EV Inverter Failure Analysis: Correlation of Power Parameters with Waveform Data

WT5000 Precision Power Analyzer

Background

When measuring root mean square (RMS) values of voltage and power from distorted waveforms, such as those produced by PWM motor drives, precision power analyzers with high sample rates are necessary to perform calculations of high frequency components. Additionally, for troubleshooting and debugging, these measurements require simultaneous observation and capture of the voltage and current waveforms. To achieve this, oscilloscopes or high-speed data loggers with insulated or differential inputs are commonly used alongside power analyzers.
Some power analyzer models can capture sampled raw waveform data or equivalent waveform data simultaneously with power parameters. While RMS and other power parameters can indicate the presence of abnormalities, observing the original waveform data provides insights into root cause.

Challenges

Until recently, Yokogawa’s solution for this scenario is the PX8000, a transient waveform capture instrument with power measurement capabilities. Waveform data and power parameters can be analyzed and correlated directly with the instrument or with its dedicated software PowerViewerPlus 760881. On the other hand, for applications with the most demanding accuracy requirements, the WT5000 precision power analyzer is increasingly recommended.
The WT5000 has a data streaming (/DS) option to capture raw waveform data. However, waveform analysis requires exporting of data IS8000 software for PC-based analysis.

Data Streaming (DS) Function of the WT5000

The IS8000 cycle statistics computation function calculates waveform parameters using observed data acquired by waveform measuring instruments, such as high-speed data loggers (ScopeCorders), DLM series oscilloscopes, and the DS function of the WT5000. This function computes waveform parameters for each cycle based on the periodic signals in the waveform data.
Waveform parameters include items related to the voltage vertical axis (peak to peak, amplitude, maximum, minimum, etc.), the time horizontal axis (rise time, fall time, frequency, period, etc.), and area-related items (Integ1TY, Integ2TY, Integ1XY, Integ2XY). The method for detecting the cycle of a specified signal is determined by each line setting—distal, mesial, and proximal lines—including a “%” setting for the input signal from 0 to 100, or a “Unit” setting where an arbitrary number representing a physical quantity can be input (see Figure 1).

Figure 1. Line settings in cycle detection

Among these items, defining the cycle detection method for root mean square values (RMS), which is a vertical axis computation item, and the average value of instantaneous power (AVERAGE), based on the multiplication of instantaneous voltage and instantaneous current using the waveform computation MATH function, enables the obtaining of more accurate results.

Computation Setting of RMS Values and Power Values

The sample rate of the WT5000 is up to 10 MS/s, however it is limited to 2 MS/s aggregate (current plus voltage) in its data streaming mode. When using WT5000 data streaming, it is important to set the rate as high as possible to maintain accuracy in RMS calculations.
Please refer to the table in Figure 2 to determine the maximum number of streaming channels for various sample rate settings.

Maximum waveform trace count

Figure 2. Correlation between the number of waveforms and the
maximum sample rates

The actual operations and settings are as follows:

1. Enable DS Function: In DAQ (WT) setting of the IS8000, turn on the DS function to capture waveform data. At this step, set the sample rate as high as possible (see Figure 3).

Figure 3. Setting DS function in the IS8000

2. Configure Math Operations: For the acquired waveform data, use MATH function to set [voltage x current] (see Figure 4).

Figure 4. Setting arithmetic operations in the IS8000
(set an arithmetic operation for power in MATH 1)

3. Cycle Statistics Computation: in the IS8000 waveform parameters menu, enable cycle statistics. Select Rms measurement for voltage (WU1) and current (WI1) traces, then select Avg measurement for Math1, as instantaneous power must be averaged to obtain the power value (see Figure 5).

Figure 5. Cycle statistics settings in IS8000 Software

4. Cycle Detection Settings: for more accurate cycle detection, set the distal, mesial, and proximal lines for the least noisy and most sinusoidal waveform (see Figure 6).

Figure 6. Channel settings in the IS8000

Verification by Reference Input (LS3300)

For maximum accuracy, cycle detection settings (distal, mesial, and proximal lines) should be verified with a reference, such as the Yokogawa LS3300, a power calibration standard. With its waveforms input to the WT5000, IS8000 cycle detection settings are adjusted until the computed power values match the WT5000 measured power. With this calibration step, the measured results of the IS8000 will be aligned to a known reference.

Figure 7. Connection between the LS3300 and the WT5000

Figure 8. WT5000 measured values from LS3300 output

Figure 9. IS8000 cycle statistics computation, using the current
waveform for cycle detection

These results show that the average values of the IS8000 cycle statistics will be almost identical with the measured values of the WT5000.

Verification by Inverter Input

Next, we verified the from PWM inverters using the measured values of the WT5000.
For voltage-type PWM inverters, the voltage waveform resembles a pulse-like pattern, making it difficult to detect the cycle of the fundamental wave frequencies. Similarly, the current waveforms have triangular waves of the carrier frequency superimposed on them, and in cases of low amplitude input, noise is more likely to be present, complicating cycle detection.
Therefore, we input a current closer to a sine wave to the unused input element (CH4), with the line filter setting to ON to create an ideal sine wave with minimal noise. This waveform was then used for cycle detection settings (see Figure 9). We also performed a comparison with the ‘Own’ setting, which is the default setting for cycle statistics computation, and the difference was evident.

Figure 10. Connection method for cycle detection reference signal

Figure 11. WT5000 waveforms and measurements
(Inverter voltage waveform (U1), current waveform (I1), and current waveform (I4) with a line filter)

From Figure 11, the measured values obtained from the WT5000 serve as reference values for comparison with those from the cycle statistic computation of the IS8000. Additionally, the waveform of element 4 at the bottom of the screen, used for cycle detection, appears as a sine wave. This makes its zero-crossing points more easily distinguishable compared to the element 1 voltage waveform, which resembles a PWM pulse, or the current waveform, which is affected by significant noise.

Figure 12. Measurement results when CH4 is used for cycle detection

Figure 13. Results with default (Own) cycle detection settings
(power values differ greatly)

Summary

We examined a case of computing power from waveform data obtained by the WT5000 data streaming function using the IS8000 cycle statistics computation.
Despite the limitation of eight waveforms at maximum for the cycle statistics computation, we found that applying a line filter to remove noise from the signal used for cycle detection and adjusting the cycle detection setting to a zero-cross level leads to results almost identical to the values measured by the WT5000 itself. Waveform data is valuable information to analyze during root cause analysis of transient events, and it is useful and important to correlate waveforms with numeric data.