INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

Integrated Signal Processing and Machine Learning Model for

Accurate Power System Fault Detection

Lalit Kumar, Arvind Kumar, Sharad Kumar, Vikas Sharma

School of Engineering & Technology, Shri Venkateshwara University, Gajraula, U.P. India

Department of Computer Applications, SRM Institute of Science and Technology, Delhi NCR Campus,

Ghaziabad, U.P. India

DOI: https://doi.org/10.51583/IJLTEMAS.2025.1411000006

Received: 14 November 2025; Accepted: 24 November 2025; Published: 01 December 2025

ABSTRACT

Accurate and timely fault detection in power systems is essential to ensure system reliability, minimize

equipment damage, and maintain continuous power delivery. This paper proposes an integrated signal

processing and machine learning (ML) model designed to enhance the precision and speed of power system

fault diagnosis. The approach leverages advanced signal processing techniques—such as Wavelet Transform

and Fast Fourier Transform (FFT)—to extract critical time–frequency features from voltage and current

signals under various fault conditions. These extracted features are then fed into optimized machine learning

classifiers, including Support Vector Machines (SVM), Random Forest (RF), and Deep Neural Networks

(DNN), to accurately identify fault types, locations, and severities. The integration of signal processing with

ML significantly improves fault detection accuracy compared to conventional methods, particularly under

noisy and dynamic operating conditions. Simulation results on standard IEEE test systems demonstrate the

model’s robustness and scalability, achieving high accuracy and reduced computational latency. The proposed

hybrid framework provides a reliable diagnostic tool for real-time monitoring and intelligent decision-making

in modern power grids. Future work will focus on extending the model for predictive maintenance and

integration with smart grid communication infrastructures.

Keywords—Power System Fault Detection, Signal Processing, Machine Learning, Wavelet Transform, Fast

Fourier Transform (FFT), Fault Classification, Smart Grid, Predictive Maintenance, Real-Time Monitoring.

INTRODUCTION

The reliability and stability of power systems are of paramount importance in ensuring uninterrupted electricity

supply to consumers and industries. However, power system faults—such as short circuits, open circuits, and

line-to-ground faults—pose significant threats to the continuous and secure operation of electrical networks.

These faults can result in severe equipment damage, power outages, and economic losses if not detected and

isolated promptly. Therefore, the development of accurate, fast, and intelligent fault detection and

classification systems has become a critical research area in modern power engineering. Traditional fault

detection techniques, which rely heavily on manual observation or rule-based protection schemes, often

struggle to cope with the increasing complexity of today’s power grids characterized by renewable energy

integration, dynamic load variations, and distributed generation. This growing complexity calls for more

adaptive and intelligent fault diagnosis methods capable of handling nonlinearity, noise, and uncertainty in

system signals. Signal processing has long been an essential tool for analyzing transient disturbances in

electrical signals caused by faults. Techniques such as the Fast Fourier Transform (FFT), Short-Time Fourier

Transform (STFT), and Wavelet Transform (WT) have been employed to decompose voltage and current

waveforms into time-frequency components, allowing for precise identification of fault-induced disturbances.

Among these, Wavelet Transform has proven particularly effective due to its ability to localize transient events

both in time and frequency domains. By extracting relevant features such as energy coefficients, spectral

content, and harmonic components, signal processing lays the groundwork for intelligent analysis and

classification of fault types. However, while these methods offer valuable insights, they often require expert

www.ijltemas.in

Page 71

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

knowledge for interpretation and lack the capability to adapt autonomously to new fault scenarios. In recent

years, Machine Learning (ML) has emerged as a powerful paradigm for pattern recognition and data-driven

decision-making in power systems. ML algorithms such as Support Vector Machines (SVM), Decision Trees,

Random Forests (RF), and Artificial Neural Networks (ANN) have demonstrated remarkable capabilities in

classifying complex nonlinear relationships between signal features and fault conditions. By learning from

historical fault data, these algorithms can generalize and accurately predict unseen fault patterns. When

combined with robust feature extraction through signal processing, ML-based models can achieve superior

performance in fault detection, classification, and localization. This integration not only improves accuracy but

also enhances system adaptability under diverse and uncertain operating environments. The proposed study

introduces an integrated framework that combines advanced signal processing techniques with optimized ML

algorithms for precise and rapid power system fault detection shown in Fig. 1. In this approach, time-

frequency features are extracted using Wavelet Transform and FFT from measured current and voltage signals.

These features serve as the input to machine learning classifiers—such as SVM, Random Forest, and Deep

Neural Networks—to accurately determine the type and location of faults. The integrated model leverages the

strengths of both domains: the analytical rigor of signal processing and the predictive intelligence of machine

learning.

Fig. 1. Integrated Framework for Power System Fault Detection

Through simulation studies on IEEE standard test systems, the proposed framework demonstrates high fault

detection accuracy, computational efficiency, and robustness under noisy conditions. The study aims to

contribute to the advancement of intelligent, data-driven monitoring systems that enhance the reliability,

resilience, and self-healing capabilities of modern smart grids.

LITERATURE REVIEW

Fault detection and classification in power systems have gained significant attention in recent years due to the

increasing complexity of modern grids and the integration of renewable energy sources. Harangaonkar et al. [1]

proposed a deep learning approach for fault detection in a 5-bus system, demonstrating that neural networks can

effectively classify different fault types with high accuracy, highlighting the potential of data-driven techniques

in small-scale systems. Building on this, Dejagah et al. [2] explored stacking-based machine learning models for

transmission line fault detection, emphasizing the improvement in classification performance when multiple ML

models are combined to leverage their complementary strengths. Similarly, Lin et al. [3] introduced a support

vector machine-based method using feature fusion for arc fault diagnosis, showing that integrating multiple

signal features enhances detection reliability and reduces false alarms. In addition to electrical signals, signal

analysis techniques have been extended to mechanical systems. Jiang et al. [4] investigated vibration signal

analysis for mechanical equipment fault diagnosis, demonstrating that feature extraction from time-frequency

domains can accurately predict failures, a concept that has been adapted in electrical fault detection frameworks.

Godhade et al. [5] presented a hybrid transformer-CNN and graph neural network model for real-time fault

www.ijltemas.in

Page 72

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

localization, highlighting the importance of combining spatial and temporal data representations for enhanced

fault detection accuracy. Beyond electrical systems, Sharma, and Kumar [6] and Vikas et al. [7] explored AI-

based security solutions, showing that hybrid deep learning models can detect anomalies in sensor networks, a

principle transferable to fault detection in smart grids where IoT devices are deployed. Industrial process fault

detection has also benefited from real-time intelligent approaches. Attouri et al. [8] proposed a real-time fault

diagnosis method for the Tennessee Eastman chemical process, illustrating that timely detection and

classification are crucial in dynamic environments with multiple interacting variables. Graph-based

optimization techniques for intrusion and anomaly detection in dynamic networks [9] further underscore the

relevance of advanced machine learning and graph modeling in handling complex system interactions, which

parallels the challenges in modern power systems. Saleh et al. [10] reviewed deep learning applications in

induction machine fault detection, summarizing methods that combine feature extraction and ML for robust

fault classification. Recent advances have also explored hybrid and fuzzy approaches. Sadik [11] analysed

boosting algorithms with Fiber Bragg Grating sensor signals, highlighting the role of ensemble learning in fault

classification. Habelalmateen et al. [12] integrated fuzzy c-means clustering with support vector regression for

calibration and fault detection, demonstrating the efficacy of hybrid soft computing methods. Mantha et al. [13]

developed an HIL testbed-based framework for automatic feature extraction and data generation, facilitating

ML/DL-based anomaly detection in power systems. Raja et al. [14] introduced a neuro-fuzzy approach for

predicting broken rotor bar faults via current spectrum analysis, reinforcing the utility of hybrid intelligent

models for complex fault scenarios. Further, security and system reliability considerations in networked

environments have implications for fault detection. Comprehensive studies on mobile ad hoc networks [15]

indicate that robust fault detection frameworks must account for dynamic system conditions and potential

cyber-physical threats. Finally, Lu et al. [16] applied weighted support vector machines for electrical fault

diagnosis, demonstrating that feature weighting improves classification performance and reduces

misclassification rates in complex power networks.

PROPOSED METHODOLOGY

The proposed methodology presents an integrated framework that combines advanced signal processing

techniques with machine learning algorithms to achieve accurate and efficient fault detection in power

systems. The methodology is designed to detect, classify, and locate different types of faults under varying

operating conditions while maintaining high reliability and computational efficiency. The overall process can

be divided into five major stages: data acquisition, signal preprocessing, feature extraction, machine learning-

based classification, and performance evaluation.

1. Data Acquisition: The first stage of the proposed methodology involves the acquisition of voltage and

current signals from different nodes or buses of the power system using sensors or Phasor Measurement Units

(PMUs). These signals capture transient disturbances caused by various fault events, including line-to-line,

line-to-ground, double line-to-ground, and three-phase faults. To ensure the robustness of the proposed

framework, fault data is simulated using standard IEEE test systems such as IEEE 14-bus, 30-bus, and 57-bus,

representing realistic operational scenarios under diverse loading conditions. This stage forms the foundation

for the methodology, providing the raw data necessary for subsequent signal analysis and feature extraction.

2. Signal Preprocessing: In the signal preprocessing stage, the acquired voltage and current signals are

cleaned and prepared for analysis. Real-time power system data often contain noise due to switching

transients, electromagnetic interference, or environmental factors, which can compromise the accuracy of fault

detection. Filtering techniques are applied to suppress unwanted noise, and normalization is performed to

maintain consistency across datasets. Preprocessing ensures that the extracted features in the subsequent stage

accurately reflect fault characteristics rather than measurement artifacts, thereby improving the reliability and

stability of the fault detection process.

3. Feature Extraction using Signal Processing: The feature extraction stage employs advanced signal

processing techniques, including the Wavelet Transform (WT) and Fast Fourier Transform (FFT), to capture

critical information from the pre-processed voltage and current signals. FFT is used to analyze the frequency

spectrum and identify harmonic distortions resulting from faults, while WT provides time-frequency

www.ijltemas.in

Page 73

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

localization, enabling the detection of transient events. The decomposition of signals into multiple frequency

bands through WT allows for the extraction of features such as energy coefficients, standard deviation, and

entropy at different levels. These features provide a comprehensive representation of both stationary and non-

stationary characteristics of faults, forming an informative dataset for machine learning classification.

4. Machine Learning-Based Classification: In this stage, the extracted features are input into

various machine learning algorithms for fault classification. The study utilizes Support Vector Machine

(SVM), Random Forest (RF), and Deep Neural Networks (DNN) to identify the type and location of faults.

SVM efficiently handles nonlinear decision boundaries, while RF improves generalization by aggregating

multiple decision trees. DNN captures complex interactions between features through its deep hierarchical

structure, enhancing classification accuracy. The models are trained using labelled datasets representing

different fault types and validated with cross-validation techniques to optimize parameters and prevent

overfitting. The trained classifiers can then accurately detect and categorize faults in real-time scenarios.

5. Performance Evaluation: The final stage involves evaluating the performance of the integrated framework

using metrics such as accuracy, precision, recall, F1-score, and computational latency. Comparative studies

with traditional fault detection methods highlight the superiority of the proposed approach in terms of

detection accuracy, robustness under noisy conditions, and real-time applicability. The methodology is

scalable to larger grid systems and adaptable for implementation in smart grid infrastructures. Future

enhancements may include integrating predictive maintenance capabilities and intelligent communication with

grid management systems to enable autonomous fault diagnosis and decision-making.

RESULT & ANALYSIS

The proposed integrated signal processing and machine learning framework was tested on standard IEEE bus

systems, including the 14-bus, 30-bus, and 57-bus networks, to evaluate its performance in fault detection,

classification, and localization. Simulated fault scenarios included single line-to-ground (SLG), line-to-line

(LL), double line-to-ground (DLG), and three-phase (3Φ) faults under various loading and fault inception

angles. The performance of the model was assessed using common metrics, including accuracy, precision,

recall, F1-score, and computational time (latency). The experiments also compared the proposed hybrid

method with conventional fault detection approaches to highlight the improvements in detection reliability and

efficiency.

1. Classification Performance: The classification performance of the proposed framework using SVM,

Random Forest, and Deep Neural Networks is summarized in Table I. Among the tested algorithms, the Deep

Neural Network achieved the highest classification accuracy, indicating its capability to capture complex

nonlinear relationships among extracted signal features.

Fault Classification Performance of ML Models

Model

Classification Accuracy

(%)

Remarks

SVM

Random Forest

93.12%

95.84%

Effective for moderate nonlinear patterns

Strong performance due to ensemble feature

learning

Deep Neural Network 97.96%

Highest accuracy; best at capturing complex

nonlinear relationships

The results show that Random Forest and DNN outperform SVM in terms of overall classification metrics.

The superior performance of DNN can be attributed to its hierarchical feature learning, which effectively

captures subtle variations in the time-frequency features extracted from signal processing.

www.ijltemas.in

Page 74

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

2. Fault Detection Time Analysis: The computational efficiency of the proposed hybrid method was

evaluated by comparing the fault detection time with traditional methods such as the Continuation Power Flow

(CPF) method. The results are presented in Table II.

Fault Detection Time Comparison

Test System

Traditional CPF

Proposed Hybrid

Improvement (%)

Detection Time (ms) Method Detection Time

(ms)

IEEE 14-Bus

42.8 ms

55.4 ms

68.2 ms

30.1 ms

34.8 ms

40.6 ms

29.67%

37.17%

40.47%

IEEE 30-Bus

IEEE 57-Bus

Fig. 2. compares the fault detection time for three IEEE test systems (14-Bus, 30-Bus, and 57-Bus). Each

system shows three bars: traditional CPF-based detection time, hybrid method detection time, and

improvement percentage. The hybrid method consistently shows significantly lower detection time, with

improvements ranging from approximately 29% to 40%.

Fig. 2. Comparison of Fault Detection Time Between Traditional CPF and Hybrid Method

3. Robustness Under Noise: To test the robustness of the methodology, Gaussian noise with varying signal-

to-noise ratios (SNR) was added to the test signals. The classification accuracy of the DNN model remained

above 95% for SNR levels as low as 20 dB, demonstrating that the integration of signal processing features

with machine learning enhances noise resilience.

Classification Accuracy of DNN under Different Noise Levels

SNR Level (dB)

Classification Accuracy

(%)

Remarks

40 dB

30 dB

20 dB

98.72%

97.85%

95.43%

Very low noise; optimal performance

Minor noise impact

Accuracy remains above 95%; strong noise

resilience

www.ijltemas.in

Page 75

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

10 dB

0 dB

92.18%

Noticeable noise; performance degradation

starts

88.67%

High noise; significant distortion in features

TABLE III. shows that even at low SNR levels, the DNN model maintains high accuracy, demonstrating the

robustness of the proposed hybrid framework against noisy measurements. This validates the effectiveness of

combining signal processing-based feature extraction with machine learning for real-world fault detection

scenarios. Fig. 3. visualizes the classification accuracy of a Deep Neural Network (DNN) at various SNR

levels (40 dB, 30 dB, 20 dB, 10 dB, and 0 dB). The accuracy decreases gradually as noise increases, ranging

from approximately 98.7% at 40 dB to around 88.7% at 0 dB. The DNN maintains above 95% accuracy at

SNR levels down to 20 dB, demonstrating robustness to noise.

Fig. 3. DNN Classification Accuracy Under Different SNR Levels

CONCLUSION

This paper presents an integrated framework combining signal processing techniques and machine learning

algorithms for accurate and efficient fault detection in power systems. By employing Wavelet Transform and

FFT for feature extraction and advanced classifiers such as SVM, Random Forest, and Deep Neural Networks,

the proposed methodology achieves high accuracy, robustness against noise, and significantly reduced

detection time compared to conventional methods. Simulation results on IEEE 14-bus, 30-bus, and 57-bus

systems demonstrate the framework’s effectiveness in classifying fault types and locating fault points under

diverse operating conditions. The study highlights the potential of hybrid signal processing and machine

learning models for real-time monitoring and intelligent decision-making in modern power grids. Future work

may focus on integrating the proposed system with predictive maintenance strategies, smart grid

communication networks, and IoT-enabled sensors, enabling autonomous fault diagnosis, early warning

systems, and enhanced grid resilience in increasingly complex and renewable-integrated power systems.

REFERENCES

1. P. V. Harangaonkar, S. Umathe, P. Daigavane, S. Naik and K. Pathe, "A Deep Learning Approach for

Fault Detection and Classification in 5 Bus System," 2025 12th International Conference on Emerging

Trends in Engineering & Technology - Signal and Information Processing (ICETET - SIP), Nagpur,

India, 2025, pp. 1-6, doi: 10.1109/ICETETSIP64213.2025.11156315.

2. M. Dejagah, H. S. Daliri, N. J. Marand, H. R. Baghaee, H. Nafisi and H. A. Abyaneh, "Stacking ML-

based Fault Detection and Classification in Transmission Lines," 2025 International Conference on

www.ijltemas.in

Page 76

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XIV, Issue XI, November 2025

Protection and Automation of Power Systems (IPAPS), Tehran, Iran, Islamic Republic of, 2025, pp. 1-

5, doi: 10.1109/IPAPS65933.2025.11092221.

3. Z. Lin, B. Luo, W. Cai and G. Liu, "Based on feature fusion and arc fault diagnosis method of support

vector machine," 2024 4th International Conference on Intelligent Power and Systems (ICIPS),

Yichang, China, 2024, pp. 412-415, doi: 10.1109/ICIPS64173.2024.10900011.

4. W. Jiang, R. Zhou and H. Jia, "Research on Mechanical Equipment Fault Diagnosis and Prediction

Technology Based on Vibration Signal Analysis," 2024 International Conference on Power, Electrical

Engineering, Electronics and Control (PEEEC), Athens, Greece, 2024, pp. 494-498, doi:

10.1109/PEEEC63877.2024.00096.

5. S. Godhade, P. Singh and J. Kumar, "Design of an improved model for real-time fault detection and

localization using hybrid transformer-CNN and graph neural networks," 2025 Third International

Conference on Microwave, Antenna and Communication (MAC), Bhopal, India, 2025, pp. 1-6, doi:

10.1109/MAC64480.2025.11139887.

6. V. Sharma and S. Kumar, "Role of Artificial Intelligence (AI) to Enhance the Security and Privacy of

Data in Smart Cities," 2023 3rd International Conference on Advance Computing and Innovative

Technologies in Engineering (ICACITE), Greater Noida, India, 2023, pp. 596-599, doi:

10.1109/ICACITE57410.2023.10182455.

7. Vikas, R. P. Daund, D. Kumar, P. Charan, R. S. K. Ingilela and R. Rastogi, "Intrusion Detection in

Wireless Sensor Networks using Hybrid Deep Belief Networks and Harris Hawks Optimizer," 2023 4th

International Conference on Electronics and Sustainable Communication Systems (ICESC),

Coimbatore, India, 2023, pp. 1631-1636, doi: 10.1109/ICESC57686.2023.10193270.

8. K. Attouri, M. Mansouri, M. Hajji, A. Kouadri, K. Bouzrara and H. Nounou, "Real-Time Fault

Detection and Diagnosis Method for Industrial Chemical Tennessee Eastman Process," 2024 10th

International Conference on Control, Decision, and Information Technologies (CoDIT), Vallette,

Malta, 2024, pp. 3009-3014, doi: 10.1109/CoDIT62066.2024.10708622.

9. Optimization of Graph Neural Networks for Real-Time Intrusion Detection in Dynamic Mobile Ad-

Hoc Networks”, Int. J. Environ. Sci., vol. 11, no. 11s, pp. 740–748, Jun. 2025, doi:

10.64252/79452g17.

10. M. A. Saleh, S. S. Refaat and J. Kammermann, "Leveraging Deep Learning for Fault Detection and

Classification of Induction Machines: A Review," IECON 2024 - 50th Annual Conference of the IEEE

Industrial

Electronics

Society,

Chicago,

IL,

USA,

2024,

pp.

1-8,

doi:

10.1109/IECON55916.2024.10905883.

11. S. A. Sadik, "Comparative Analysis of Boosting Algorithms for Contactor Fault Classification from

Fiber Bragg Grating Sensor Signals," 2024 47th International Conference on Telecommunications and

Signal

Processing

(TSP),

Prague,

Czech

Republic,

2024,

pp.

187-190,

doi:

10.1109/TSP63128.2024.10605955.

12. M. I. Habelalmateen, S. R. Kasarla, J. Sravanthi, P. Kavitha and S. M. Sundaram, "A Fuzzy C-Means

With Support Vector Regression for Calibration and Fault Detection in Modern Instruments," 2024 4th

International Conference on Mobile Networks and Wireless Communications (ICMNWC), Tumkuru,

India, 2024, pp. 1-5, doi: 10.1109/ICMNWC63764.2024.10872349.

13. A. A. Mantha, A. Hussain and G. Ravikumar, "HIL Testbed-based Auto Feature Extraction and Data

Generation Framework for ML/DL-based Anomaly Detection and Classification," 2024 IEEE Power &

Energy Society Innovative Smart Grid Technologies Conference (ISGT), Washington, DC, USA, 2024,

pp. 1-5, doi: 10.1109/ISGT59692.2024.10454202.

14. H. A. Raja, T. Vaimann, B. Asad, A. Kallaste and M. U. Sardar, "A Neuro-Fuzzy Approach for Broken

Rotor Bar Fault Prediction in Induction Motors via Current Spectrum Analysis," 2025 IEEE

Symposium on Diagnostics for Electric Machines, Power Electronics and Drives (SDEMPED), Dallas,

TX, USA, 2025, pp. 01-07, doi: 10.1109/SDEMPED53223.2025.11154194.

15. A Comprehensive Analysis of Security Mechanisms and Threat Characterization in Mobile Ad Hoc

Networks”, IJLTEMAS,

vol.

14,

no.

pp.

732–737,

Jun.

2025, doi:

10.51583/IJLTEMAS.2025.140500079.

16. W. Lu, Q. Sun and A. Li, "Electrical Fault Diagnosis Using Weighted Support Vector Machine," 2024

First International Conference on Software, Systems and Information Technology (SSITCON),

Tumkur, India, 2024, pp. 1-5, doi: 10.1109/SSITCON62437.2024.10796619.

www.ijltemas.in

Page 77