New approaches to help improving the interpretation of frequency response analysis (FRA) in power transformers

Frequency response analysis (FRA) is a common technique used to detect mechanical deformations in power equipment such as power transformers. The measurements record a unique “fingerprint” of the transformer, which can be compared to a reference measurement. However, the interpretation of the frequency responses sometimes remains very vague and still does not make it possible to locate the incipient fault. The main objective of the research project is to contribute to a method for interpreting FRA curves in a more objective way. For this, machine-learning algorithms are developed using a transformer model designed in the laboratory. The model is unique in the sense that it allows the non-destructive interchange of healthy and distorted winding sections and, hence, reproducible and repeatable FRA measurements. Different categories of faults (axial displacement, radial strain and short circuits) and other conditions (temperature, presence of earthed structures near the coils, etc.) affecting the frequency response of the transformer can therefore be considered.