Low air pressure partial discharge recognition using statistical analysis of time-domain pulse features

Loading...
Thumbnail Image
Date
2019
Authors
Shahabi, Saeed
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The heavy mechanical, hydraulic and pneumatic based systems in modern aircraft are going to be replaced by electrical systems to make them more compact and light. The growth of electrical equipment in modern aircraft is achieved by increasing the voltage of electrical power supplies that may cause partial discharges (PD). At higher voltages, the partial discharges can be disruptive and cause insulation failure. Furthermore, an aircraft experiences a wide range of operating pressures during ascending and descending. The air pressure at high altitudes drops as low as 30%. It is widely known from Paschen's law that the dielectric strength of air decreases with altitude and hence increases the risk of partial discharges (PD). The performance of electric power system components of an aircraft must be reliable at high altitude under sub-atmospheric pressures. Electric actuators used in more-electric aircraft are fed by inverter drives that generate pulse width modulated (PWM) voltages. Under sub-atmospheric pressures, these PD signals are covered by the interfering signals from inverter that makes them di cult to be detected. Because of this, PD activity measurement under sub-atmospheric pressures has been a topic of interest for the evaluation of aircraft insulation system. The ultimate goal of this dissertation is to show a powerful diagnostic method for the evaluation of insulation condition and PD source recognition under sub-atmospheric pressures. A method based on the combination of wavelet and energy techniques is proposed to detect PD pulses in an extremely polluted noisy environment under typical aircraft's operating air pressure. For separation of PD sources, the time-domain features are calculated from PD pulse signal. Three of the features are selected to make a three dimensional (3D) space and the calculated features of all PD signals are mapped in the 3D space for separation of superimposed PD sources. The statistical moments of feature distributions are used for classification of PD sources. In order to have a good combination of available statistical moments and high speed of classification, kernel support vector machine (KSVM) algorithm is employed as a classifier for PD recognition under sub-atmospheric pressures.
Description
Keywords
Partial Discharge, Low Air Pressure
Citation