A Novel Deep-Learning Framework to Identify and Locate Single and Multiple Partial Discharge Events
Rakesh Das, Arup Kumar Das, Soumya Chatterjee, Arpan Kumar Pradhan, Subrata Biswas, Sovan Dalai, Biswendu Chatterjee, Kesab Bhattacharya
Abstract
In this contribution, a novel deep-learning framework is proposed for accurate localization of single and multiple partial discharge (PD) events employing optical sensor signature. An experimental setup has been fabricated to generate single as well as multiple events at different locations and for each case, PD signature is recorded using five optical sensors. The local fluctuations in the acquired PD signatures are initially analyzed using 1-D local binary pattern. The local binary pattern-transformed PD signatures corresponding to each sensor are then fed to a configured multichannel fusion hybrid deep-network consisting of a convolutional neural network and bidirectional long short-term memory network to classify the location of PD events. Investigations revealed that the proposed network is able to classify PD events with good accuracy. In addition, the performance of the proposed network is found to be better compared to existing methods for PD detection using optical sensors.