Caracterização de atributos de sinais para utilização de técnicas de reconhecimento de padrões na avaliação do isolamento de instalações e de equipamentos elétricos.

Abstract

Contamination of outdoor insulators employed in the overhead high voltage transmission lines in the polluted geographic regions is one of major causes of electric power system failure. The partial discharges within the body of an insulation material also progressively lead to the irreversible deterioration in the performance of the electrical equipment. Data acquisition systems for automatically monitoring the signals related with insulator contamination levels as well as partial discharges, can help in reducing the electrical system failures In view of the practical difficulties in the development of specific analytical models for these phenomena, signal, pattern recognition techniques have been employed to interpret the acquired signals to classify the state of insulator contamination and predict the potential occurrence of system failure. The main emphasis of the present investigation is the characterization of the attributes of the monitored signal, with a view to obtain improved performance the existing pattern recognition techniques. Digital signal processing techniques like digital Fourier transforms and Wavelet transforms have been employed as an aid to characterize the signal attributes. Various evaluations of the number and types of the characterized signal attributes are described. The procedures are outlined for choosing the most significant mother wavelet to be utilized in the recognition of the pattern of the attributes of the acquired signals. Two case studies are presented. In the first one, a method is developed to predict the level of contamination of the overhead high voltage transmission line insulator strings based on linear and nonlinear pattern recognition of the attributes of the acquired signals. This case can be considered to represent an analysis of data fusion originating from multiple sensors employed for data monitoring. In the second case study, signal pattern recognition techniques are employed to diagnose the defects in polymeric insulators utilizing the attributes of the partial discharge signals.