Murdoch University Research Repository

Welcome to the Murdoch University Research Repository

The Murdoch University Research Repository is an open access digital collection of research
created by Murdoch University staff, researchers and postgraduate students.

Learn more

Development of a quantitative health index and diagnostic method for efficient asset management of power transformers

Islam, Md Mominul (2017) Development of a quantitative health index and diagnostic method for efficient asset management of power transformers. PhD thesis, Murdoch University.

PDF - Whole Thesis
Download (3MB) | Preview


Power transformers play a very important role in electrical power networks and are frequently operated longer than their expected design life. Therefore, to ensure their best operating performance in a transmission network, the fault condition of each transformer must be assessed regularly. For an accurate fault diagnosis, it is important to have maximum information about an individual transformer based on unbiased measurements. This can best be achieved using artificial intelligence (AI) that can systematically analyse the complex features of diagnostic measurements.

Clustering techniques are a form of AI that is particularly well suited to fault diagnosis. To provide an assessment of transformers, a hybrid k-means algorithm, and probabilistic Parzen window estimation are used in this research. The clusters they form are representative of a single or multiple fault categories. The proposed technique computes the maximum probability of transformers in each cluster to determine their fault categories.

The main focus of this research is to determine a quantitative health index (HI) to characterize the operating condition of transformers. Condition assessment tries to detect incipient faults before they become too serious, which requires a sensitive and quantified approach. Therefore, the HI needs to come from a proportionate system that can estimate health condition of transformers over time. To quantify this condition, the General Regression Neural Network (GRNN), a type of AI, has been chosen in this research. The GRNN works well with small sets of training data and avoids the needs to estimate large sets of model parameters, following a largely non-parametric approach. The methodology used here regards transformers as a collection of subsystems and summarizes their individual condition into a quantified HI based on the existing agreed benchmarks drawn from IEEE and CIGRE standards. To better calibrate the HI, it may be mapped to a failure probability estimate for each transformer over the coming year. Experimental results of the research show that the proposed methods are more effective than previously published approaches when diagnosing critical faults. Moreover, this novel HI approach can provide a comprehensive assessment of transformers based on the actual condition of their individual subsystems.

Item Type: Thesis (PhD)
Murdoch Affiliation(s): School of Engineering and Information Technology
Supervisor(s): Lee, Gareth and Hettiwatte, Sujeewa
Item Control Page Item Control Page


Downloads per month over past year