Developing a New Decision-making and Interpretation Technique to Support Industry for Transformer Frequency Response Data Analysis using Machine Learning and Bolstered Error Estimation

Faculty Development Competitive Research Grant Program 2021-2023

Frequency Response Analysis (FRA) is the most sensitive, non-destructive, fast, reliable and industrialized method in transformer winding deformation detection; even though, its interpretation is still remained as a challenge for industrial/transformer operators and expertise. Making decision on transformer FRA data is yet unautomated and practically performed based on expertise knowledge. Hence, the main objectives of this proposal are,
•To identify decision boundaries for statistical indicators in inter-turn/inter-disk short circuit fault as well as transformer winding radial and axial displacement,
•Developing an automated system and programmed structure to categorize the normal, suspicious and critical conditions of transformer in inter-turn/inter-disk short circuit and also winding radial and axial displacement faults,
•Developing univariate and multivariate classification confidence level estimation for decision-making on transformer FRA data in industry using Bolstered Error Estimation technique,
•Developing and providing a free-access web-based online engine with a desirable interface for public access to interpret industry/academic FRA data.