Abstract

The Power Transformer Health Index (PTHI) has become an increasingly suitable condition evaluation method for driving maintenance strategies for electrical utilities. Many researchers considering PTHI lack practical field experience and Asset Management (AM) practice. AM practice has not been considered or reported widely as a vital factor for improving PTHI.
The thesis aims to integrate the researcher's technical field experience besides the AM standard ISO55000 framework and obtain a new PTHI health condition model. Developing a good analysis and understanding of the actual data, the PTHI model, and the benefits of driving PT asset management maintenance strategy is vital. This research employs a triangulation research methodology, a combination of qualitative, descriptive studies, and quantitative analysis, strengthening the overall validity, quality, reliability, accuracy and findings of the research PTHI model. This research uses detail condition assessment using conventional HI, Fuzzy Logic HI and eight Machine Learning (ML) algorithms, including Binary Logistic Regression (BLR), K-Nearest Neighbors (KNN), Naive Bayes (NB), Random Forest (RF), Decision Trees (DT), Artificial Neural Networks (ANN), and Support Vector Machines (SVM) for developing PTHI health condition models as binary and multi-class classification problems and in addition, using linear and polynomials regression models.
The research contributed to multiple publications and a better understanding of ML evaluation metrics for the imbalanced dataset. For the first time, the F1 score was introduced in this field in this research area compared with other researchers who used only overall accuracy. This research has also updated the transformers diagnostic condition assessment HI factors data interpretation list. The detailed analysis and the study led to two research impacts: the technical impact of using the PTHI web app tool to support AM and field maintenance engineers and the financial impact of the PTHI model. The financial analysis shows a vast saving impact for implementing the ML PTHI model, with results of 59.9% for considering early replacement plan decisions. The research recommended developing the international best practices of AM standard ISO 55000 framework and culture for the electrical utility. This development will support the effective PTHI model for implementing the data-driven decision-making approach in diverse situations and breaking the silo among departments.