A ~10kV spark discharge is a pretty intense level. When from an ESD generator, it's, well, that's ESD, at about the highest level normally seen (8kV contact / 15kV air discharge).
You don't mention what pattern the Hipot produces (single or repetitive, what rate / waveform), but even simply that more metal and wiring is involved in the test, makes the EMP that much more intense than commercial-level ESD.
It's no surprise then, that commercial-level interfaces (serial, USB, anything in the PC itself for that matter) can potentially fail in this scenario.
Now, EMC topics are complicated. Much must go into consideration to reach a determination of what's going on, and how to address it. A single wire unaccounted for can spoil all the filtering and shielding effort in the world. Sneak paths can carry EMI circuitously through a system; distance helps, but isn't the end-all where connections are also required.
So, understand that, within these margins, I cannot adequately convey:
- The analysis method to assess a given system;
- A comprehensive solution that is guaranteed to address your problem at any cost; or
- an optimal solution that addresses the problem adequately at minimal cost / number of changes.
For a comprehensive solution, you would really need to bring an EMC expert on site; they would be able to inspect your wiring and determine what grounding, filtering, shielding and isolating solutions would be most suitable.
That said, I can still offer some hints.
- The spark is the source, and when I say "EMP", it really is. The rise time of an air spark can be a fraction of a nanosecond. In that very instant, wave energy propagates at the speed of light, away from the spark, as electric field is discharged through the now-conductive path. In a nanosecond, the wavefront travels 30cm or so (or a bit less when conducted within cables). That is, from one nanosecond to the next, the wavefront -- propagating this change in voltage from 10kV (at the wire; less at a distance, of course) to zero within about the span of your hand. So, that's a lot of field intensity, it's a lot of rate-of-change, and it induces currents into everything nearby.
- Where that EMP creates voltage drops across cables, particularly at connectors where the cable shield joins to a metal enclosure (hopefully), some voltage can drop across that gap, and disrupt the low-level signals within. If that drop is some 10s to 100s of V, further malfunction can easily result (e.g. triggering CMOS latchup of interface ICs).
- Most communication standards are made to reject impulsive noise (error detection or correction), and data may be retried automatically, but persistent interference may knock out those attempts too, and the link drops.
- So, depending on the spark pattern, it might be okay to just limit it to single pulses, or it may be that even a single pulse is enough to knock out your system.
- The EMP expands outward from the spark location, carried up any connecting wires, and eventually bouncing around on them as the wave hits a reflection (open or short circuit). This reverberation generally stretches the transient, adding ringing (so it's not just a one-sided impulse, but both polarities in rapid succession). Cables have a lower impedance, so that 10kV discharge might be on 100 or 50 ohms, i.e. a peak current through the spark of 50-200A, for as long as the cables are (some meters <--> 10s of ns, plus ringing). These are big pulses!
- Electric and magnetic induction also ensures EMP on nearby (unconnected) wires/cables, through gaps in panels, etc. (An effective shield must have slots of maximum length much less than the risetime of the pulse: a slot length of a few inches is enough to let a direct ESD strike affect low-level circuitry behind that it.)
(It's not clear to me if you've also grounded the Hipot unit to the same local ground terminal, so that there is a spark-discharge path closed around the unit, or if there's no such connection and the loop area in fact includes facility ground, and probably mains wiring as well. A physical diagram would be helpful here.)
What can you do about it?
- As much as possible, confine energy to the local area. Make short loops. Instead of laying out as much HV wire as available, bunch it up (fold the wire over on itself) to keep distances short. Ground the Hipot machine to the DUT, and then earth both. Or earthing isn't really important (give or take safety practices in the Hipot manual, applicable local and work rules, of course). Do not make circuitous discharge paths, like grounding Hipot to one earth (e.g. the one electrode) and DUT to another (facility or mains earth).
- Add filtering and decoupling to the affected (high voltage) leads, so that energy isn't carried up them, and the energy stored on their length doesn't become radiated to surroundings. This is not exactly easy to do at high voltages (or recommended: a filter necessarily stores energy, too, and you'd need pretty big capacitors -- 10kV and some nF, say -- to do much of a job here; the filter acts to increase overall spark energy, just delayed a bit in time; this may further damage DUTs that would otherwise be repairable).
- Combined with filtering, shield the local area. Put the DUT and Hipot source (or at least its HV lead), within a metal enclosure. Decouple all wires leading through that enclosure. This terminates EM fields to the enclosure, preventing radiation, and limiting scope to only that which gets conducted out along cables. Filtering, and terminating cable shields to enclosure, addresses conducted routes.
- Where grounding and filtering isn't feasible -- or still isn't enough -- consider isolation instead. Use separate sources -- different branches of mains for example (the Hipot being on 220V vs. PC's 120V seems likely it has this already, but you'd have to see the facility wiring to know if they're actually routed separately). Use a low-capacitance isolation transformer to power one or both. Consider optoisolators for serial/USB -- but, mind that these too can have capacitance across them, making them unsuitable for the high-frequency energy in question here. Best case: get a fiber optic isolated unit. Fiber has very little capacitance along it.
This can reduce the amount of current propagating up the PC's cables (and onto its chassis, and in turn onto other connected cables), giving it a better chance of survival.
Changing the PC to something smaller (less capacitance, less current flows onto the chassis in general) and less connected (preferably, not connected at all) is also an option: you might consider replacing it with a laptop, notebook, tablet, etc.
If you're using USB hubs, mind that they need to be equally robust. There may be industrial types available that are rated for higher immunity levels -- but beware that EMC ratings are only in one test setup, not yours specifically, and there's always the possibility that they've self-reported those claims without testing them, or testing under realistic conditions.
As you can see, everything depends wildly on what is connected, how and where, and seemingly subtle or irrelevant details (like which earth connection is used where) make vast differences to the outcome. An EMC specialist is likely a heavy-weight solution, but is the simplest (administratively speaking) option with the highest chance of success.
