Online vs offline partial discharge testing: which does your switchgear need?

6 min read · Updated 30 July 2026

Partial discharge — the small, repetitive sparking inside or across degrading insulation — is the early warning that matters most on MV switchgear, because it is how insulation announces itself before a flashover. Two families of test listen for it, and the industry's naming does buyers no favours: "online" and "offline" sound like the same product at different convenience levels. They are different instruments answering different questions.

Online: the switchroom as it actually runs.

Online (in-service) PD testing reads the equipment live, closed, and carrying its normal load — TEV sensors on the enclosure cladding for discharge inside the insulation, airborne ultrasound at openings for surface tracking and corona. Its strengths follow from the condition it works in: the whole switchroom can be screened in a visit, nothing is opened or isolated, and the insulation is being judged under the genuine stresses of service — operating voltage, load heating, real humidity and contamination. Its honest limits: readings are comparative (panel against panel, site against background), location is approximate, and the output is a severity-rated indication, not a laboratory measurement. It answers: which of my panels deserves attention, and how urgently?

Offline: precision, at the price of an outage.

Offline PD testing takes the asset out of service: isolated, de-energised, then energised from a test set under controlled, often stepped voltage while calibrated instruments measure discharge directly. Done well, it quantifies activity, characterises it, and locates it — genuinely diagnostic information that a screen cannot produce. The costs are equally real: an outage per asset, isolation and switching work, specialist equipment and time — and one subtle limitation that gets little airtime: the test condition is not the operating condition. Equipment tested cold and unloaded is not experiencing the thermal and load stresses of service, so the two methods can each see what the other misses. It answers: what exactly is wrong with this asset, and how bad is it?

The order is the strategy.

Put the two questions side by side and the sequence writes itself. The screen is cheap per panel, live, and comprehensive — so it goes first, across everything, without touching production. The diagnosis is expensive per asset and needs an outage — so it goes second, aimed only at what the screen flagged, ideally alongside the thermal picture of the same boards. Run in that order, the outage window arrives with a short, justified target list. Run in reverse — or run offline everywhere "to be thorough" — and you pay diagnostic prices to learn that most panels were fine, which the screen would have said for a fraction of the cost.

Where we stand, declared.

We deliver the online screen, as one of the two methods inside the Electrical Health Assessment — and when a finding justifies offline diagnosis, the honest next step is to say so and scope it, not to stretch a screening instrument into claims it cannot carry. The boundary runs in both directions: a clean screen is not a laboratory certificate, and a screen finding is not yet a repair bill. Each step stands on its own evidence — which is, in the end, the only pricing logic that survives being questioned.