An Intelligent Machine Learning based Fault Detection and Classification Technique for Secondary Distribution Network using contactless Magnetic Measurements of a Distribution Transformer /

Abstract

Electric power distribution system is of significant importance in insuring uninterrupted power delivery to end users. Overhead distribution conductors are commonly subjected to various types of short circuit faults mainly caused by growing vegetation, storms and otherenvironmental factors. Rapid fault detection and classification is of core importance fothe utility in order to quickly dispatch repairing teams to insure power restoration to the affected areas reducing the power outage duration and also provide maintenance to the valuable assets such as the distribution transformer in order to prevent further catastrophicfailure of the expensive equipment. In this research thesis a non-contact magnetic flux leakage measurements associated with distribution transformer are utilized to trainmachine learning algorithms in order to perform short circuit fault detection and classification and actively monitor the distribution transformer for such faults. For thispurpose, modelling and electromagnetic simulation of a utility distribution transformehave been performed using the method of finite element analysis. Three sensor position have been explored to measure the flux leakage outside the steel tank of the transformerand utilize it for fault detection and classification using three state of the art machine learning algorithms namely k-nearest neighbour with dynamic time warping, supportvector machine and artificial neural networks. In total 10 fault scenarios are explored andthe artificial neural network achieved 98% accuracy in correctly detecting and classifyingall 10 fault scenarios alongside the normal case scenario. The algorithm performed itsprediction for 669 case scenarios in less than 0.16 seconds making it the fastest algorithmto detect and classify the faults.