Distinguish partial discharge from electrical noise by using phase relationship, repetition, multi-sensor comparison and time, frequency and amplitude discrimination, then confirming with background and de-energised checks where appropriate. A single screenshot is never enough: classify the signal pattern, verify it with an independent measurement, and document the confidence level and escalation limit.
Why Noise Can Resemble a Partial Discharge Signal
Electrical noise shares many properties with partial discharge signals. Switching transients, communication signals, corona from unrelated conductors and interference from load equipment produce pulses that can appear in the same sensor band and even at similar amplitude. Without structured discrimination, the same sensor reading can be interpreted as a discharge in the asset or as a harmless external signal.
The goal is not to remove all noise but to classify the source. The classification uses several independent dimensions of evidence, and the conclusion is only as strong as the weakest dimension, so the workflow must be systematic rather than visual.
Common noise sources in a substation include power electronics, motor drives, radio and telecommunication transmitters, lightning arresters in operation and corona on adjacent conductors. Each has a characteristic signature: a fixed frequency, a repetitive timing, or an amplitude that tracks a specific load. Identifying the possible sources on the site is the first step in any discrimination plan, because it tells the operator which checks are most likely to resolve the question.
Use Phase Relationship and Repetition as Evidence
Partial discharges in AC insulation tend to repeat with a stable relationship to the applied voltage cycle, clustering at specific phase positions that reflect the discharge mechanism. Phase-resolved analysis therefore looks for repetition and phase stability: a signal that appears at the same phase window on every cycle is stronger evidence than a random burst. Random, non-repetitive pulses are more likely to be noise.
Repetition also applies over time: a real discharge pattern is usually present at every measurement under the same conditions, while noise may appear and disappear with load or weather. Record multiple cycles and multiple measurements, and compare the pattern stability before forming a conclusion.
Stability across measurement sessions is especially important: a signal that returns at the same phase window on different dates, with the same sensor position and similar load, is much stronger evidence than a burst seen once. Store the captures with their timestamps so the pattern can be re-examined later with fresh interpretation tools.
PRPD pattern shape is a classification clue: different discharge mechanisms produce characteristically different distributions of pulses across the AC cycle, while many noise sources produce either broadband scatter or narrow, fixed-phase spikes that do not match a discharge mechanism. Pattern recognition is an aid to classification, not a proof, and it must be combined with the other dimensions before the conclusion is written.
Compare Multiple Sensors and Measurement Locations
The strongest field discriminator is sensor comparison. Place two sensors at different positions on the same asset, or on different phases, and compare the signals. A signal that appears on the sensor near the suspected source and attenuates with distance behaves like a local discharge; a signal that appears identically on all sensors or changes with the environment behaves like external noise.
Multi-sensor comparison also distinguishes between phases: a discharge in one phase produces a different pattern than interference coupled from the whole bay. The comparison requires the sensors to be synchronised, which is why simultaneous multi-channel acquisition is a core feature rather than an option.
Sensor placement should be planned before the measurement: choose positions that are mechanically stable, electrically representative and repeatable, and mark them for future surveys. The attenuation test, where the sensor is moved progressively away from the suspected source, is one of the most direct ways to show that a signal is local rather than environmental.
Apply Time, Frequency and Amplitude Discrimination
Use the signal’s characteristics as a fingerprint. Compare the time-of-arrival between sensors to estimate the source location, examine the frequency content against the expected spectrum for the asset and sensor, and check whether the amplitude scales consistently with distance or test conditions. Each dimension adds evidence; a signal that matches the expected discharge signature in several dimensions is more credible than one that matches in only one.
Document the discrimination applied in the report: the sensor positions, the frequency bands examined, the amplitude values and the phase windows. The record is what allows a reviewer to test the conclusion, and it prevents a borderline signal from being re-litigated without data.
Time-of-arrival analysis between two synchronised sensors can estimate the source position along a cable or inside a GIS compartment, which turns a classification question into a location question. Frequency discrimination uses the band where the sensor and the discharge mechanism are most sensitive; gating excludes a known periodic source such as a communication burst. Each technique adds a line of evidence that the final report should state explicitly.
Perform Background and De-Energised Checks Where Appropriate
Measure the background before and after the asset measurement at the same sensor positions, with the asset in service and, where the procedure allows, with nearby equipment states documented. If the signal appears in the background with the same character, it is not coming from the asset under test. A de-energised check, where permitted, is the decisive test: a signal that disappears when the asset is de-energised is associated with the asset or its immediate circuit, while a signal that remains is environmental.
De-energised checks require the same safety controls as any isolation and grounding procedure, and the interpretation must account for the change in operating stress: some discharges appear only at operating voltage, so the absence of a signal when de-energised is expected and does not by itself confirm the asset is healthy.
Procedural controls also reduce ambiguity: schedule repeat measurements at the same load and weather conditions, record the state of adjacent equipment, and avoid measuring during known switching activity unless that is the specific question. These controls may not be possible on every site, but the ones that are possible should be applied consistently and recorded.
Document Confidence and Escalation Limits
Every PD conclusion should carry a confidence level based on the evidence dimensions confirmed, and a defined escalation rule. State which checks were performed, what the pattern indicated, and what confidence the combination supports: screening-level, confirmation-level or diagnosis-level. Define the conditions that trigger a repeat measurement, an offline confirmation or engineering review.
| Evidence dimension | Discharge-like behaviour | Noise-like behaviour |
|---|---|---|
| Phase relationship | Stable, repetitive phase windows | Random across the cycle |
| Repetition over time | Present at each repeat measurement | Appears and disappears with load or weather |
| Multi-sensor comparison | Localised to the suspected source | Present identically on all sensors |
| Background check | Absent from background at same position | Present in background without the asset |
| De-energised check | Disappears when de-energised | Remains after de-energisation |
The table is a classification aid, not a proof. A single dimension can be misleading, and the conclusion should be based on the weight of independent evidence, with the remaining uncertainty stated in the record.
The confidence statement should distinguish screening evidence from confirmation evidence: a signal that repeats at a stable phase window, localises with a multi-sensor comparison and survives a background check supports a confirmation-level conclusion, while a signal that meets only one or two criteria supports a repeat measurement and closer monitoring. The escalation rule should be written so that no single dimension, however alarming, triggers a decision that the evidence cannot support.
Close the record with the action and its owner: the repeat date, the responsible engineer, the follow-up test to be scheduled and the condition that would trigger escalation. A classification record without an action plan leaves the confidence statement without consequence, which is how borderline findings are lost between maintenance intervals.
Frequently Asked Questions
What is a background test?
A background test measures the sensor response with the same setup but without the asset contributing, or with the surrounding conditions documented, so the operator can identify which parts of the reading are environmental. It is the reference against which the asset measurement is compared.
Why is a multi-sensor comparison important?
Because a local discharge attenuates with distance and appears differently on different sensors, while external noise tends to couple identically across the measurement points. Comparing synchronised sensors separates the two more reliably than any single reading.
How much confidence is enough to escalate?
Confidence should match the action: screening-level evidence justifies a repeat or a controlled follow-up, while escalation to engineering review is justified when the pattern is confirmed by multiple independent dimensions and suggests a developing defect. Define the thresholds in your procedure and record the evidence that reached them.
For the framework that selects PD methods by asset and outage condition, see the partial discharge testing guide. When you need instruments that support multi-sensor noise discrimination, review PD test equipment options and request a technical proposal with your sensor and access requirements.
The discipline that separates discharge from noise is reproducible evidence, and the instruments and procedures you choose should be judged by how well they preserve it.