Development of a Data-Driven Approach for Permanent Fault Location in Underground Power Cables

Industrial plants frequently encounter various contingencies in their electricity systems, which can be attributed to the loss or failure of specific components such as lines, cables, or individual equipment. In response to these faults, protective devices isolate the affected areas from the network. However, identifying the exact location of the fault can be a time-consuming process. Operators rely on their expertise and available tools, but when dealing with buried underground cables or hard-to-reach equipment, it can take several hours to trace the fault's origin. To address this challenge, this research project aims to develop an innovative solution capable of detecting fault locations based on field measurement data. The primary focus is on creating advanced algorithms that can effectively identify faults in both buried high-voltage (HV) feeder cables used in potash mines and medium-voltage (MV) cables in underground mining operations. The project's ultimate goal is to demonstrate the feasibility of a practical device capable of instantly reporting, displaying, or transmitting fault location information in response to contingencies. The main beneficiaries of this project are expected to be prominent mining companies in Saskatchewan and Canada. They can integrate these cutting-edge innovations into their industrial plants, enhancing their operational efficiency and minimizing downtime. Efforts to achieve efficient and cost-effective fault localization in electrical cables have led to increased interest in pinpointing cable faults through local measurements. This approach strikes a balance between precision and hardware costs, making it attractive for the industry. This research introduces a novel method for online fault localization in medium-voltage (MV) cables, utilizing measurements of local sheath current. The innovative technique leverages Artificial Neural Networks (ANNs), a subset of artificial intelligence (AI), to enhance fault detection and location accuracy. It builds upon the principles of reflectometry, analyzing how electrical signals propagate along the cable and reflect when encountering faults. Medium-voltage cables present unique challenges, but this method's cost-efficient approach uses a single sensor at the sheath grounding joint to measure the sum of three-phase local sheath currents. This method's strength lies in its ability to estimate wave travel delay, a crucial factor in determining fault location, especially when wave arrivals are obscured or attenuated. In the realm of high-voltage (HV) cable systems, swift and accurate fault localization is crucial for restoring power promptly. Achieving high accuracy in fault localization while managing measurement costs is a delicate balance. This thesis proposes an efficient framework for HV cable fault localization, analyzing sheath currents in modal mode and comparing them to traditional core conductor measurements. The discovery that collective sheath current across phases exhibits similar characteristics opens the door to using fewer, lower-rated sensors as a cost-effective alternative. However, working with sheath current measurements presents challenges, particularly in wavefront recognition within the sheath. To address this, the thesis introduces a Convolutional Neural Network (CNN) tailored for precise sheath current-based fault localization. These approaches excel in achieving high localization accuracy while mitigating measurement costs and maintaining consistent performance across various operational scenarios, even with limited training data. Empirical validation through a comprehensive case study on the PSCAD/EMTDC platform highlights the effectiveness and feasibility of these novel frameworks, shedding light on its key insights and implications.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.