We've got a 4x8 China router with ATC running (a confirmed legitimate copy of) Mach3, Delta servos and WiXHC MKX-IV Mach3 motion card. 

We cut a lot of adhesive backed foam- the first few tools come and do surface features, then I run a separate toolpath for a drag knife to cut through the adhesive and paper backing.  The drag knife toolpath has a modified post processor that I've removed all traces of M03 so the spindle doesn't grenade with an unbalanced tool.

It seemed pretty well sorted after a few teething pains (it was delivered with the Y axis reversed, so it was cutting mirror images of parts  ).  After a few weeks of running flawlessly, it started crashing during tool change occasionally (maybe once a week, using it daily).  Since then it has been doing it with increasing frequency, but still inconsistently.  Maybe once every couple days. 

Here's how it goes down-

I'll run Example1.txt file with T3, T4, and T5.  After that is finished, I'll Close G-Code, go into MDI tab, and input T7 M6 to change the tool to a empty toolholder (I don't keep the drag knife loaded in the tool holder to prevent the possibility of getting loaded and spindle turned on by accident).  I then manually change the empty toolholder to the drag knife, load and run ExampleDRAG1.txt.  I then turn off the vacuum table and dust collector, close the G-Code, change back to the empty toolholder, jog the head near the end of the work envelope, and change to a fresh sheet of material. 
I then load Example2.txt, and most of the time it runs without issue.  When it errors though, it goes to return T7 to its clip, and then suddenly jumps about a cm.  Sometimes X, sometimes Y, and just a few minutes ago it might have done X and Z (or the Z might have just from the impact with the tool holder bar.  It also cracked the 3/4" thick aluminum mounting plate for the spindle).

We have gone into Windows and turned off all the energy saving options (so the hard drive, USB, and the rest of the computer won't go to sleep after inactivity), but it's still doing it and it seems like the rate at which it's happening is increasing.  We haven't found any pinched/kinked/stretched wires or anything else visibly wrong, though we haven't pulled the sheet metal and fully chased everything. 

So my question is- what should we be looking at and have you ever seen anything similar?

And here's a couple pictures

Hi,
sounds like electrical noise getting into one or another of the servos.

When the fault occurs does it always occur at the same stage of the tool change cycle?. What I'm thinking is that say when
'procedure 4' happens, which is say 'rotate tool table', where a particular device, maybe a solenoid or motor operates that device
generates some electrical noise which can couple into an adjacent circuit, servo step/direction for instance.

The general rule for dealing with noise is identify the device causing it and suppressing the noise if possible and breaking any path
that would allow electrical noise to propagate to nearby circuits. Keeping noise sensitive circuit seperated from noisy motors is a good idea
but preventing noise from being radiated is still a better proposition than trying to stop that noise from causing problems elsewhere.

Spend a bit of time trying to identify the noisy part. Best not to change too many things, just one at a time and then test. Once you've
established that the change either has no effect on the fault or that it does, then and then only progress to the next item on your
list of suspects.

Craig