Mathematical Morphology-Aided Bidirectional Long Short-Term Memory Network-Based Partial Discharge Pulse Sequence Classification

In this article, a novel approach for automated identification of partial discharge (PD) defects inside an insulation system is proposed employing PD pulse sequence analysis (PSA). The sequence of PD pulses is directly related to the type of PD defect in an insulation system. Therefore, the pattern of PD pulse sequence has been analyzed in this article to diagnose different types of defects. For this contribution, three common types of artificial defects have been emulated and pulse sequence pattern corresponding to each type of PD defect has been recorded. Following this, mathematical morphology (MM) has been used to analyze the PD pulse sequence pattern and based on morphological operations; several novel features have been extracted in this article to discriminate different PD pulses. The extracted features were fed to a bidirectional long short-term memory (Bi-LSTM)-based deep neural network (DNN) classifier. It has been noticed that the proposed Bi-LSTM network achieved an accuracy of 98.76% in discriminating different types of PD defects. Comparative study with other deep learning methods also indicates that the proposed MM aided Bi-LSTM is suitable automated classification of PD pulse sequence.