Full metadata record
| DC Field | Value | Language |
|---|---|---|
| dc.contributor | Department of Building Environment and Energy Engineering | en_US |
| dc.contributor.advisor | Du, Yaping Patrick (BEEE) | en_US |
| dc.creator | Jia, Chuanzhen | - |
| dc.identifier.uri | https://theses.lib.polyu.edu.hk/handle/200/14570 | - |
| dc.language | English | en_US |
| dc.publisher | Hong Kong Polytechnic University | en_US |
| dc.rights | All rights reserved | en_US |
| dc.title | Low-voltage AC series arc fault detection and performance assessment | en_US |
| dcterms.abstract | Arc faults represent a major hidden risk in low-voltage AC systems, as they can generate persistent high-temperature plasma leading to insulation failure and fire hazards. Conventional protective devices often fail to detect these faults because arc signatures are transient, nonlinear, and strongly influenced by operating environments. Arc fault detection devices (AFDDs) are developed to prevent the electrical fires caused by arc faults. But, complex scenarios—such as various load conditions and line masking—further distort current signals, making detection highly unreliable. Existing standards address only limited arc-fault cases and lack comprehensive evaluation metrics, leaving a gap between laboratory testing and real-world residential environments. However, the circulation of these products in the market cannot effectively detect arc-fault events and prevent fires, thereby posing serious threats to human life and property safety. | en_US |
| dcterms.abstract | To overcome the gaps, this thesis aims to systematically evaluate the performance of existing AFDDs, develop novel arc fault detection algorithms, and propose a unified explainable assessment scheme for improved reliability and applicability. | en_US |
| dcterms.abstract | Standard tests cannot fully capture the complexity of real-life situations. To investigate the impact of diverse load types and line conditions on arc-fault signals, this thesis develops a sophisticated experimental platform and designs various additional tests to analyze the behavior of AFDDs under different complex real-world operating conditions. Based on the platform, existing AFDDs are systematically evaluated under both standard requirements and additional test scenarios to identify key weaknesses in practical implementations. | en_US |
| dcterms.abstract | To address these challenges, two refined arc fault detection algorithms are proposed targeting complex load types and long cable lengths, thereby enhancing the accuracy, reliability, and real-world applicability of arc fault detection. | en_US |
| dcterms.abstract | Existing arc fault detection methods struggle with low accuracy in complex load environments and lack adaptability to simultaneous or mixed load operations. Prior studies offer no clear guidance on optimal sampling parameters. To address the challenges, this thesis proposes a lightweight AC arc fault detection framework based on event-based load classification. The method first analyzes arc characteristics under different load conditions and performs load classification using transient events, categorizing loads into resistive, inductive, and switchable types. For each category, sequential floating forward selection is employed to optimize feature subsets, and a KNN-based classifier is used to achieve efficient and accurate detection. The experimental results demonstrate that the proposed method performs effectively in practical arc fault detection under multi-load conditions and has stronger applicability in arc fault detection than other reported methods, including those using deep learning models. The proposed framework avoids the high computational cost of deep learning while maintaining strong detection performance. | en_US |
| dcterms.abstract | AFDDs exhibit significant drops in detection accuracy when operating under long cable lengths. Regarding complex masking tests with line impedance, a novel approach to AC arc fault detection is introduced, combining multi-band feature fusion with Catboost. Due to the different attenuation degrees of high-frequency features under various load types in long-distance cables, the startup power variations are analyzed to capture their transient characteristics for classification. Recursive Feature Elimination and CatBoost classifiers are used to extract optimal features and detect arc-fault events, improving detection performance without requiring deep learning models. The proposed method achieves 97% accuracy in arc fault detection, even when models are trained only under ideal (0m) conditions, demonstrating excellent generalization under long-cable conditions. An ablation study and model comparison are conducted to validate the effectiveness of the proposed method. | en_US |
| dcterms.abstract | The high accuracy does not guarantee true feature extraction. Existing arc fault diagnosis models often lack trustworthiness and interpretability, relying too heavily on accuracy while ignoring whether true features are captured. The study introduces a unified explainable assessment scheme that defines ground-truth fault features and integrates explainable artificial intelligence (XAI) to assess model reliability. Two XAI methods, Occlusion Sensitivity and SHAP, are used to assess how models identify key fault features for feature pool-based methods and deep learning methods. Several traditional machine learning methods and deep learning methods across two arc fault datasets with different sample times and noise levels are utilized to test the effectiveness of the proposed assessment scheme. Through this approach, the arc fault diagnosis models are easy to understand and trust, allowing practitioners to make informed and trustworthy decisions. Based on the results, a lightweight balanced neural network is developed to guarantee competitive accuracy and soft feature extraction score. The proposed method enhances model transparency, supports real-world deployment with low computational cost, and guides model selection and ensemble strategies. | en_US |
| dcterms.abstract | In conclusion, this thesis presents a comprehensive framework that integrates a newly designed testing platform, two advanced arc fault detection algorithms, and a robust explainable assessment scheme. Building upon existing standards and arc fault detection algorithms, this thesis systematically addresses their limitations and introduces novel methods to enhance detection performance. The proposed framework enables reliable identification of arc-fault signals in real-world environments characterized by diverse loads and complex disturbances. In addition to improving detection accuracy and robustness, this work provides practical insights into standardized test requirements and algorithm evaluation. These contributions support manufacturers in developing safer and more reliable arc fault detection products and contribute to reducing the incidence of electrical fires. | en_US |
| dcterms.extent | xiv, 122 pages : color illustrations | en_US |
| dcterms.isPartOf | PolyU Electronic Theses | en_US |
| dcterms.issued | 2026 | en_US |
| dcterms.educationalLevel | Ph.D. | en_US |
| dcterms.educationalLevel | All Doctorate | en_US |
| dcterms.LCSH | Electric fault location | en_US |
| dcterms.LCSH | Electric currents | en_US |
| dcterms.LCSH | Electric power systems -- Protection | en_US |
| dcterms.LCSH | Hong Kong Polytechnic University -- Dissertations | en_US |
| dcterms.accessRights | open access | en_US |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/14570