How Genetic Algorithms Improve Multiclass Gas Pipeline Leak Detection: A Review of Multi-Domain Machine Learning Classification

on

Gas pipeline systems constitute one of the most critical infrastructures supporting modern energy distribution. Detecting pipeline leaks at an early stage is essential for preventing catastrophic failures, minimizing environmental damage, reducing economic losses, and protecting human life. While numerous leak detection techniques have been proposed, accurately distinguishing among multiple leak conditions remains considerably more challenging than simply detecting whether a leak exists. This study introduces a multiclass gas pipeline leak detection framework that combines acoustic signal analysis from both time and frequency domains with genetic algorithm-based feature selection and conventional machine learning classifiers. Using the GPLA-12 benchmark dataset, the proposed methodology demonstrates that carefully engineered multi-domain features coupled with evolutionary optimization can achieve high classification accuracy while maintaining low computational complexity. The research provides an efficient alternative to computationally intensive deep learning approaches, particularly for industrial environments where labeled datasets and computing resources are limited.

Bibliographic Information

Item Information
Article Title Multiclass gas pipeline leak detection using multi-domain signals and genetic algorithm-optimized classification models
Authors Wiwit Suprihatiningsih; Dedik Romahadi; Hadi Pranoto; Rikko Putra Youlia; Fajar Anggara; Rizky Rahmatullah
Journal Teknomekanik
Volume 9
Issue 1
Publication Date February 2026
Pages 25–41
DOI https://doi.org/10.24036/teknomekanik.v9i1.38372
Publisher Universitas Negeri Padang
License Creative Commons Attribution 4.0 International (CC BY 4.0)
e-ISSN 2621-8720
p-ISSN 2621-9980
Keywords gas pipeline leak detection; multiclass classification; acoustic signal; genetic algorithm

1. Research Background

  • Gas pipeline networks are indispensable components of modern energy infrastructure. Pipelines provide an efficient and economical means of transporting natural gas and other gaseous resources over long distances. Because these systems operate continuously and frequently traverse densely populated or environmentally sensitive regions, maintaining their structural integrity is essential for public safety, environmental protection, and uninterrupted industrial operations.
  • Pipeline leaks can rapidly escalate into serious safety and environmental incidents. Leaks may occur due to corrosion, material degradation, welding defects, accidental third-party damage, or operational failures. Undetected leakage increases the risk of explosions, fires, environmental contamination, production interruptions, and significant financial losses. Consequently, industries require detection systems capable of identifying leak conditions as early as possible.
  • Most conventional monitoring systems focus primarily on binary leak detection. Existing methods commonly determine whether a leak exists by monitoring variations in pressure, flow rate, vibration, or acoustic emissions. Although these techniques successfully identify abnormal operating conditions, they often lack the capability to distinguish among different leak types occurring under varying operating environments.
  • Multiclass leak identification presents a considerably more complex engineering challenge. Real industrial environments involve different pipe materials, operating pressures, sensor locations, environmental noise levels, and leak mechanisms. These variations produce highly heterogeneous acoustic signals that are difficult to classify accurately, particularly when only limited labeled datasets are available for supervised learning.
  • Deep learning models are not always the most practical solution for industrial deployment. Convolutional neural networks and other deep architectures have demonstrated promising performance in leak detection studies. However, their successful implementation generally depends on large annotated datasets, extensive computational resources, long training times, and specialized hardware. These requirements can limit adoption in many industrial settings.
  • Conventional machine learning remains attractive for data-constrained applications. Algorithms such as Support Vector Machines, Random Forests, Naïve Bayes, Decision Trees, K-Nearest Neighbors, Linear Discriminant Analysis, and Artificial Neural Networks with carefully engineered features often provide competitive performance while requiring substantially less computational effort. Their relatively simple architecture also facilitates interpretation and deployment on resource-constrained systems.
  • Feature engineering plays a decisive role in improving multiclass classification performance. Acoustic leak signals contain valuable information in both the time and frequency domains. Combining temporal waveform characteristics with frequency-domain representations obtained through Power Spectral Density analysis enables a richer description of leak behavior than relying on either domain individually.
  • Evolutionary optimization offers an effective strategy for feature selection. Instead of using every extracted feature, Genetic Algorithms can identify compact feature subsets that maximize class separability while reducing redundancy and computational cost. Feature optimization becomes particularly valuable when training datasets are relatively small and model efficiency is an important practical consideration.
  • The present research addresses an important methodological gap. The study develops an integrated multiclass leak detection framework that combines multi-domain acoustic feature extraction with Genetic Algorithm-based feature selection and systematic benchmarking of seven conventional machine learning classifiers. The objective is to achieve high classification accuracy while maintaining computational efficiency suitable for real-world industrial implementation.

2. Research Objectives

  • To develop an end-to-end multiclass gas pipeline leak detection framework capable of distinguishing twelve different gas leak conditions using acoustic signal analysis.
  • To construct a comprehensive multi-domain feature representation by integrating time-domain acoustic descriptors with frequency-domain features extracted using Power Spectral Density analysis.
  • To employ a Genetic Algorithm as a wrapper-based feature selection technique for identifying the most informative feature subset while reducing dimensionality and computational complexity.
  • To benchmark the performance of seven conventional supervised machine learning classifiers—Naïve Bayes, K-Nearest Neighbors, Random Forest, Linear Discriminant Analysis, Decision Tree, Support Vector Machine, and Artificial Neural Network—using the GPLA-12 benchmark dataset.
  • To further optimize the highest-performing classifier through Genetic Algorithm-based refinement of both feature subsets and model hyperparameters.
  • To demonstrate that conventional machine learning combined with carefully engineered acoustic features can achieve competitive multiclass classification performance compared with more computationally demanding deep learning approaches in limited-data environments.

3. Why This Research Matters

  • Improves industrial safety. Earlier and more accurate identification of leak types enables operators to respond rapidly before failures escalate into catastrophic accidents involving explosions, fires, or environmental contamination.
  • Supports reliable pipeline integrity management. Accurate multiclass classification provides engineers with richer diagnostic information than simple leak detection, allowing maintenance activities to be prioritized according to the specific characteristics of detected leak events.
  • Reduces computational requirements. The proposed framework demonstrates that carefully engineered features and conventional machine learning algorithms can achieve excellent classification accuracy without requiring the extensive computational resources associated with deep neural networks.
  • Addresses challenges associated with limited datasets. Many industrial applications cannot easily generate thousands of labeled fault examples. By emphasizing efficient feature engineering and evolutionary optimization, the proposed methodology remains suitable for practical scenarios where training data are scarce.
  • Enhances explainability. Conventional machine learning models combined with explicit feature selection provide greater transparency regarding classification decisions than many black-box deep learning architectures, supporting engineering interpretation and model validation.
  • Demonstrates the value of multi-domain acoustic analysis. Integrating temporal and spectral information produces a richer representation of leak characteristics, improving discrimination among multiple leak classes while maintaining computational simplicity.
  • Provides a practical framework for industrial deployment. Because the proposed approach relies on established signal processing techniques, publicly available benchmark data, Genetic Algorithm optimization, and conventional classifiers, it offers a feasible pathway toward implementation in resource-constrained monitoring systems for oil and gas infrastructure.

4. Research Methodology

  • Research Design

    The study adopted an experimental machine learning framework for multiclass gas pipeline leak classification. Rather than developing a new sensing system, the researchers focused on improving classification performance by integrating multi-domain acoustic feature extraction with evolutionary feature optimization and supervised machine learning. The complete workflow consisted of three sequential stages: dataset preparation, multi-domain feature extraction, and Genetic Algorithm (GA)-assisted feature selection followed by classifier evaluation and optimization.

  • Dataset

    The experiments were conducted using the publicly available GPLA-12 benchmark dataset, which contains acoustic signals collected from gas pipeline leakage experiments. The dataset consists of 684 acoustic signal samples, each represented by 1,460 variables, with 57 samples for each of the twelve leak categories. The leak scenarios represent different operating conditions involving pressure levels, environmental noise conditions, and microphone configurations, creating a balanced multiclass classification problem.

  • Computing Environment

    All experiments were implemented using MATLAB 2024b under a fully supervised learning framework. Model training and evaluation were performed on a Windows 11 workstation equipped with a 14th-generation Intel Core i9 processor and an RTX 4080 Super graphics card. However, GPU acceleration was intentionally not utilized, demonstrating that the proposed framework can operate efficiently using CPU resources alone.

  • Multi-Domain Feature Extraction

    The proposed framework combines information from two complementary signal domains. Raw acoustic signals preserve transient waveform characteristics within the time domain, while frequency-domain information is obtained using Power Spectral Density (PSD) estimation through the Welch method. The resulting spectral descriptors characterize how signal energy is distributed across different frequency components associated with various leak conditions.

    The extracted PSD features are concatenated with the original time-domain signal representation to form a comprehensive multi-domain feature set. This integration enables the classification model to exploit both temporal behavior and frequency characteristics, producing a richer and more discriminative feature representation than either domain individually.

  • Machine Learning Models

    Seven conventional supervised machine learning algorithms were evaluated to determine their suitability for multiclass gas leak identification under limited-data conditions. The benchmark models included:

    • Naïve Bayes (NB)
    • K-Nearest Neighbors (KNN)
    • Random Forest (RF)
    • Linear Discriminant Analysis (LDA)
    • Decision Tree (DT)
    • Support Vector Machine (SVM)
    • Artificial Neural Network (ANN)

    These algorithms were selected because they represent a broad spectrum of statistical learning approaches while remaining computationally efficient and well suited to relatively small datasets.

  • Genetic Algorithm-Based Feature Selection

    A wrapper-based Genetic Algorithm was employed to identify the optimal subset of features for classification. Each chromosome represented a binary vector indicating whether individual features were included or excluded from the classification model. Candidate feature subsets were evaluated according to classifier performance, and successive generations evolved through selection, crossover, and mutation operations.

    The optimization process continued for up to 300 generations or until no further improvement in classification performance was observed. By removing redundant or less informative variables, the Genetic Algorithm simultaneously reduced feature dimensionality and improved model discrimination capability.

  • Performance Evaluation

    Model performance was assessed using a five-fold cross-validation strategy. The dataset was divided into five equally sized subsets, allowing each subset to serve once as the testing set while the remaining four subsets were used for training. This validation strategy reduces evaluation bias and provides more reliable performance estimates when working with relatively small datasets.

    Classification performance was measured using several widely accepted metrics, including Accuracy, Precision, Recall (Sensitivity), Specificity, and F-score. These complementary evaluation criteria provide a comprehensive assessment of multiclass classification capability beyond overall accuracy alone.

5. Key Findings

Genetic Algorithm Optimization Significantly Improved Classification Performance

The incorporation of Genetic Algorithm-based feature selection substantially enhanced classifier performance by identifying compact feature subsets with greater discriminative capability. Instead of relying on the complete feature space, the optimized models achieved superior classification performance while simultaneously reducing computational complexity. This demonstrates that feature quality is more important than feature quantity for multiclass acoustic leak identification.

The Proposed Framework Achieved 96.35% Classification Accuracy

The highest-performing optimized model achieved a multiclass classification accuracy of 96.35% on the GPLA-12 benchmark dataset. This level of performance demonstrates that carefully engineered acoustic features combined with evolutionary optimization can accurately distinguish among twelve different gas pipeline leak conditions without relying on computationally intensive deep learning architectures.

Multi-Domain Features Outperformed Single-Domain Representations

Integrating time-domain waveform information with frequency-domain Power Spectral Density features produced a more informative representation of leak characteristics than using either domain independently. Temporal descriptors preserved transient signal behavior, while spectral features captured frequency-related anomalies associated with different leak mechanisms. Their combination increased the separability of multiclass acoustic patterns.

Conventional Machine Learning Remained Highly Competitive

The benchmarking study demonstrated that conventional supervised machine learning algorithms remain highly effective when combined with appropriate feature engineering and optimization techniques. Rather than requiring complex deep neural networks, the proposed framework achieved comparable classification accuracy using simpler models that demand significantly fewer computational resources and less training data.

Feature Selection Reduced Computational Cost

The Genetic Algorithm eliminated redundant and non-informative variables while preserving classification accuracy. This reduction in feature dimensionality decreased computational requirements, making the proposed methodology more suitable for industrial environments where processing speed and hardware limitations are practical considerations.

The Framework Demonstrated Practical Industrial Applicability

Because the complete workflow relies on standard acoustic sensing, publicly available signal processing techniques, conventional machine learning algorithms, and CPU-based computation, the methodology provides a practical solution for real-world gas pipeline monitoring systems operating under resource constraints.

6. Scientific Contribution

  • Introduces an integrated multiclass leak classification framework that combines time-domain acoustic descriptors, frequency-domain Power Spectral Density features, Genetic Algorithm optimization, and conventional supervised machine learning into a unified methodology.
  • Demonstrates the effectiveness of multi-domain acoustic feature fusion for improving discrimination among twelve different gas pipeline leak classes while maintaining computational efficiency.
  • Validates Genetic Algorithm as an effective wrapper-based feature selection approach capable of simultaneously reducing feature dimensionality and improving classification accuracy.
  • Provides a comprehensive comparative evaluation of seven widely used machine learning classifiers under identical experimental conditions, offering valuable insight into their relative suitability for multiclass gas leak identification.
  • Shows that conventional machine learning can compete with deep learning approaches when feature engineering and evolutionary optimization are carefully integrated, particularly in environments characterized by limited labeled datasets and constrained computational resources.
  • Establishes a reproducible benchmark framework using the GPLA-12 dataset that can support future research in acoustic leak detection, feature optimization, and industrial fault diagnosis.

7. Industrial Implications

  • Enhances pipeline safety management. Accurate multiclass identification enables operators to recognize different leak conditions more rapidly, supporting timely maintenance decisions and reducing the probability of catastrophic pipeline failures.
  • Supports predictive maintenance strategies. Reliable leak classification provides richer diagnostic information than simple anomaly detection, allowing maintenance personnel to prioritize inspections according to leak characteristics rather than merely detecting abnormal events.
  • Reduces computational infrastructure requirements. The proposed methodology operates efficiently using CPU-based computation without requiring specialized GPU hardware, making deployment feasible in industrial monitoring systems with limited computing resources.
  • Facilitates implementation in resource-constrained environments. Industries that cannot generate large labeled datasets or invest in high-performance computing infrastructure may still achieve high classification accuracy using conventional machine learning enhanced through Genetic Algorithm optimization.
  • Supports intelligent condition monitoring. Combining acoustic sensing with optimized machine learning contributes to the development of smarter pipeline integrity management systems capable of continuous automated monitoring.
  • Provides a scalable framework for industrial digitalization. The proposed workflow can serve as a foundation for future intelligent monitoring systems incorporating additional sensing technologies, real-time analytics, and Industry 4.0 infrastructure while maintaining computational efficiency.

8. Research Limitations

  • The study was evaluated using a single benchmark dataset. All experiments were conducted exclusively on the GPLA-12 dataset. Although this benchmark contains twelve balanced leak classes and provides a suitable platform for comparative evaluation, additional validation using other datasets and real industrial pipeline systems would further demonstrate the robustness and generalizability of the proposed framework.
  • The research focused exclusively on supervised multiclass leak classification. The proposed methodology was designed to classify predefined leak categories based on labeled acoustic signals. It does not address unknown or previously unseen leak mechanisms that may occur in practical industrial environments.
  • Pipeline leak localization was beyond the scope of the study. The framework identifies the type of leak but does not estimate the physical location of leakage along the pipeline. Integrating classification with localization techniques would provide more comprehensive support for maintenance operations.
  • Leak severity estimation was not investigated. Although the classifier distinguishes among twelve leak conditions, it does not quantify leakage magnitude, discharge rate, or damage severity. These capabilities would be valuable for prioritizing emergency responses and maintenance planning.
  • The framework relies solely on acoustic sensing. Other monitoring variables commonly used in pipeline integrity management, such as pressure, flow rate, temperature, or vibration sensor fusion, were intentionally excluded from the proposed methodology. Multi-sensor integration may further improve classification reliability under complex operating conditions.
  • Domain adaptation was not considered. The study does not investigate how the trained models perform when acoustic signals originate from different pipeline materials, operating pressures, sensor layouts, or environmental conditions. These domain shifts remain important challenges for industrial deployment.
  • Open-set recognition was outside the research objectives. The proposed classifier assumes that every testing sample belongs to one of the twelve known leak categories. Detecting entirely new or abnormal fault conditions beyond the training classes remains an important area for future investigation.

9. Future Research Opportunities

  • Develop hybrid monitoring systems that integrate acoustic sensing with additional measurement modalities such as pressure, flow rate, vibration, ultrasonic sensing, or fiber-optic monitoring to improve leak diagnosis under complex industrial environments.
  • Investigate pipeline leak localization techniques capable of simultaneously identifying both the leak category and its spatial position within the pipeline network.
  • Extend the proposed framework to estimate leak severity, leakage rate, and damage progression, providing richer information for predictive maintenance and emergency response planning.
  • Evaluate transfer learning and domain adaptation strategies to improve model robustness across different pipeline materials, operating pressures, environmental conditions, and sensor configurations.
  • Explore open-set recognition approaches capable of identifying previously unseen leak mechanisms that are not represented within the original training dataset.
  • Investigate lightweight deep learning architectures combined with Genetic Algorithm-based feature optimization to determine whether hybrid approaches can further improve classification accuracy while maintaining computational efficiency.
  • Validate the methodology using real industrial pipeline monitoring systems operating under continuously changing environmental conditions to assess long-term reliability and practical deployment performance.

10. Potential for Public Policy Citation

This research provides valuable evidence for organizations responsible for energy infrastructure regulation, industrial safety, and environmental protection. The proposed multiclass leak detection framework demonstrates that accurate and computationally efficient monitoring systems can be developed using conventional machine learning enhanced through evolutionary optimization. Such findings may support policy initiatives promoting intelligent infrastructure monitoring, predictive maintenance, and digital transformation within the oil and gas sector.

Government agencies responsible for pipeline safety may reference this research when developing technical guidelines for continuous condition monitoring, leak detection standards, and emergency response systems. By emphasizing early leak identification with relatively modest computational requirements, the proposed methodology is particularly relevant for organizations seeking cost-effective monitoring solutions suitable for both large-scale transmission pipelines and distributed industrial facilities.

The study also contributes to broader policy discussions concerning critical infrastructure resilience, environmental risk mitigation, industrial digitalization, and sustainable energy transportation. The ability to accurately classify leak conditions before catastrophic failures occur aligns with national strategies aimed at improving industrial safety, minimizing environmental pollution, and supporting more resilient energy distribution networks.

11. Who Should Read This Paper?

  • Researchers working in machine learning, pattern recognition, signal processing, and intelligent fault diagnosis.
  • Mechanical, petroleum, and pipeline engineers responsible for pipeline integrity management and industrial monitoring systems.
  • Researchers investigating acoustic sensing technologies for industrial fault detection and predictive maintenance.
  • Data scientists interested in feature engineering, evolutionary optimization, and conventional machine learning for engineering applications.
  • Professionals developing Industry 4.0 monitoring systems, intelligent sensors, and digital twin technologies for critical infrastructure.
  • Graduate students studying machine learning applications in mechanical engineering, industrial engineering, artificial intelligence, or energy systems.
  • Government agencies, industrial regulators, and policy makers responsible for pipeline safety, environmental protection, and energy infrastructure resilience.

12. Final Thoughts

This study demonstrates that high-performance multiclass gas pipeline leak detection does not necessarily require computationally intensive deep learning architectures. Through the integration of carefully engineered multi-domain acoustic features and Genetic Algorithm-based feature selection, the proposed framework achieves excellent classification performance while maintaining computational simplicity suitable for practical industrial deployment.

One of the most significant contributions of the research is its demonstration that conventional machine learning remains highly competitive when supported by effective feature engineering and evolutionary optimization. Rather than relying solely on increasingly complex neural network architectures, the study illustrates the continued relevance of interpretable and computationally efficient learning algorithms for engineering applications characterized by limited labeled datasets and constrained computing resources.

Beyond the reported classification accuracy, the research establishes a practical workflow that can serve as a foundation for future intelligent pipeline monitoring systems. The methodology is sufficiently flexible to accommodate additional sensing modalities, advanced optimization techniques, and real-time monitoring capabilities as industrial digitalization continues to evolve.

Overall, the paper makes a meaningful contribution to intelligent infrastructure monitoring by combining signal processing, evolutionary computation, and supervised machine learning into a practical multiclass leak diagnosis framework. It offers both scientific value and industrial relevance, particularly for organizations seeking accurate, interpretable, and resource-efficient solutions for safeguarding critical pipeline infrastructure.

13. Suggested Citations

Teknomekanik (UNP) Style

Suprihatiningsih, W., Romahadi, D., Pranoto, H., Youlia, R. P., Anggara, F., & Rahmatullah, R. (2026). Multiclass gas pipeline leak detection using multi-domain signals and genetic algorithm-optimized classification models. Teknomekanik, 9(1), 25–41. https://doi.org/10.24036/teknomekanik.v9i1.38372

APA (7th Edition)

Suprihatiningsih, W., Romahadi, D., Pranoto, H., Youlia, R. P., Anggara, F., & Rahmatullah, R. (2026). Multiclass gas pipeline leak detection using multi-domain signals and genetic algorithm-optimized classification models. Teknomekanik, 9(1), 25–41. https://doi.org/10.24036/teknomekanik.v9i1.38372

IEEE Style

W. Suprihatiningsih, D. Romahadi, H. Pranoto, R. P. Youlia, F. Anggara, and R. Rahmatullah, "Multiclass gas pipeline leak detection using multi-domain signals and genetic algorithm-optimized classification models," Teknomekanik, vol. 9, no. 1, pp. 25–41, Feb. 2026, doi: 10.24036/teknomekanik.v9i1.38372.

Harvard Style

Suprihatiningsih, W., Romahadi, D., Pranoto, H., Youlia, R.P., Anggara, F. & Rahmatullah, R., 2026. Multiclass gas pipeline leak detection using multi-domain signals and genetic algorithm-optimized classification models. Teknomekanik, 9(1), pp.25–41. Available at: https://doi.org/10.24036/teknomekanik.v9i1.38372.

Vancouver Style

Suprihatiningsih W, Romahadi D, Pranoto H, Youlia RP, Anggara F, Rahmatullah R. Multiclass gas pipeline leak detection using multi-domain signals and genetic algorithm-optimized classification models. Teknomekanik. 2026;9(1):25-41. doi:10.24036/teknomekanik.v9i1.38372.

Chicago (Author–Date)

Suprihatiningsih, Wiwit, Dedik Romahadi, Hadi Pranoto, Rikko Putra Youlia, Fajar Anggara, and Rizky Rahmatullah. 2026. "Multiclass Gas Pipeline Leak Detection Using Multi-Domain Signals and Genetic Algorithm-Optimized Classification Models." Teknomekanik 9 (1): 25–41. https://doi.org/10.24036/teknomekanik.v9i1.38372.

MLA (9th Edition)

Suprihatiningsih, Wiwit, et al. "Multiclass Gas Pipeline Leak Detection Using Multi-Domain Signals and Genetic Algorithm-Optimized Classification Models." Teknomekanik, vol. 9, no. 1, 2026, pp. 25–41. Crossref, https://doi.org/10.24036/teknomekanik.v9i1.38372.

14. Editorial Note

Engineering Research Insights is an independent scientific review series that highlights recently published engineering research with the objective of improving knowledge dissemination across academia, industry, and government. This review summarizes the scientific contribution, methodology, engineering significance, and potential practical applications of the reviewed article while maintaining the integrity of the original publication.

The bibliographic metadata presented in this review were verified using the official article webpage published by Teknomekanik. The scientific discussion—including the research background, objectives, methodology, findings, limitations, and future research directions—was prepared exclusively from the content of the published research article. No scientific interpretations were derived from external literature beyond the verified bibliographic metadata.

Readers are encouraged to consult the original publication for complete methodological details, mathematical formulations, experimental procedures, figures, tables, and supplementary discussions.

15. SEO Meta Description

A comprehensive review of the Teknomekanik research article on multiclass gas pipeline leak detection using multi-domain acoustic signals, Genetic Algorithm feature selection, and conventional machine learning classifiers. Learn how the proposed framework achieves 96.35% classification accuracy while maintaining computational efficiency for industrial pipeline monitoring.

16. SEO Keywords

gas pipeline leak detection, multiclass leak classification, genetic algorithm, feature selection, machine learning, acoustic signal processing, power spectral density, PSD, pipeline monitoring, fault diagnosis, intelligent pipeline inspection, supervised learning, support vector machine, random forest, artificial neural network, Industry 4.0, predictive maintenance, engineering research review, Teknomekanik, pipeline integrity management

17. Recommended URL Slug

multiclass-gas-pipeline-leak-detection-genetic-algorithm-review

Reviewed by Engineering Research Insights
Independent Engineering Literature Review Series


Comments

Post a Comment