How to track down <100µA earth leakage on DC rails

Question

On our subsea robot, we have an isolated power supply (converting 3-phase AC into 320 Vdc).

For security reasons, we have a line monitor monitoring the resistance between protective earth and both 0 V and 320 Vdc voltage rails. Some equipment have their carcass connected to protective earth. In normal operations, the resistance between earth and 0 V or earth and 320 Vdc is supposed to be >20 MΩ (the maximum value the line monitor can measure).

From time to time, we have some small leakage (resistance in the 100 kΩ - 19 MΩ range).

For now, to find the fault, we just plus/unplug things to try to find out where the fault is. Sometimes it works well, but sometimes it takes a lot of time, especially when it is in some unexpected place (at those levels of sensitivity, some materials that look like isolators are just big resistors, dust and humidity starts to matter, ...).

The way the line monitor works is by injecting generating a +-50 Vdc voltage between earth and either 0 Vdc or 320 Vdc for 2 seconds, and measuring the resulting current (20 MΩ = 2.5 µA).

My question is: is there any efficient way to track down where the fault is? (The fault might be in the wiring, on a PCB (custom design or of the shelf), or in a black box device (like a motor)).

Also note that in most cases, the fault is also present when the power supply is turned off, so methods working only when only the line monitor is active are also welcome (most diagnostics are done this way, as it avoids working with 320 Vdc present nearby).

Test equipment available: scope, isolated scope (battery powered), multi meters (including one with µA range), lab power supplies (up to >500 V), mega-ohm-meter (injecting 250 Vdc or 500 Vdc, measuring resistance up to 10 GΩ), function generator, programmable load, current clamp down to 10 mA dc. If you have a solution requiring some other equipment, feel free to suggest, I might be able to convince the CTO to buy it.

Solutions working only for the bigger leakages (resistance <2 MΩ) are also of interest.

Answer 1

is there any efficient way to track down where the fault is?

It will be hard if you can't isolate or turn off sections of power. Think of a circuit, if you have several loads in parallel you can't know where the majority of the current is going without turning off loads. If you can turn off and isolate the loads then it is possible

In a grounding network it will be more difficult to find the resistance fault because it's likely that things will be hard to separate. You have a few choices here to determine where the current is coming from but it will be hard to deduce where the insulation break is because you can make conductivity measurements at different points and then try and find the point of least resistance.

The third idea is if there is supposed to be a specific resistance and there is isolation, you could turn everything off and apply a known voltage (with a bench supply) between the two points and then see if you can find where the current is going, this will only work if you can do this without destorying electronics. Ground should be fine to drive a potential between chassis and wherever you want to measure the leak, because everything will float.

It also depends on if the leak is AC or DC, if it's AC it will be harder to locate because it could be something capacitvely coupled that is driving the leak, where you could have to bits of metal with insulator between them and still get current flowing. You could also have bulk capacitance, if this is your problem then you'll need shielding. If you suspect AC leakage, another thing you could do is try and shield different areas of insulation and see if that helps.

Answer 2

I'd use an audio frequency network analyzer (ie, a PC soundcard and some python) plus an amp and a current transformer clamp.

Low DC currents are hard to measure without breaking the circuit and inserting a current sensing resistor, and the goal is to avoid unplugging everything. On the other hand, low AC currents are easy to detect with a current transformer clamp.

However, with an AC excitation voltage, all the parasitic capacitances between everything and the chassis, plus the noise filter caps, will create AC leakage currents which will add up to the leakage current you want to observe.

Thus:

I'd use the soundcard to output a pure sine wave, and do a frequency sweep, say 20Hz to a few kHz. Amplify it to +/-50V with an off the shelf audio amp, and feed that signal as common mode voltage relative to Earth on your 320VDC supply. The supply can be on or off, it makes no difference, as long as the voltage is common mode.

Then you can use the current transformer, with a low noise preamp, to measure AC current through any cable in the robot.

The trick is to use the soundcard to synchronize the sampling of this current transformer signal with the generated sine wave. Any soundcard which has audio inputs and ouputs will do that. Then project on phase and quadrature, and you get the measured current as a vector or a complex number.

Now stray capacitance will create imaginary (phase shifted) leakage current, but the actual leakage you're looking for will create real (in phase) current.

Then if you plot real and imaginary values of measured current in real time, you can visualize leakage due to resistance and leakage due to capacitance separately. As an example, I stepped the leakage resistance on the plot.

Due to leakage capacitance, imaginary component will be much higher than real component, but still well within the dynamic range of any respectable 24 bit soundcard... I'd recommend a pro USB soundcard with microphone preamps, they're not expensive, and with stereo inputs you can even use two current transformers!

Then you can clamp the current transformer on any cable in the robot and see in which direction the current is going, until you find the culprit.

In fact you can probably use off the shelf audio analyzer software to do that, although it will probably plot phase and magnitude, which would be useless in your case.