Building a mains power quality logger for unstable grid supply

Equipment at two sites was failing at a rate the supplier could not account for. The suspicion was supply quality, but nobody had measurements — only the observation that failures clustered after outages. This build was to get the measurements.

The first approach was unsafe

The obvious cheap design is a resistive divider straight into an ADC input. It is documented widely and it is the wrong answer here: it bonds the logic ground of the device to mains neutral. That is tolerable where neutral is reliably at earth potential. At these sites it was not, which is part of why we were measuring in the first place.

Splitting the measurement

Magnitude and timing turned out to have different requirements. Magnitude needs amplitude accuracy but tolerates latency; frequency and dropout detection need edge timing but almost no amplitude resolution. Measuring them with one front end meant paying for both properties in a single expensive part.

Magnitude path

Split-core current transformer into a burden resistor and a 12-bit ADC. Isolated by construction, no galvanic path to the logic side.

Timing path

Optocoupled zero-cross detector into a hardware capture input. Gives cycle-accurate frequency and unambiguous dropout timing.

The 47 sub-200-millisecond dropouts were the finding that mattered. They are too short to register as an outage to anyone on site, and long enough to reset equipment mid-write. The failure pattern stopped being mysterious.

Open questions

• CT calibration drift over temperature is being tracked monthly and is not yet characterised.
• Three units is not a sample. These figures describe two sites, not a grid.
• The logger records; it does not yet correlate against equipment failure timestamps automatically.