RE: 48kW DRSSTC and RELIABILITY

["Tesla list" <tesla@xxxxxxxxxx>]

Original poster: Steve Conner <steve@xxxxxxxxxxxx>


we are finding is
exceptionally difficult to get a unit to have enough performane vs. cost
(bang for the buck), yet at the same time be reliable enough to sell as
a commercial unit without it becoming a warranty nightmare.

OK, now the philosophical question: How do you know the coil won't be 
reliable enough? Did you build 100 identical coils and run them for 
100,000 hours and take note of when each one failed? I'll bet you 
didn't. You probably took the manufacturer's MTBF figures (or the 
ones that the military publish) and combined them using the mil spec 
equations.

But if you're running your IGBTs over current, the manufacturer's 
implicit warranty that the device will last for a reasonable time is 
void, and so presumably is the mil spec reliability calculation. You 
may not even know you are running them out of spec under certain 
streamer loading conditions, unless you use active current limiting. 
(even then, if the active current limiting is unreliable, you still 
may not know.)

My approach to SSTC reliability was like Terry said: to brainstorm 
everything that could possibly go wrong, decide what out of those 
possibilities were actually likely to happen- as opposed to 
theoretical flights of fancy- and then add protection circuits against them.

For me, overrunning the IGBTs turned up as a major weak point, for 
the reason I explained above. So I fitted a current limiter and set 
it to the peak current rating of the IGBTs I was using. I bench 
tested the current limit extensively until I was confident it worked 
and would always kick in when needed.

To keep the impressive spark performance we all expect from a DRSSTC 
now, I just used one size bigger IGBTs than usual. The output o f the 
coil ended up limited by tracking down the inside of the secondary 
former, rather than the current limit.

The other major failure route I identified was bad drive. This 
happens when a ground arc makes the resonator jump to a funny mode, 
in combination with a naive feedback circuit that doesn't know it 
ought to reject the resulting crazy frequencies. You end up toggling 
the IGBTs on and off way too fast, probably hard switching high 
currents too, and they explode from massive switching losses.

Steve Ward fixed this with his "HF Protection" circuit as far as I 
know: I chose to use the PLL approach. This is my only quibble with 
Terry's DF-DRSSTC by the way: its feedback circuit has no way of 
rejecting bad feedback and ensuring clean switching.

Finally, there was all the boring stuff like UVLO and soft-start and 
shutdown to ensure the IGBTs never get bad drive signals: they either 
get good drive or nothing. These techniques are common place in 
industrial SMPS and motor drives. There was also EMC and shielding, 
and I put a lot of thought and experimental work into laying out the 
PCB and wiring harness so that RF from ground arcs and flashovers 
wouldn't go where it wasn't wanted.

So I now have a driver with no flaws "That I know of". The next step 
is to find the ones I don't know about. For this I'm trying to use 
the open source development model. I released the circuit and PCB 
artwork in the public domain and encourage everyone else to play with 
it and help me find the bugs ;-) So far, the complexity has scared 
most people off, but I believe that every part is in there for a 
sound reason and have no plans to try simplifying it.

Lastly, I should mention that I owe one to Steve Ward and Jimmy 
Hynes. They were kind enough to share their DRSSTC experimental 
results and ideas with me, so my design could build on all their 
findings too. This was the main reason I chose to make it an ope n design.

Steve Conner
http://www.scopeboy.com/