Pattern recognition of partial discharge faults using convolutional neural network (CNN)

Partial Discharge (PD) analysis is one the most widely used methods to monitor and determine the fault conditions of electrical equipment, especially in high-voltage environments such as power transformers and power generators. Conventional method of PD analysis that is widely used in multiple studies and commercial equipment usually rely on a feature extraction technique such as the Phase Resolved Partial Discharge (PRPD) pattern to assist PD experts to inspect the faults in the system. This study proposes a CNN based method to recognize the PRPD patterns for different types of PD. The differences of each type of PD, data pre-processing steps and visualization of PD waveforms in PRPD patterns are discussed in details. The obtained PRPD pattern images are then used to train a pattern recognition model and the results show that the proposed method can effectively classify different types of PD under consideration.