Diagnosis model of noise-type defects for dry-type transformer based on time–frequency-space tensors and improved prototypical network under small sample conditions

Abstract

This paper presents a diagnostic method for dry-type transformer noise-type defects using time–frequency-space tensors and an improved prototypical network. The method converts single-channel time domain sound signals into time–frequency matrices using a parameter-optimized variational mode decomposition algorithm and the Hilbert-Huang transform, and arranges the matrices in space. Sample enhancement via sound source virtual rotation and dimensionality reduction through hyper-principal component analysis are employed to enhance the model’s generalization capability, computational speed, and recognition accuracy. The improved prototypical network, fortified with a custom residual network as the encoder, improves the model’s ability to learn complex patterns and simultaneously diminishes the model’s computational complexity. Experimental results show that the proposed method has good recognition accuracy, computation speed, and stability with few samples. The recognition accuracy is 96.0% ± 2.1% with 15 samples per type of defect. This method demonstrates good applicability to sound signals collected by microphone arrays with rotational symmetry structure.