Reducing PD Interpretation Errors: Seven Judgement Mistakes That Produce Wrong Conclusions
1. Measurement error versus judgement error
These have different causes and different fixes.
A calibrated, well-executed test can still produce a wrong conclusion. The seven mistakes below are the usual route.
2. Mistake 1 — Classifying noise as PD
How it looks: a discharge-like signal that is accepted because it is "repeatable", without checking whether it moves with the test voltage.
Why it misleads: corona from nearby hardware, VLF source switching noise and communication equipment all repeat. Repeatability alone proves nothing.
Counter-check: raise the test voltage in steps and watch the signal. PD responds to voltage in a characteristic way; most external noise does not. Confirm the signal disappears when the source is isolated.
3. Mistake 2 — Treating a single reading as a verdict
How it looks: one PD value at one voltage, reported as the condition of the asset.
Why it misleads: PD is statistically distributed and varies with temperature, moisture and recent load history. One reading describes one moment.
Counter-check: take at least two measurements at the same voltage point, and note the spread. If the spread is large, the conclusion is about the spread, not the value.
4. Mistake 3 — Ignoring phase information
How it looks: reporting apparent charge in pC without reference to the phase-resolved pattern.
Why it misleads: magnitude alone cannot separate a surface discharge from an internal void, and cannot distinguish PD in the cable body from PD at a termination — the two have very different consequences.
Counter-check: always record and attach the phase-resolved pattern alongside the magnitude. The pattern is the evidence for defect identification.
5. Mistake 4 — Comparing incompatible measurements
How it looks: "PD has doubled since the last test", where the last test used a different instrument, bandwidth or method.
Why it misleads: apparent charge is defined by the measurement circuit. Values from different configurations are not on the same scale.
Counter-check: compare only like-for-like — same instrument family, bandwidth, method, voltage sequence and coupling configuration. Where the method changed, record the change and treat the series as discontinuous.
6. Mistake 5 — Reading magnitude without location
How it looks: a PD level quoted as a severity figure with no distance, joint reference or termination identified.
Why it misleads: the same apparent charge means different things at a joint, at a termination and mid-span. Without a location, the result cannot become a work order.
Counter-check: pair every reported PD figure with its position — from PD mapping, correlated against joint positions from as-built records.
7. Mistake 6 — Ignoring voltage dependence
How it looks: a PD level quoted at one voltage, with no inception or extinction voltage recorded.
Why it misleads: the relationship between PDIV, PDEV and the operating voltage determines the margin. A discharge that starts well above operating voltage and extinguishes far below it behaves very differently from one that is active at service stress.
Counter-check: record PD inception and extinction voltage as part of the test, and report the margin to operating voltage, not just the pC figure.
8. Mistake 7 — No baseline, no trend
How it looks: a conclusion drawn from a single test campaign, with no reference to the commissioning measurement.
Why it misleads: PD activity that is stable over years is a different asset risk from PD activity that has grown over months.
Counter-check: compare against the asset's own baseline first, then against similar assets. Where no baseline exists, state that the conclusion is a snapshot and set the next measurement date.
9. Six checks before signing a PD conclusion
A conclusion that cannot pass all six is not yet a conclusion — it is an observation, and should be reported as one.
FAQ
Do interpretation mistakes mainly come from software? No. Automatic pattern classification is a support tool, not a decision. Most wrong conclusions come from how the human reading the output frames the question — absence of a baseline, an unstated location, an invalid comparison.
How many measurements are needed before concluding? At least two at each voltage point, plus the inception and extinction voltages. The purpose is not statistical rigour for its own sake: it separates a real phenomenon from a transient.
Can a PD result be too small to matter? Yes — and that is a legitimate conclusion, provided the noise floor, the voltage margin and the location are recorded. "No PD above the detection threshold at the stated voltage" is a defensible finding; "no PD" is not.
What is the most common single mistake? Concluding without a baseline. It converts a condition assessment into a snapshot, and makes the next test the first comparable measurement rather than the second.
How should uncertainty be documented? Record the noise floor, the measurement spread, the test voltage sequence, the calibration factor and the location confidence. A conclusion with its uncertainty stated is more useful than a precise number without one.
Related reading
- How to Interpret Partial Discharge Test Results — magnitude, PDIV/PDEV and pass/fail criteria
- Common Mistakes in On-Site PD Testing — the field measurement errors that precede interpretation
- PD Testing Equipment for Commissioning — calibration and measurement hardware
- Partial Discharge vs Tan Delta — choosing the diagnostic that answers your question