Title :
Application of ultrasonics and neural network techniques to the evaluation of stator bar insulation
Author :
Gleizer, H. ; Nelson, J.K. ; Azizi-Ghannad, S. ; Embrechts, M.J.
Author_Institution :
Dept. of Electr. Power Eng., Rensselaer Polytech. Inst., Troy, NY, USA
Abstract :
The combination of non-invasive acoustic techniques and neural network backpropagation methods is a promising tool to assess the aging condition of electrical insulation of hydrogenerator busbars. The efficacy of the method has been confirmed for laboratory aged samples. The networks showed quite reasonable generalization by the low percentage of misclassifications verified during the test phase. Mathematical preprocessing of the acoustic data plays an important role in the neural networks´ performance. Global normalization is indicated for a reduced number of measurements, while local normalization following combined Fourier and Wavelet transforms can result in more robust systems for a sufficiently large amount of data
Keywords :
ageing; backpropagation; delamination; electric machine analysis computing; feedforward neural nets; hydroelectric generators; insulation testing; machine insulation; machine testing; stators; ultrasonic materials testing; Fourier transforms; acoustic data; aging condition; backpropagation methods; electrical insulation; global normalization; hydrogenerator busbars; local normalization; mathematical preprocessing; neural network techniques; noninvasive acoustic techniques; stator bar insulation; ultrasonics; wavelet transforms; Acoustic measurements; Acoustic testing; Aging; Backpropagation; Dielectrics and electrical insulation; Fourier transforms; Laboratories; Neural networks; Robustness; Wavelet transforms;
Conference_Titel :
Electrical Insulation and Dielectric Phenomena, 1995. Annual Report., Conference on
Conference_Location :
Virginia Beach, VA
Print_ISBN :
0-7803-2931-7
DOI :
10.1109/CEIDP.1995.483710
