LASALVIA, L. A. B.; http://lattes.cnpq.br/7991407879058293; LASALVIA, Leonardo Antonio Barbosa.
Resumen:
This paper presents the development of a predictive maintenance technique for detecting defects in porcelain insulators, using the method of acoustic emission to generate chart patterns that identify the state of integrity of electrical equipment. These insulators are widely used in buses substations which form the electrical power system. In order to achieve this goal, acoustic emission testing in the field
were conducted in substation Garden, the Companhia Hidroelétrica do São Francisco to obtain the audible noise which in turn were stored, processed and assembled into a database for later implementation computing. We used the wavelet Transform continues in order to identify the most appropriate wavelet family for the study of the sound signal captured in the field trials. Were analyzed and compared
the discrete Daubechies wavelet families, Coiflets, Symlets, Discrete Meyer, Biorthogonal and Reverse Biorthogonal. The next step was to decompose a signal multilevel analysis using wavelet Packet Transform matrices for generating energy of wavelet coefficients. Finally, to add reliability, automation, ability to generalize and
adapt to new situations, we used an artificial neural network, perceptron with three layers, associated with Resilient Propagation learning algorithm to classify the desired patterns (upright insulators and insulating defective) from the energy matrices generated by wavelet Packet Transform, thus validating the method. Results were recorded above 85% accuracy.