Time-frequency sparsity map on automatic partial discharge sources separation for power transformer condition assessment

Partial discharge (PD) measurements can evaluate integrity of transformers' insulation systems. Current research focuses on multiple PD sources separation to identify the types of insulation defects that may coexist in a transformer. This paper proposes a time-frequency (TF) sparsity map for revealing and separating different PD sources. TF sparsity map is developed based on decomposing signals into time and frequency domains at multiresolutions. Two decomposition methods, conventional wavelet transform-based signal decomposition and novel mathematical morphology (MM)-based signal decomposition are implemented in this paper. After sparsity values are calculated from the decomposed signals in time and frequency domains, sparsity trends are determined to provide unique representation of PD sources. By taking roughness of the trends, an accurate separation of multiple PD sources is obtained on a TF map. A density-based clustering is then evoked to form clusters related to different PD sources. The proposed method has been verified by signals acquired from multiple PD source models and substation transformers. Results show that an accurate representation of PD pulses in the presence of multiple PD sources and subsequently separation of PD sources can be achieved. Comparisons of wavelet transform and MM-based signal decomposition methods on TF sparsity maps construction and multiple PD sources separation are also provided.