Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1226

Development of a Predictive Maintenance

Algorithm for a Diesel Generator using
Machine Learning
Olokede, Oluwagbemiga1; Evans Ashigwuike2
1,2
Department of Electrical Engineering, University of Abuja

Publication Date: 2025/03/29

Abstract: This study develops a predictive maintenance framework for a 500kVA diesel generator using advanced machine
learning techniques, aiming to enhance reliability and operational efficiency. The research involves the collection of real-
world operational data at one-minute intervals over two months, focusing on critical parameters such as bearing
temperature, engine vibration, and coolant temperature. Two machine learning models—XGBoost and Multi-Layer
Perceptron (MLP)—were trained to classify generator conditions into distinct maintenance categories with high accuracy.
A meta-learning ensemble approach was implemented, integrating the predictions from these models to leverage their
complementary strengths and enhance robustness. The results demonstrate exceptional performance, with both individual
and ensemble models achieving precision, recall, and F1-scores near 1.00 across multiple fault scenarios. The meta-learning
framework proved particularly effective, showcasing improved reliability over standalone models. This study’s
contributions are twofold: it advances the state of predictive maintenance by employing hybrid modelling techniques and
addresses a critical gap in the proactive management of high-capacity diesel generators. The research underscores the
practical applicability of machine learning in industrial contexts, offering a scalable and sustainable solution to minimise
downtime, reduce maintenance costs, and optimise equipment longevity. By integrating robust data analysis with cutting-
edge machine learning, this framework establishes a foundation for proactive, data-driven maintenance strategies in
industrial settings, aligning with the broader goals of Industry 4.0 and sustainable industrial practices.

How to Cite: Olokede, Oluwagbemiga; Evans Ashigwuike. (2025). Development of a Predictive Maintenance Algorithm for a
Diesel Generator using Machine Learning. International Journal of Innovative Science and Research Technology,
10(3), 1417-1427. https://doi.org/10.38124/ijisrt/25mar1226.

I. INTRODUCTION networks, and ensemble learning have further elevated the

accuracy and robustness of PdM systems, surpassing
Predictive maintenance (PdM) has emerged as a vital traditional statistical methods [4].
component of Industry 4.0, significantly transforming
traditional maintenance strategies. By leveraging advanced Explainable artificial intelligence (XAI) is becoming
analytics and machine learning (ML), PdM predicts increasingly relevant in PdM to ensure that insights generated
equipment failures before they occur, minimising downtime by ML models are interpretable and actionable. This has
and enhancing operational efficiency. Unlike reactive and catalysed trust and adoption in sectors such as manufacturing,
scheduled maintenance, which either wait for failures or aerospace, and energy, where safety and reliability are
follow predetermined schedules, PdM integrates real-time paramount. Furthermore, PdM strategies are aligning with
data, historical trends, and computational models to sustainability goals, as predictive capabilities reduce energy
recommend maintenance only when necessary. This consumption and material wastage, aligning with green
approach not only reduces costs but also extends the lifespan manufacturing initiatives [5], [6].
of equipment by avoiding unnecessary interventions [1], [2].
Despite its advantages, challenges persist, including the
The advent of IoT and the proliferation of data heterogeneity of datasets, scalability across diverse
collection technologies have been pivotal in driving PdM industries, and integration with legacy systems. Continuous
adoption. Sensors embedded in industrial systems provide research is addressing these gaps, with novel frameworks
real-time monitoring of critical parameters such as incorporating product quality parameters and multivariate
temperature, vibration, and pressure, enabling high statistical models, making PdM a versatile and adaptive
granularity in data acquisition [3]. The integration of IoT with solution for complex industrial environments [7].
ML allows for seamless data processing and enhanced
decision-making. Recent advancements in XGBoost, neural

IJISRT25MAR1226 www.ijisrt.com 1417

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1226

In this study, we aim to develop a predictive [12] explored the development of a microgrid control
maintenance algorithm for a diesel generator using machine system leveraging deep reinforcement learning techniques.
learning techniques. Machine learning models, specifically The research validates the effectiveness of reinforcement
XGBoost and Multi-Layer Perceptron (MLP), were learning in optimising control decisions in systems that
developed and trained to classify the generator’s condition include a diesel generator. Despite its robust control
into distinct maintenance categories with high precision and framework, the study does not directly address predictive
recall. Additionally, a meta-learning ensemble approach was maintenance analytics for fault detection or preventive
implemented to integrate the predictions of these models, maintenance classification.
capitalising on their complementary strengths to enhance
robustness. Key steps included data preprocessing to handle [13] proposed a machine learning approach to forecast
missing values and noise, feature selection guided by a capacitor bank requirements for improving grid efficiency.
correlation heatmap, and iterative model optimisation. Model While this research underscores the potential of machine
performance was evaluated using precision, recall, F1-score, learning in real-time decision-making and energy
and accuracy metrics to ensure reliability across diverse fault optimisation, its focus is more aligned with grid management
scenarios. than predictive maintenance strategies.

The growing complexity of industrial systems, [14] present a Maximum Power Point Tracking (MPPT)
including high-capacity diesel generators, necessitates algorithm that integrates real-time analytics for enhanced
advanced maintenance strategies to prevent unexpected control of industrial power systems. The study highlights the
failures and optimise resource utilisation. Traditional role of predictive analytics in improving system performance
maintenance approaches are increasingly inadequate in but is more focused on system control than on predictive
addressing the dynamic and multifaceted nature of modern maintenance for diesel generators.
equipment. This study addresses these gaps by integrating
cutting-edge machine learning techniques into a predictive [15] investigate the optimisation of post-disaster
maintenance framework. By using ensemble learning, it microgrid control using multi-agent deep reinforcement
aligns with best practices recommended in recent literature, learning. Their findings demonstrate the predictive
which highlight the efficacy of hybrid models in improving capabilities of reinforcement learning algorithms in dynamic
prediction accuracy and reducing false positives [8], [9]. environments. However, the primary focus is on control
response strategies rather than routine maintenance
II. LITERATURE REVIEW prediction.

The empirical review draws from several relevant [16] conduct a ferrographic study of wear particles in
publications to contextualise the development of a predictive used oil from power generation machinery, contributing to
maintenance algorithm for a 500kVA diesel generator using operational parameter monitoring. Although this approach
machine learning techniques. A consistent theme across the provides valuable insights into wear and fault detection, it
reviewed literature is the pursuit of more effective and does not incorporate advanced machine learning methods for
efficient maintenance strategies through the intelligent predictive maintenance.
analysis of operational data. In exploring optimal power
distribution, [10] investigates the integration of distributed [17] examines islanding detection using distributed
diesel generators into power systems in Iraq, with a focus on generator systems and an artificial bee colony algorithm.
addressing operational challenges and developing algorithms While the study offers important insights into system
for optimal power distribution. While the research provides monitoring and health assessment, it diverges from the
valuable insights into the integration and operational development of predictive maintenance strategies using
efficiency of diesel generators, it lacks a direct focus on machine learning techniques.
predictive maintenance, limiting its applicability for fault
detection and classification strategies essential for [18] explore hybrid renewable energy systems
maintenance optimisation. integrating photovoltaics, wind turbines, diesel engines, and
batteries. The study focuses on system optimisation for rural
[11] examined load forecasting at a microgrid level electrification but lacks emphasis on predictive maintenance
using machine learning algorithms to optimise grid load analytics for diesel generators.
management. The study demonstrates the efficacy of
predictive techniques in managing operational states, [19] apply reinforcement learning to optimise the tuning
providing a foundation for forecasting methodologies. of grid-connected inverter controllers in microgrids.
However, its emphasis on load balancing rather than Although the research demonstrates the predictive potential
maintenance prediction diverges from the specific objectives of machine learning, its focus remains on inverter control
of maintenance category classification. rather than predictive maintenance classification for
generator conditions.

IJISRT25MAR1226 www.ijisrt.com 1418

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1226

The reviewed works reveal significant advancements in

applying machine learning for load forecasting, energy
optimisation, and control strategies within energy systems.
However, a gap emerges in the specific development of
predictive maintenance frameworks tailored for large diesel
generators. Most studies, including those by [10], [11],
prioritise energy management and system control over fault
detection and maintenance strategies. Furthermore, while
methodologies such as reinforcement learning [12] and fuzzy
logic systems [14] offer robust decision-making capabilities,
their application to predictive maintenance for diesel
generators remains limited or absent.

The current research aims to address these gaps by

developing a comprehensive predictive maintenance
algorithm that integrates operational parameter monitoring
and fault scenario characterisation with advanced machine
learning techniques. By leveraging models such as XGBoost
and Multi-Layer Perceptron Neural Networks, combined into
a meta-learning ensemble framework, this research will
enhance predictive accuracy and robustness. The proactive
classification of generator conditions into distinct
maintenance categories will offer actionable insights,
advancing maintenance strategies beyond what existing
models provide.
Fig 1: Research Methodology Algorithm
III. RESEARCH METHODOLOGY
This design bridges the gap between theoretical
A. Research Design machine learning and practical maintenance needs, creating a
This study used a quantitative approach to develop a system that can turn complex operational data into useful
predictive maintenance system for a 500kVA diesel maintenance recommendations.
generator. The project combined multiple machine learning
models to predict maintenance needs more accurately than B. Data Collection
traditional methods as shown in Figure 1. Data was collected from a diesel generator at a
manufacturing company in Abuja, Nigeria, over two months.
We used two main machine learning techniques: The system recorded measurements every minute to track the
XGBoost and Multi-Layer Perceptron neural networks. generator's performance.
These were chosen to analyse complex equipment data and
identify potential failures. The system collected operational The monitoring system captured several key
data every minute for two months, providing detailed insights parameters: oil contamination and viscosity, bearing
into the generator's performance under various conditions. temperature, engine vibration, mechanical noise, coolant
temperature, heat dissipation, and various electrical readings.
We adopted an ensemble learning approach, combining These measurements provided a comprehensive view of the
different models to overcome individual limitations. This generator's operational state.
method merged the strengths of each model to create more
reliable predictions. The framework was designed to be As shown in Table 3.1, the data was organised into
practical and adaptable to similar industrial equipment. distinct maintenance categories. These included Oil Change
Required (M001), Engine Alignment Adjustment (M010),
Cooling System Maintenance (C001), No Immediate
Maintenance (N000), Short-Circuit Maintenance (S001),
Overload Protection Maintenance (O002), Frequency
Adjustment Maintenance (F003), Voltage Regulation
Maintenance (V004), and Electrical Diagnostics (E005).

High-precision sensors monitored the generator in real

time, recording both normal operations and fault conditions.
Each reading included a timestamp for detailed analysis. We
then processed the data to ensure its quality before using it in
our machine learning models.

IJISRT25MAR1226 www.ijisrt.com 1419

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1226

C. Data Preprocessing and Feature Engineering approach allows the model to predict various maintenance
The first step in developing the predictive maintenance categories accurately.
model involved cleaning the data to remove anomalies and
inconsistencies, as shown in Figure 2. This process included
addressing missing values, removing outliers, and
standardising the format of timestamps and sensor readings
to ensure accurate predictions.

Fig 2: Data Preprocessing and Feature Engineering Block Fig 3: XGBoost Model Algorithm
Diagram
We optimised the model's performance by carefully
After cleaning, the most relevant parameters that selecting key settings, including learning rate, tree depth, and
influence maintenance needs were identified. These boosting rounds, using Grid Search Cross-Validation. To
parameters include bearing temperature, oil viscosity, engine prevent overfitting, we included regularisation parameters.
vibration, and coolant temperature. These parameters were The data was split with 80% for training and 20% for testing.
chosen based on their relationship with specific fault We addressed any imbalance in maintenance categories using
scenarios. We then standardised the numerical values through the scale_pos_weight parameter and employed multi:softmax
scaling and normalisation to ensure all features had equal for handling multiple maintenance categories. The model's
importance during model training. The maintenance accuracy and reliability were thoroughly evaluated through
categories were converted into numerical values to make classification reports.
them compatible with our machine learning algorithms.
E. Multi-Layer Perceptron Neural Network
D. XGBoost Model The Multi-Layer Perceptron (MLP) neural network
This research used XGBoost for maintenance prediction employed a deep learning approach for predictive
because of its effectiveness with tabular data. As shown in maintenance analysis, designed to capture complex
Figure 3, the model uses multiple decision trees, with each relationships within the diesel generator data. The
new tree improving upon the previous ones' results. This architecture balanced model complexity with
generalisability, as shown in Figure 4.

Table 1: Data Categorisation

Category Code Parameter No Fault Cause
Maintenance Scenario
Range Range
Oil Change M001 Oil Contamination (%) 0–3 >5 Aging oil or contamination
Required [20] by debris
Oil Viscosity (cSt) 8–12 <7 Degradation due to heat or
contaminants
Bearing Temperature 80–90 95–110 Heat transfer inefficiency
(°C)
Engine Alignment M010 Engine Vibration 0.5–0.8 >1.2 Misalignment, wear, or
Adjustment [21] (mm/s) imbalance

IJISRT25MAR1226 www.ijisrt.com 1420

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1226

Mechanical Noise (dB) 30–50 >60 Increased wear or loose

components
Cooling System C001 Coolant Temperature 75–85 >95 Radiator blockage, pump
Maintenance [22] (°C) failure
Heat Dissipation (%) >90 <80 Reduced cooling efficiency
No Immediate N000 All Parameters Within Normal Not Normal operating
Maintenance [23] Ranges Applicable conditions
Short-Circuit S001 AC Current (A) 45–55 >70 (Short Cable or winding faults
Maintenance[24] Circuit)
Voltage Fluctuations <3 >5 Sudden electrical failures
(%)
Overload Protection O002 AC Current (A) 45–55 60–70 Excessive load
Maintenance [25] (Overload)
Bearing Temperature 80–90 >95 Overheating due to
(°C) overload
Frequency F003 Output Frequency (Hz) 49.9–50.1 <49.5 or Governor faults or
Adjustment >50.5 unbalanced loads
Maintenance
Voltage Regulation V004 Voltage Fluctuations 0.5–3 >5 Alternator or AVR
Maintenance [26] (%) malfunction
Electrical E005 Current Imbalance (%) <1 >3 Load mismanagement or
Diagnostics [26] aging components
Winding Temperature 30–60 >70 Winding aging, winding
(°C) insulator loss, overloading,
faulty winding cooling
system

Fig 4: MLP Neural Network Architecture

 Network Architecture  Activation Functions

The MLP neural network consisted of three primary The hidden layers employed Rectified Linear Unit
layers: an input layer, two hidden layers, and an output layer. (ReLU) activation functions for computational efficiency and
The input layer accommodated the preprocessed generator gradient optimization. The output layer used a softmax
operational data. The first hidden layer contained 128 neurons activation function for multi-class classification across
for capturing feature interactions, while the second hidden maintenance categories with probabilistic interpretations.
layer used 64 neurons for refining feature representations.
 Training Parameters
 Layer Configuration The network implementation used the Adam optimizer
Each hidden layer transformed the input features with categorical cross-entropy as the loss function. A
progressively. The first hidden layer's 128 neurons served as validation split of 20% enabled performance assessment and
a feature extraction mechanism, identifying patterns across early stopping. Dropout regularization at 20% in both hidden
operational parameters. The subsequent 64-neuron layer layers prevented overfitting. The training process ran for 100
distilled these representations into sophisticated feature epochs with batch sizes of 16, balancing learning
mappings for maintenance prediction. convergence with computational efficiency.

IJISRT25MAR1226 www.ijisrt.com 1421

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1226

F. Meta-Learning Ensemble Hybrid models and highlighted which maintenance categories were
The Meta-Learning Ensemble Hybrid approach difficult to distinguish.
combined multiple machine learning models to enhance
maintenance forecast accuracy. This strategy integrated the The evaluation process included a detailed statistical
XGBoost classifier and MLP neural network into a hybrid summary for each maintenance category through a
model, as illustrated in Figure 1. classification report. This report measured the accuracy of
positive predictions, the model's ability to find relevant cases,
The ensemble used a stacking approach where a meta- and provided a balanced assessment through the F1-score. To
model, typically a logistic regression classifier, combined ensure reliable validation and prevent overfitting, the study
predictions from base models. This integration leveraged employed a five-fold cross-validation technique. This method
XGBoost's strength in handling tabular data and MLP's divided the data into five parts, maintaining the original
capability for non-linear relationships. The prediction fusion distribution of maintenance categories while testing the
technique stacked outputs from both models into a new model's performance across different data combinations.
feature set for the meta-model, generating final predictions
with improved accuracy and reliability. The final analysis compared the performance of three
models: the XGBoost Classifier, Multi-Layer Perceptron
G. Performance Evaluation and Validation Neural Network, and Meta-Learning Ensemble Hybrid
The evaluation process assessed the predictive Model. This comparison revealed each model's strengths and
maintenance algorithm's reliability and effectiveness through limitations in predicting maintenance requirements for the
several key measurements. The main metric used was diesel generator. Through these comprehensive evaluation
classification accuracy, which showed how often the model methods, the study established the reliability and
correctly predicted maintenance needs. The analysis also effectiveness of the predictive maintenance system.
included precision, recall, and F1-scores for each
maintenance category to provide detailed insight into the IV. RESULTS AND DISCUSSION
model's performance.
A. Exploratory Data Analysis
A confusion matrix served as an essential analytical The research commenced with a comprehensive data
tool, comparing predicted maintenance categories against visualisation approach to explore the intricate relationships
actual requirements. This matrix tracked correct predictions, between various operational parameters of the 500kVA diesel
incorrect category assignments, missed predictions, and generator. Figure 5 presents a correlation heatmap that
correct identification of non-maintenance scenarios. These illuminates the interdependencies between key variables
measurements helped identify any biases in the predictive monitored during the study.

Fig 5: Correlation Heatmap Graph

The trend analysis depicted in Figure 6 provides a vibration. This visualisation offers insights into the dynamic
temporal representation of critical parameters, namely behaviour of these essential indicators throughout the
bearing temperature, winding temperature, and engine generator's operational lifecycle.

IJISRT25MAR1226 www.ijisrt.com 1422

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1226

Fig 6: Bearing Temperature, Winding Temperature and Engine Vibration Trend

Figure 7 illustrates the distribution of bearing A comprehensive multivariate analysis is presented in

temperatures, which is crucial for understanding the thermal Figure 8, which showcases a pairplot exploring the
characteristics and potential stress points within the relationships between bearing temperature, engine vibration,
generator's mechanical system. A detailed examination of the oil contamination, and coolant temperature. This
temperature distribution can reveal patterns indicative of visualisation enables a nuanced understanding of the complex
impending mechanical deterioration. interactions among these critical operational parameters.

Fig 7: Bearing Temperature Distribution

IJISRT25MAR1226 www.ijisrt.com 1423

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1226

Fig 8: Pairplot of Bearing Temperature, Engine Vibration, Oil Contamination and Coolant Temperature

B. Model Prediction Results As presented in Table 2, the MLP Neural Network

The predictive maintenance framework evaluation achieved optimal accuracy at 1.0, while both the XGBoost
employed three advanced machine learning approaches: and Meta-Learning Hybrid Model demonstrated exceptional
XGBoost, Multi-Layer Perceptron (MLP) Neural Network, performance at 0.9994.
and a Meta-Learning Hybrid Model. Table 3 presents a
comprehensive classification report detailing the Table 2: Model Performance Report
performance metrics across eight maintenance categories. Model Accuracy
XGBoost 0.9994
The classification results demonstrated exceptional MLP 1.0
predictive capabilities. The XGBoost model achieved perfect Meta-Learning Hybrid Model 0.9994
precision, recall, and F1-scores of 1.0 across seven classes,
with class 5 showing a marginal reduction in recall to 0.99. The confusion matrices illustrated in Figures 9, 10, and
The MLP Neural Network and Meta-Learning Hybrid Model 11 provide visual validation of the models' classification
exhibited comparable performance, achieving perfect or near- performance, demonstrating precise categorisation across
perfect metrics across all categories. maintenance scenarios.

Table 3: Model Classification Report

Class Precision Recall F1-Score Support Model
0 1 1 1 200 XGBoost
1 1 1 1 200 XGBoost
2 1 1 1 200 XGBoost
3 1 1 1 200 XGBoost
4 1 1 1 200 XGBoost
5 1 0.99 1 200 XGBoost
6 1 1 1 200 XGBoost
7 1 1 1 200 XGBoost
0 1 1 1 200 MLP
1 1 1 1 200 MLP
2 1 1 1 200 MLP
3 1 1 1 200 MLP
4 1 1 1 200 MLP
5 1 1 1 200 MLP

IJISRT25MAR1226 www.ijisrt.com 1424

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1226

6 1 1 1 200 MLP
7 1 1 1 200 MLP
0 1 1 1 200 Meta-Learning Hybrid Model
1 1 1 1 200 Meta-Learning Hybrid Model
2 1 1 1 200 Meta-Learning Hybrid Model
3 1 1 1 200 Meta-Learning Hybrid Model
4 1 1 1 200 Meta-Learning Hybrid Model
5 1 0.99 1 200 Meta-Learning Hybrid Model
6 1 1 1 200 Meta-Learning Hybrid Model
7 1 1 1 200 Meta-Learning Hybrid Model

C. Discussion of Results
The research successfully developed an advanced
predictive maintenance framework for a 500kVA diesel
generator through sophisticated machine learning
methodologies. Data visualisation revealed intricate
relationships between operational parameters, with the
correlation heatmap (Figure 4.1) demonstrating complex
interconnections among generator variables. This
visualisation proved essential for understanding the
multidimensional nature of mechanical system behaviour.

Trend analysis of bearing temperature, winding

temperature, and engine vibration (Figure 4.2) revealed
critical temporal variations, indicating that continuous
monitoring provides early indicators of mechanical stress.
Fig 9: XGBoost Confusion Matrix
The bearing temperature distribution (Figure 4.3) further
validated this approach by identifying thermal characteristics
indicative of emerging mechanical anomalies.

The machine learning models demonstrated exceptional

performance. The XGBoost model achieved 0.9994
accuracy, with near-perfect precision across maintenance
categories. The MLP Neural Network achieved optimal
accuracy at 1.0, while the Meta-Learning Hybrid Model
matched XGBoost's performance. These results validate the
effectiveness of advanced machine learning techniques in
predictive maintenance applications.

D. Comparative Analysis
This study advances the field of predictive maintenance
through innovative integration of XGBoost, MLP, and meta-
Fig 10: MLP Confusion Matrix learning ensemble techniques. Previous research, such as the
2021 IoT-enabled predictive maintenance study for diesel
generators, focused primarily on real-time monitoring
without incorporating advanced ensemble techniques [27].
Similarly, the 2023 case study employing Random Forest and
Support Vector Machines, while comprehensive, lacked the
hybrid robustness achieved through MLP and XGBoost
integration [28].

The research's significance lies in its innovative

methodology, delivering superior precision and recall across
multiple fault categories while demonstrating scalability
through hybrid ensemble modelling. This approach
transcends previous studies that relied on standalone models
or IoT-based diagnostics, establishing a new benchmark for
fault diagnosis in critical systems. The framework's
integration of diverse machine learning capabilities ensures
Fig 11: Hybrid Model Confusion Matrix

IJISRT25MAR1226 www.ijisrt.com 1425

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1226

broader applicability and enhanced reliability for complex E. Conflict of Interest

industrial machinery. The authors declare no conflict of interest. This research
received no external funding, and the development of the
V. CONCLUSION AND RECOMMENDATIONS predictive maintenance framework was conducted solely for
academic purposes. The manufacturing company that
A. Summary of Research provided access to their diesel generator for data collection
This research developed and validated an advanced had no role in the study design, data analysis, interpretation
predictive maintenance framework for a diesel generator of results, or the writing of this manuscript.
using sophisticated machine learning techniques. The study
integrated multiple approaches, including XGBoost, Multi- REFERENCES
Layer Perceptron Neural Network, and a Meta-Learning
Hybrid Model. Data collection spanned two months, [1]. Z. M. Çınar, A. A. Nuhu, Q. Zeeshan, O. Korhan, M.
capturing operational parameters at one-minute intervals. The Asmael, and B. Safaei, “Machine Learning in
models achieved exceptional accuracy, with the MLP Neural Predictive Maintenance towards Sustainable Smart
Network reaching perfect accuracy (1.0) and both XGBoost Manufacturing in Industry 4.0,” Sustainability, vol.
and Meta-Learning Hybrid Models achieving 0.9994 12, p. 8211, Dec. 2020, doi: 10.3390/su12198211.
accuracy. The framework demonstrated robust capabilities in [2]. C. Riccio, M. Menanno, I. Zennaro, and M. M.
predicting maintenance requirements across eight distinct Savino, “A New Methodological Framework for
categories, establishing a significant advancement in Optimizing Predictive Maintenance Using Machine
predictive maintenance technology. Learning Combined with Product Quality
Parameters,” Machines, vol. 12, p. 443, Dec. 2024,
B. Conclusion doi: 10.3390/machines12070443.
The research successfully established the effectiveness [3]. L. K. Narayanan, L. S, H. R, D. Jayalakshmi, and V.
of machine learning-based predictive maintenance for R. Vimal, “Machine Learning-Based Predictive
industrial diesel generators. The developed framework Maintenance for Industrial Equipment Optimization,”
demonstrated unprecedented accuracy in maintenance in 2024 International Conference on Trends in
prediction, surpassing traditional maintenance approaches. Quantum Computing and Emerging Business
Through comprehensive data analysis and advanced model Technologies, 2024, pp. 1–5. doi:
integration, the study validated the potential for significant 10.1109/TQCEBT59414.2024.10545280.
improvements in equipment reliability and operational [4]. A. Shaala, D. Baglee, and D. Dixon, “Machine
efficiency. The meta-learning approach proved particularly learning model for predictive maintenance of modern
effective, combining the strengths of multiple algorithms to manufacturing assets,” in 2024 29th International
enhance prediction reliability. These findings represent a Conference on Automation and Computing (ICAC),
significant contribution to the field of predictive 2024, pp. 1–6. doi:
maintenance, offering practical solutions for industrial 10.1109/ICAC61394.2024.10718768.
equipment management. [5]. A. Ucar, M. Karakose, and N. Kırımça, “Artificial
Intelligence for Predictive Maintenance Applications:
C. Recommendations Key Components, Trustworthiness, and Future
Trends,” Applied Sciences, vol. 14, p. 898, Dec. 2024,
 Implementation of real-time monitoring systems doi: 10.3390/app14020898.
integrated with the developed predictive maintenance [6]. L. Cummins et al., “Explainable Predictive
framework to enable immediate fault detection and Maintenance: A Survey of Current Methods,
response. Challenges and Opportunities,” IEEE access, p. 1,
 Extension of the data collection period beyond two Dec. 2024, doi: 10.1109/access.2024.3391130.
months to capture seasonal variations and long-term [7]. A. P. Kane, A. S. Kore, A. N. Khandale, S. S. Nigade,
degradation patterns in generator performance. and P. P. Joshi, “Predictive Maintenance using
 Development of a standardised implementation protocol Machine Learning,” Dec. 2022. doi:
for deploying the predictive maintenance framework 10.48550/arXiv.2205.09402.
across various industrial settings and generator [8]. R. Angel, “Predictive Maintenance with Machine
specifications. Learning.,” Grenze International Journal of
 Integration of additional sensor technologies and data Engineering & Technology (GIJET), vol. 10, 2024.
streams to enhance the model's predictive capabilities and [9]. L. K. Narayanan, L. S, H. R, D. Jayalakshmi, and V.
expand its application to diverse industrial equipment. R. Vimal, “Machine Learning-Based Predictive
Maintenance for Industrial Equipment Optimization,”
D. Research Limitations in 2024 International Conference on Trends in
Study limitations include the focus on a single generator Quantum Computing and Emerging Business
type and a two-month data collection period. Future research Technologies, 2024, pp. 1–5. doi:
opportunities include expanding the dataset, incorporating 10.1109/TQCEBT59414.2024.10545280.
diverse generator models, and exploring additional machine
learning techniques to further validate and extend these
findings.

IJISRT25MAR1226 www.ijisrt.com 1426

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1226

[10]. S. Khalaf, “Integration of Distributed Diesel [21]. M. Eltohamy, “Optimal Utilization of Distributed
Generators in Power System, Iraq Case Study,” Generation,” 2021. [Online]. Available:
Cardiff University, 2021. [Online]. Available: https://www.researchgate.net/profile/Mohammed-
https://orca.cardiff.ac.uk/id/eprint/149179/ Eltohamy/publication/379026349_OPTIMAL_UTILI
[11]. T. A. C. Guimarães, “Load Forecast on a Micro Grid ZATION_OF_DISTRIBUTED_GENERATION/
Level Through Machine Learning Algorithms,” [22]. S. Khalaf, “Integration of Distributed Diesel
University of Porto, 2020. [Online]. Available: Generators in Power Systems,” Cardiff University,
https://search.proquest.com/openview/f0b184775ad6 2021. [Online]. Available:
a29cd4529bf9bbb8d549 https://orca.cardiff.ac.uk/id/eprint/149179/1/Thesis%
[12]. N. F. P. Dinata, M. A. M. Ramli, M. I. Jambak, and 20Final.pdf
M. A. B. Sidik, “Designing an Optimal Microgrid [23]. L. P. Strydom, “Power System Design Guidelines to
Control System Using Deep Reinforcement Learning: Enhance Reliability of Cellular Networks in Africa,”
A Systematic Review,” ScienceDirect, 2024, North-West University, South Africa, 2014. [Online].
[Online]. Available: Available:
https://www.sciencedirect.com/science/article/pii/S2 https://repository.nwu.ac.za/handle/10394/15587
215098624000375 [24]. W. K. Chae and et al., “Design and Field Tests of an
[13]. S. K. Rajput, D. Kulshrestha, N. Paliwal, and V. Inverted Based Remote Microgrid on a Korean
Saxena, “Forecasting Capacitor Banks for Improving Island,” Energies (Basel), vol. 8, no. 8, p. 8193, 2015,
Efficiency of Grid-Integrated PV Plants: A Machine [Online]. Available: https://www.mdpi.com/1996-
Learning Approach,” ScienceDirect, 2025, [Online]. 1073/8/8/8193/pdf
Available: [25]. S. J. A. Haider, “Optimal Predictive Maintenance
https://www.sciencedirect.com/science/article/pii/S2 Strategies,” 2007. [Online]. Available:
352484724008230 https://sites.ualberta.ca/~jed3/Theses/Haider-
[14]. H. Agomuo and B. O. Ogbonna, “Development of MEngReport-UofA-2007.pdf
MPPT Algorithm for Improved Control of Industrial [26]. C. J. Lanigan, “Implementation of a condition
Power Systems, Case Study: Ocean Marine Security,” monitoring program for High Voltage (HV) assets for
IJAEM, 2024. the Santos GLNG Project,” 2013. [Online]. Available:
[15]. H. Nie, Y. Chen, Y. Xia, S. Huang, and B. Liu, https://sear.unisq.edu.au/24698/1/Lanigan_2013.pdf
“Optimizing the Post-Disaster Control of Islanded [27]. D. R. Nayak, A. G. Mohapatra, B. Keswani, A.
Microgrid: A Multi-Agent Deep Reinforcement Mohanty, P. K. Tripathy, and A. K. Samantaray, “IoT
Learning Approach,” in IEEE Xplore, 2020. [Online]. enabled predictive maintenance of diesel generator in
Available: the context to Industry 4.0,” in 2021 19th OITS
https://ieeexplore.ieee.org/document/9172071 International Conference on Information Technology
[16]. A. Adebayo, B. S. Oluwadare, and J. T. Stephen, (OCIT), 2021, pp. 364–368. doi:
“Ferrographic Study of Wear Particles in Used Oil of 10.1109/OCIT53463.2021.00078.
a Machinery System in Power Generating Plant,” [28]. D. S. Satwaliya, H. P. Thethi, A. Dhyani, G. R. Kiran,
IJSTRE, [Online]. Available: M. Al-Taee, and M. B. Alazzam, “Predictive
http://www.ijstre.com/Publish/4202019/12145446.pd Maintenance using Machine Learning: A Case Study
f in Manufacturing Management,” Dec. 2023, doi:
[17]. L. O. Mogaka, “Rotating Machine Based Distributed 10.1109/icacite57410.2023.1018301.
Generator Islanding Detection and Power
Prioritisation Using Artificial Bee Colony
Algorithm,” JKUAT, 2017. [Online]. Available:
http://ir.jkuat.ac.ke/handle/123456789/2576
[18]. M. G. M. Almihat and M. T. E. Kahn, “Design and
Implementation of Hybrid Renewable Energy
(PV/Wind/Diesel/Battery) Microgrids for Rural
Areas,” AJOL, 2023.
[19]. T. L. Vu, A. Singhal, and K. Schneider, “Tuning Phase
Lock Loop Controller of Grid Following Inverters by
Reinforcement Learning to Support Networked
Microgrid Operations,” in IEEE Xplore, 2023.
[20]. A. Adebayo, B. S. Oluwadare, and J. T. Stephen,
“Ferrographic Study of Wear Particles in Used Oil of
a Machinery System in Power Generating Plant,”
International Journal of Scientific and Technical
Research in Engineering (IJSTRE), 2019, [Online].
Available:
http://www.ijstre.com/Publish/4202019/12145446.pd
f

IJISRT25MAR1226 www.ijisrt.com 1427