What is the practical difference between an offline and an online partial discharge test on a motor, and which one should I specify?
They answer two different questions, and a mature test programme uses both. Offline testing quantifies; online testing trends.
Offline (IEC 60034-27-3): The winding is isolated from the drive and energised from a separate low-noise HV source. Each phase is tested with the other two grounded, and voltage is raised in steps (typically 0.2 U0 increments up to 1.0 U0, then held). Because the background noise can be pushed below 10 pC, results are calibrated in picocoulombs and are directly comparable between machines, between phases and between years. This is the correct method for a commissioning baseline, for PDIV / RPDIV determination and for acceptance after a rewind.
Online (IEC 60034-27-4): Sensors stay installed and the motor is measured under real thermal, mechanical and voltage stress - including the conditions that actually initiate discharge. The trade-off is that readings are uncalibrated (mV or dB), noise-rich and influenced by load, so online data is only meaningful against its own history on the same machine.
Recommended split: Offline for baseline, diagnosis and post-repair acceptance; online for continuous trending between outages. Never compare an online mV reading with a laboratory pC value - the two scales are not interchangeable.