Partial discharge diagnostics on inverter-fed drives of electric vehicles

In principle, the methods from IEC/TS 61934 can be used to measure PD at steep voltage edges. It is important here that, unlike IEC 60270, PD is not measured in the low frequency range below \(\mathrm{1\,MHz}\). Drive converters cause interference up to some hundreds of MHz, thus causing any measurement below these interference frequencies impossible. For this reason, IEC/TS 61934 suggests measurements in the higher frequency range of approximately \(\mathrm{1\,GHz}\) and the use of appropriate high-pass or band-pass filters.

Three electrical sensor methods can be used to couple out the PD signals. These are decoupling via a coupling capacitor combined with a coupling quadripole, via an antenna or a high-frequency current transformer (HFCT). The measurement via external coupling capacitors is avoided in this work, since these do not occur in real applications for drive motors. The use of parasitic capacitances as coupling capacitance is difficult, since they are usually not precisely known and can change with rotor position. Antennas can be used to measure PD reliably. However, metallic housing parts can lead to shielding of the antenna. HFCT can be inserted directly into the feed lines. There is no shielding effect due to housing parts. However, the inductive motor windings lead to attenuation of the PD signal. Depending on the objective of the measurement and the device under test, both methods are suitable. In this work, both methods will be compared in order to evaluate a selection of suitable sensors for PD measurement. The focus in this work is on passive sensors for better comparability. An optimization with active components would be conceivable after the basic consideration. Table 1 shows the sensors used. The ball probe and stub probe are part of a commercially available near-field probe set for EMC test laboratories. There, they are normally used to detect sources of interference.

In order to suppress the interference range of the inverter, it is essential to use high-pass filters. In several own works [3‐5] it has been shown that passive high-pass filters of 7th order directly at the sensor provide sufficient attenuation of the interference signal. Acoustic and optical methods can be used as alternative PD measurement methods, which serve as validation methods in this work.

A wide toolbox is available for the evaluation of the measured signals. Good results could be achieved, if the interfering signals of the source are already pre-filtered with high-pass filters before the data acquisition. Some PD signal power in the lower frequency range is also suppressed using this method. Only in the higher frequency range does the signal components of the PD pulse outweigh the interference.

In [5] some evaluation possibilities in the time domain, frequency domain and in the combined time-frequency domain were presented. The simplest way to evaluate the prefiltered signals is directly in the time domain.

For the verification of possible influencing factors on the partial discharge at the test specimens from Sect. 4 different voltage sources are required. In order to be able to simulate steep-edged square wave voltages of modern inverters in the laboratory environment, a 3‑level neutral point clamped inverter with IGBTs was built in [5] and a full-bridge inverter with SiC MOSFETs in [3]. Compared to commercially available inverters, the voltage shape can thus be adjusted much more flexibly, since there is access to all characteristic parameters. To ensure a sufficiently large voltage reserve for devices under test (DUTs) with a high PD inception voltage (RPDIV) and in case of voltage overshoots, semiconductors with a reverse voltage of 1.7 kV are selected.

Furthermore, test generators for EMC testing technology can be utilized to generate standardized, pulse-shaped voltages. Burst generators, for example, simulate transient noise interference with high slew rates. This allows to investigate the influence of high-frequency interference up into the MHz range.

For PD measurements, according to IEC/TS 61934, the following properties should be used to characterize the voltage pulses generated by the test voltage sources:

The self-developed full-bridge SiC inverter is used as a universal test voltage source in this paper. Figure 1 shows a photo of the 19′′ rack-mount chassis of the inverter. The inverter is constructed using two identical SiC half-bridge modules. The two MOSFETs of a half-bridge module are driven via a plug-in gate driver board.

The SiC inverter is fed from a DC link. A high-voltage generator is used as DC voltage source. Since the generator cannot be used as a sink, additional discharge resistors are integrated in the DC link setup. A capacitor bank of film capacitors in an additional chassis serves as energy storage for stabilizing the DC link voltage. Snubber film capacitors are located directly in the inverter chassis to provide the quick current pulses for the fast-switching SiC semiconductors. This avoids overshoots directly on the semiconductors and thus contributes to safety.

Table 2 shows all adjustable parameter values of the SiC inverter and shows the classification of the inverter according to IEC/TS 61934. As a special feature, the slew rate can be varied by the gate drivers. This is implemented using a resistive open-loop gate driver topology with a switchable external gate resistor array. In [3], the influence of the slew rate on the RPDIV was investigated on twisted-pair DUTs in this way. Via a graphical user interface, the parameters can be changed during operation on the laboratory PC.

The measurements at sine voltage are used to validate the sensors. This is necessary because it is very important to be able to distinguish the PD signals from possible interferers. At 50 Hz, standardized PD measuring devices can be used for comparison, which was also done in this work. The procedure is described in [4]. The results show that the sensors can detect PD in accordance with the standard PD measurement device.

Figure 5 shows the results for a sinusoidal voltage with a frequency of 50 Hz and an amplitude of 1295 V. This voltage is slightly above the PD inception voltage as a steady state for the occurrence of PD is to be achieved. Any self-healing processes [3] should therefore not be considered. The measurement was performed in envelope mode to determine a distribution of PD pulses. It can be clearly seen that PD can be detected with HFCT1, HFCT2 as well as with the ball probe. In contrast, the stub probe is significantly less sensitive and does not show any peak signal. It can also be seen that the HFCT provides a higher signal amplitude than the ball probe. However, as shown in [4], the amplitude level of the HFCT sensors is strongly dependent on the damping of the winding and consequently on the positioning of the sensors and the locality of the PD.

Since electrical machines are subjected to steep-edged voltages during inverter operation, their effect must be investigated. For this purpose, there are already some works such as [1] or [11]. Nevertheless, since here according to IEC/TS 60034-27‑5 the winding is open it is important for this work to analyze the used stator in more detail. For this reason, an impedance analyzer is first connected instead of the inverter from Fig. 3 to determine possible resonance points. The line resistance, the inductance of the winding and the supply line and the leakage capacitances are effective here and form an LCR series resonant circuit in the first approximation.

Figure 6 shows on the left axis in double logarithmic scaling the magnitude of the absolute impedance. On the right axis the phase response is plotted. At a frequency of 91.7 kHz a resonance point is clearly visible.

Figure 7 shows the measured voltage waveform with the inverter connected according to Fig. 3 and a DC link voltage of 575 V. V1 in black shows the measurement at the input of the device under test and V2 in green at the open winding. As an overview, a whole period is plotted in Fig. 8, the ringing is resolved higher. The measured frequency of the overshoot is 84 kHz. The deviation from the behavior shown in Fig. 6 can be explained by the fact that the cables to the inverter are added to the setup, which slightly shift the resonance point due to their inductance.

Now the stator and the motor are examined for PD according to Fig. 3. The self-developed SiC converter is used for this purpose. The fundamental frequency of the square wave voltage is set to 1 kHz. The slew rate is set to \(6.5\,\mathrm{kV}/\upmu\mathrm{s}\). Figure 8 shows the results on the falling edge of the stator. The falling edge is only used as an example. With regard to the PD behavior, there are no discernible differences between the falling and rising edge. The PD events were superimposed in envelope mode. In the upper graph, the resonance can be seen in green. In the middle, the PD events measured with HFCT1 are plotted. In the lower plot the results of the near field probes are shown. All measurements were recorded synchronously at the same time. PD can thus be detected with HFCT as well as with the ball probe.

Figure 9 shows the same measurement on the motor. Since the probes cannot be inserted into the stator here, they were placed above the open terminal box. It is noticeable that the ball probe can detect PD despite the closed housing. A verification with a UV PD localizer ruled out the possibility of external PD. It would be possible that the PD is emitted via the connection cables, or that it is propagated through the opening of the terminal box to the motor. With regard to the phase position, the measurements are comparable to the measurement with sinusoidal voltage. This is an interesting result, since the frequencies of the voltages are so far different from each other. The resonance frequency at the motor is slightly smaller than at the stator. The damping, on the other hand, is much greater. As a result, the overshoot decays more quickly and its amplitude is also not as pronounced.