Reverse recovery occurs within a switching and commutation network containing devices, buses, inductances and capacitances. Part of the pulse can circulate locally. The transformer sees the portion carried by its connected path, which depends on that network. IEC 61378-1 covers industrial converter-transformer specification, design and testing [1].
Draw the high-frequency current loop, not only the power-frequency schematic. Include device junction behavior, bus geometry, winding parasitics and any filtering or local capacitive path relevant to the pulse.
Establish the observation boundary
| Measurement location | What it observes | Interpretation limit |
|---|---|---|
| Device lead | Local recovery and commutation current | Not automatically winding current |
| Secondary terminal lead | Current at that boundary | May include local capacitive components |
| Winding model branch | Defined internal contribution | Depends on model representation |
| Primary terminal | Transferred and network-filtered response | Does not reveal every local secondary pulse |
Use synchronized measurements or a validated network model to compare these locations. A pulse visible at one boundary can have a different magnitude and shape elsewhere.
Integrate the pulse that belongs to the conductor
For an assumed isolated triangular pulse of peak IP and total duration τ, ∫i²dt = IP²τ/3. If identical nonoverlapping pulses reach the winding at repetition frequency fr, their mean-square contribution is IP²τfr/3. Current is in amperes, duration in seconds and repetition in hertz.
This relation applies to the measured or modeled winding-visible pulse. Substituting a semiconductor data-sheet peak without the propagation network does not establish winding RMS.
Include ringing and baseline current when relevant. Cross terms can matter when pulses overlap another correlated current component; do not add unrelated RMS quantities arithmetically.
Verify measurement capability
Sensor bandwidth, placement and acquisition rate must resolve the pulse and its timing. A probe with inadequate response can understate a peak, while pickup from nearby switching fields can create an apparent current spike.
Keep probe transfer behavior and the analysis window with the result. A wideband peak and a power-frequency RMS reading answer different questions. Avoid using one to declare the other negligible.
Calculate the pulse contribution at the winding boundary
Assume isolated triangular pulses actually reaching the winding, peak 100 A, total width 10 µs and repetition rate 1000 Hz, with no overlapping baseline current. Mean-square contribution is 100² × 10 × 10⁻⁶ × 1000/3 = 33.33 A², giving 5.774 A RMS. Charging a 200 A device-lead peak to the same winding instead gives 11.547 A RMS and four times the pulse I² contribution. Use the winding-visible pulse for conductor duty and a wideband loss model for its frequency-dependent heating.
Assess the relevant loss mechanism
Even a small total RMS contribution can require examination where high-frequency current distribution changes local loss. Use a frequency-appropriate conductor or winding model and the actual spectrum. Conversely, a dramatic device peak of very short duration should not be charged entirely to the winding without evidence that it follows that path.
Use the winding-boundary waveform to connect device recovery with transformer duty. Preserve the recovery loop, sensor capability and pulse repetition assumptions when converting the observed contribution into RMS or frequency-dependent loss.
Retain an explicit measurement model and input uncertainties when comparing measured and calculated duty [2].
References
- IEC, IEC 61378-1:2011, Converter transformers – Part 1: Transformers for industrial applications.
- JCGM, JCGM 100:2008, Evaluation of measurement data — Guide to the expression of uncertainty in measurement, 2008.