The use of wavelet packet transform for digital-signal processing of PD data acquired from tests on inverter-fed specimen

When feeding an induction motor by means of an IGBT inverter, the electrical and thermal stresses on the electrical insulation can be different from those deriving from a pure sinusoidal voltage waveshape, which can lead to sensibly shortened device life-time. A fundamental role on the premature failure of the electrical insulation is played by partial discharges in the dielectric; in order to clarify such aspect, measurements under real operating conditions are needed. Under such conditions traditional acquisition on limited bandwidth (∼400 kHz) oscilloscopes, according to IEC 270, is extremely difficult due to the presence of steep front pulses in correspondence with the power devices commutations which can completely hide the information given by the partial discharge; in such cases ultra-wide band (bandwidth ∼1 GHz) detection must be performed. However, due to the extremely noisy environment, a digital signal-processing method must be adopted in order to separate the PD pulse from the transient voltage signal due to the commutations. In a previous work the authors have shown the first attempt of using the wavelet packet transform and the Daubechies wavelet family for the extraction of PD pulses. However, since the accuracy of the results can depend on the choice of a suitable mother wavelet, the aim of the present work is to deeply investigate on the WPT by comparing the performances of different wavelet families.