitivity to critical motor failures.
non-invasive and efficient method for fault detection,
localization, and severity estimation in electric motors. These outcomes confirm that ML-based methods offer a
powerful alternative to traditional diagnostic techniques
Action to be taken by providing faster, more accurate, and automated fault
identification. The simulation-based approach proved to
Appropriate actions can be taken based on the fault be safe and efficient for validating diagnostic algorithms,
diagnosis, such as initiating maintenance procedures, making it a valuable tool for predictive maintenance and
reducing motor power, or activating safety systems[10]. enhancing the operational safety of electric drive systems.

4. RESULTS REFERENCES

 Normal Operating Current Range: [1] Wangjie Lang, , Yihua Hu , Chao Gong , Xiaotian Zhang ,
During normal operation, the PMSM motor draws a Hui Xu, and Jiamei Deng, “Artificial Intelligence-Based
current in the range of 14 A to 17 A, indicating stable and Technique for Fault Detection and Diagnosis of EV
fault-free conditions. Motors: A Review”, IEEE transactions on
transportation electrification, vol. 8, no. 1, march
 Fault Condition Current Range: 2022.
Under fault conditions, the motor exhibits an abnormal
[2] H. Henao et al., “Trends in fault diagnosis for electrical
current range of 30 kA to 100 kA, significantly deviating
machines:A review of diagnostic techniques,” IEEE
from the normal range, which clearly signals the presence
Ind. Electron. Mag., vol. 8, no. 2, pp. 31–42, Jun. 2014.
of a fault.
[3] S. S. S. R. S. Duvvuri, “Rotor inter-turn short-circuit
fault diagnostics relevant modeling of slip-ring

© 2025, IRJET | Impact Factor value: 8.315 | ISO 9001:2008 Certified Journal | Page 739
 International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN: 2395-0072

induction motor,” in Proc. IEEE 8th Power India Int. Mayuri V. Rajput, UG Student,
Conf. (PIICON), Dec. 2018, pp. 1–6. Department of Electrical Engineering,
Government College of Engineering
[4] R. Natarajan, “Failure identification of induction Aurangabad, Chhatrapati
motors by sensing unbalanced stator currents,” IEEE Sambhajinagar, Maharashtra, India.
Power Eng. Rev., vol. 9, no. 12, pp. 35–36, Dec. 1989.

[5] R. K. Patel and V. K. Giri, “Feature selection and

classification of mechanical fault of an induction
motor using random forest classifier,” Perspect. Sci.,
vol. 8, pp. 334–337, Sep. 2016.

[6] E. H. El Bouchikhi, V. Choqueuse, and M. Benbouzid,

“Induction machine faults detection using stator
current parametric spectral estimation,” Mech. Syst.
Signal Process., vols. 52–53, pp. 447–464, Feb. 2015.

[7] D. T. Hoang and H. J. Kang, “A motor current signal-

based bearing fault diagnosis using deep learning and
information fusion,” IEEE Trans. Instrum. Meas., vol.
69, no. 6, pp. 3325–3333, Jun. 2020.

[8] K. A. Loparo, M. L. Adams, W. Lin, M. F. Abdel-Magied,

and N. Afshari, “Fault detection and diagnosis of
rotating machinery,” IEEE Trans. Ind. Electron., vol.
47, no. 5, pp. 1005–1014, Oct. 2000.

[9] Q. Yao, J. Wang, L. Yang, H. Su, and G. Zhang, “A fault

diagnosis method of engine rotor based on random
forests,” in Proc. IEEE Int. Conf. Prognostics Health
Manage. (ICPHM), Jun. 2016, p. 3.

[10] Q. Hu, X.-S. Si, A.-S. Qin, Y.-R. Lv, and Q.-H. Zhang,
“Machinery fault diagnosis scheme using redefined
dimensionless indicators and mRMR feature
selection,” IEEE Access, vol. 8, pp. 40313–40326,2020.

BIOGRAPHIES

Dr Sandhya S. Kulkarni, Professor,

Department of Electrical Engineering,
Government College of Engineering
Aurangabad, Chhatrapati
Sambhajinagar, Maharashtra, India.

Aniket M. Lakhe, UG Student,

Department of Electrical Engineering,
Government College of Engineering
Aurangabad, Chhatrapati
Sambhajinagar, Maharashtra, India.

© 2025, IRJET | Impact Factor value: 8.315 | ISO 9001:2008 Certified Journal | Page 740