Back to search

University of Saskatchewan

Development of a Data-Driven Framework for Partial Discharge Faults Detection and Location in Power Cables

Abstract

dc:description.abstract

Power cables are widely used in medium-voltage (MV) and high-voltage (HV) power systems. However, this widespread power cable deployment is not without problems. As a result of different reasons, incipient faults such as partial discharge (PD) gradually grow inside the insulation until they cause complete failure. Cable failures due to PD have been one of the major problems in power systems. To prevent power outages, PD should be detected before reaching a critical point. Therefore, it is necessary to monitor and improve the condition of power cables to avoid significant disruptions. Monitoring and detecting PD remains a challenge for utility owners due to various factors, including environmental conditions (e.g., background noise presence), sensitivity limitations (as PD pulses attenuate while propagating toward cable ends), and the intrinsic nature of PD events, which are characterized by very small pulses typically in the millivolt range. To address the aforementioned challenges, numerous researchers have explored various methodologies, with a growing emphasis on data-driven approaches, particularly machine learning (ML) and deep learning (DL) techniques. However, these methods generally suffer from two key limitations. First, they often struggle to extract discriminative features for reliable PD identification and separation from PD-like interference signals. Second, DL-based classification frameworks typically operate as black-box systems, making it difficult to interpret misclassifications or understand the rationale behind a given output. To address these drawbacks, this study adopts a novel approach for the PD identification task. In this approach, first, a DL-based feature extraction framework is employed to obtain a highly abstract representation of PD pulses and generate a set of synthetic features optimized for classification. Second, to enhance interpretability and overcome the limitations of DL-based classifiers, a knowledge-based classification scheme is introduced. This classifier is based on rough set theory (RST) and incorporates a novel feature analysis algorithm (FAA) for distinguishing PD from noise. Unlike DL-based classifiers, which may yield unreliable results when faced with data inconsistencies, the proposed RST framework as the primary classifier, refraining from making decisions when classification rules are not satisfied. In such cases, the FAA is activated to perform a more detailed analysis using both shallow and deep feature evaluations. The efficacy of the proposed identification framework is verified through simulations and experimental studies. Following PD identification, appropriate remedial actions must be taken based on the severity of the PD to address the defective cable segment and prevent potential failure and service interruptions. In this context, a reliable and precise localization method is essential for minimizing inspection time and reducing unnecessary costs. However, PD localization presents several challenges and complexities. Depending on the measurement scheme, various localization methodologies have been proposed. This study adopts a single-end localization approach, which simplifies implementation by avoiding the complexities associated with double-sided or distributed measurement schemes. A key challenge in such systems is the identification of incident pulses and their corresponding reflections within the measured signal, especially in the presence of impulse noise and PD-like interference. These disturbances make it difficult to reliably distinguish relevant incident and reflection pulses. To overcome these challenges, this study introduces a localization method that leverages traveling wave characteristics and transfer function (TF) analysis to determine the PD source, even in complex scenarios involving multi-path propagation, impulse noise, and simultaneous PD events. The proposed approach incorporates a cable-specific attenuation characteristic into a two-step signal segmentation algorithm, followed by the application of a U-Net model to precisely estimate PD pulse arrival times. Additionally, a statistical analysis is conducted to evaluate the method's maximum localization capability based on the noise level and cable length. The performance of the method is assessed under both homogeneous and inhomogeneous cable configurations.

Degree

thesis:*
Name thesis:degree_name
Doctor of Philosophy (Ph.D.)
Level thesis:degree_level
Doctoral
Discipline thesis:degree_discipline
Electrical Engineering
Department thesis:degree_department
Electrical and Computer Engineering
Grantor
University of Saskatchewan
Year dc:date.issued
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Shamsoddini, Morteza
Advisor dc:contributor.advisor
  • Ko, Seokbum
Committee members dc:contributor.committeemember
  • Berscheid, Brian
  • Zareipour, Hamid
  • Faried, Sherif
  • Chen, Li
  • Zhang, Chris
ORCID person:identifier.orcid

Subjects

dc:subject × 7

Rights

Language dc:language.iso
en

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/10388/17518

Chain of custody

source
Harvested from
University of Saskatchewan
Base URL
harvest.usask.ca/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Shamsoddini, Morteza. Development of a Data-Driven Framework for Partial Discharge Faults Detection and Location in Power Cables. Doctoral thesis, University of Saskatchewan, 2025. https://hdl.handle.net/10388/17518