Abstract

In industrial environments, the partial discharge (PD) identification process can be limited by the simultaneous presence of PD sources and electrical noise. Therefore, it is advisable to previously carry out a source separation process based on the characteristic parameters of the captured signals in order to differentiate and analyze each source present during the measurement. After that, and to automate the diagnosis, clustering techniques are usually applied to label the points associated with the clusters established. Unfortunately, one of the main problems that usually occurs in most separation processes is the shape that clusters can take in the separation maps (2D or 3D), hindering the correct labeling by the clustering technique in use. In this paper, a novel clus- tering technique called Communication with Local Agents (CLA) is proposed to discriminate multiple PD sources and electrical noise. To evaluate the performance of CLA, three experimental configurations have been imple- mented, with multiple PD sources and electrical noise. The results show that the CLA technique outperforms other clustering techniques in terms of average error rate. In the three experiments carried out, CLA obtained the lowest error values: 1.2% and 12.27% in the two measurement processes where three clusters were present, and 6.97% in the measurement where up to four different clusters were generated.