Interpretation of transformer winding deformation fault by the spectral clustering of FRA signature

Frequency response analysis (FRA) has been accepted as a widely used tool for the power industry. The interpretation of FRA can be achieved by the conventional mathematical indicators-based method, which is mostly used in the past. The newly developing artificial intelligence (AI)-based method also provides an alternative interpretation. However, in most reported AI techniques, the features of FRA signatures are directly input into the AI model to obtain the classification results. Few studies have concentrated on the separability of winding deformation faults. In this context, a spectral clustering algorithm is studied to aid in FRA interpretation. The electrical model simulation and experimental tests are performed. The FRA data processing results verify the feasibility, effectiveness and superiority of the proposed method.

Automated summary

Frequency response analysis (FRA) is widely used in the power industry, with interpretation traditionally relying on mathematical indicators or newer AI methods. A spectral clustering algorithm has been investigated for separating winding deformation faults, showing feasibility, effectiveness, and superior results in FRA interpretation.