Deep residual neural network for EMI event classification using bispectrum representations

Mitiche, Imene and Jenkins, Mark David and Boreham, Philip and Nesbitt, Alan and Stewart, Brian G. and Morison, Gordon; (2018) Deep residual neural network for EMI event classification using bispectrum representations. In: 26th European Signal Processing Conference (EUSIPCO). IEEE, Piscataway, NJ. (In Press)

[img]
Preview
Text (Mitiche-etal-EUSIPCO2018-Deep-residual-neural-network-for-EMI-event-classification)
Mitiche_etal_EUSIPCO2018_Deep_residual_neural_network_for_EMI_event_classification.pdf
Accepted Author Manuscript

Download (1MB)| Preview

    Abstract

    This paper presents a novel method for condition monitoring of High Voltage (HV) power plant equipment through analysis of discharge signals. These discharge signals are measured using the Electromagnetic Interference (EMI) method and processed using third order Higher-Order Statistics (HOS) to obtain a Bispectrum representation. By mapping the time-domain signal to a Bispectrum image representations the problem can be approached as an image classification task. This allows for the novel application of a Deep Residual Neural Network (ResNet) to the classification of HV discharge signals. The network is trained on signals into 9 classes and achieves high classification accuracy in each category, improving upon our previous work on this task.