If you've ever wondered why two small-current grounding fault locators look identical on the outside but differ by a factor of two in price, the answer is usually hidden in the analog front end. The CTs, the analog-to-digital converter, the clock that synchronizes everything-that's where the real cost and the real performance live. In a system with an ungrounded or resonant-grounded neutral, the fault current on the faulty feeder is remarkably small. A few amps under normal conditions, and when the fault resistance is high, it can drop to tens of milliamps. At that level, the device isn't just measuring a signal; it's trying to pull a whisper out of a noisy room. And the quality of that pull depends entirely on sampling accuracy.
The first thing to understand is what you're actually trying to capture. When a single-phase-to-ground fault happens, the transient zero-sequence current is the most useful fingerprint. It rises sharply, oscillates for a few milliseconds, and then settles into a steady state that's much smaller. The oscillation frequency is typically several hundred hertz to a few kilohertz. This transient burst is where the faulty feeder separates itself from the healthy ones. If your sampling rate is too low, you don't see the peak. You don't see the ringing. You get a smoothed-over version that looks a lot like what the healthy feeders see. In a 10 kV network with five feeders, for instance, when the grounding resistance reaches 1 kΩ, you need a sampling rate of at least 6 kHz to keep the transient recording error under 10%. Drop below that, and the waveform you're analyzing is no longer the waveform that actually occurred.
Amplitude accuracy is the next hurdle. Most zero-sequence current transformers are designed to be accurate at rated current, which might be 5 A or more. But at 0.5 A, many of them exhibit ratio errors of 20% to 80% and phase errors of 10° to 50°. That's not a minor deviation. If your selection method compares the amplitude of zero-sequence currents among feeders, a 20% error can easily flip the ranking. The feeder that should be flagged as faulty suddenly looks healthy, and a healthy feeder takes the blame. Even if the CT is perfect, the ADC introduces its own problems. A 12-bit converter with poorly designed input scaling can lose the low-amplitude details. Quantization noise can bury the very signal you need when the fault resistance is high.


Phase accuracy is just as important, and it's often overlooked. Many algorithms rely on the fact that the faulty feeder's zero-sequence current is roughly opposite in phase to the healthy feeders'. Theoretically, that's a 180° difference. But if your CT introduces a 15° phase shift, and your ADC clock adds another 5°, you've compressed that distinction. Now a healthy feeder and a faulty feeder might only differ by 140° or 150°, which is much harder to discriminate, especially when you have multiple feeders with similar lengths and load profiles. The problem gets worse in digital substations, where zero-sequence current is synthesized from three separate phase currents. Any mismatch in the accuracy or synchronization of those three channels propagates directly into the zero-sequence value. We've seen cases where a 2° phase mismatch between phases created a 15% error in the synthesized zero-sequence current-enough to confuse the selection logic.
Then there's sampling rate and synchronization. For transient-based methods, the Nyquist criterion is just the starting point. You want enough points per cycle to capture the shape of the transient, not just its existence. If the dominant transient frequency is 1.5 kHz, a 4 kHz sampling rate gives you fewer than three points per cycle. That's dangerously close to the edge, and any anti-aliasing filter with a gentle roll-off will attenuate the very frequencies that carry the fault signature. Higher sampling rates help, but they also demand more processing power and generate more data. In a multi-feeder system, synchronization between channels becomes critical. A timing skew of just a few microseconds between channels translates into a phase error. That phase error can mimic a real fault characteristic, triggering a false selection or missing a genuine fault.
So what does this mean when you're choosing a device? It means you can't treat sampling accuracy as a secondary specification. It's not something to trade away for a lower price or a smaller enclosure. The algorithm-whether it's wavelet-based, correlation-based, or something else-can only work with the data it receives. If the data is distorted, no amount of clever math will recover the truth. When you evaluate fault locators for export markets, ask about the CT accuracy class at low currents, not just at rated current. Ask about the ADC resolution and the actual sampling rate per channel. Ask about channel-to-channel synchronization. A device with 0.5% current accuracy and 10 kHz sampling will outperform a cheaper unit with 2% accuracy and 4 kHz sampling in virtually every real-world scenario, particularly in systems with high-resistance grounding or long cable runs where the fault signal is weakest.
In the end, sampling accuracy is the foundation. Get it right, and the selection works reliably. Get it wrong, and you're installing a device that looks good on paper but fails when it matters most.