학술논문

Terahertz Detection of Interface Defects Within Composite Insulators Using a Gated Recurrent Neural Network
Document Type
Periodical
Source
IEEE Transactions on Dielectrics and Electrical Insulation IEEE Trans. Dielect. Electr. Insul. Dielectrics and Electrical Insulation, IEEE Transactions on. 31(2):704-712 Apr, 2024
Subject
Fields, Waves and Electromagnetics
Engineered Materials, Dielectrics and Plasmas
Insulators
Reflection
Logic gates
Air gaps
Imaging
White noise
Testing
Complementary ensemble empirical mode decomposition (CEEMD)
composite insulator
gated recurrent unit (GRU)
nondestructive testing
terahertz
Language
ISSN
1070-9878
1558-4135
Abstract
The interface performance of composite insulators is related to their overall electrical performance. Accurate identification of interface defects is an important means to ensure the safe and reliable operation of insulators. In recent years, terahertz technology has been applied to the defect detection field of composite insulators due to its strong penetration, high imaging resolution, and nondestructive. However, the traditional terahertz detection method has shortcomings of high labor cost, strong subjectivity, and low efficiency. Therefore, the complementary ensemble empirical mode decomposition-gated recurrent neural (CEEMD-GRU) algorithm for terahertz detection of interface defects within composite insulators was proposed in this study to realize intelligent recognition and accurate imaging of interface defects. First, samples of silicone rubber-epoxy resin plate with air-gap defects of various thicknesses were prepared, and terahertz reflection scanning experiments were carried out on them. The relative error of defect thickness between calculated results based on terahertz time-domain waveform analysis and actual measured results was less than 6%. Then, the terahertz reflection spectral signal was decomposed by complementary ensemble empirical mode decomposition (CEEMD) to obtain the intrinsic mode function (IMF) subsequences. Calculating the energy ratio, Rényi entropy, Petrosian fractal dimension (PFD), and Hurst index of each subsequence, characteristic matrices were constructed and input into gated recurrent unit (GRU) to realize the intelligent identification of interlayer air-gap defects, and the detection rate reached 99.25%. Finally, the trained model was loaded for defect detection and imaging, and the maximum relative error of the defect area between recognition values and actual values was 5.27%, which verified the effectiveness of CEEMD-GRU in the intelligent recognition and detection of terahertz signals. This study provides a novel idea and method for the application of terahertz technology in the nondestructive testing of composite insulators.