Ryan Miller wrote:
>I have a board that I recently redesigned that contains nine
>T90 relays that switch 110VAC. The board goes in an
>industrial controller and switches lights and contactors
>(mag starters). The relays all have flyback diodes across
>the coils and snubbers across the contacts.
.......

Hi Ryan, for several years I consulted with a company that built
industrial controllers that switched multiple 220 VAC inductive loads.
They sold 1000s of these units each year, and saw every manner of
failure. Many previous comments by other picisters have indicated
that 1 or 2 things might fix this sort of problem, but in my
experience, we found it required multiple lines of attack and each
was only marginally effective. A heavy inductive switching environment
is probably the very worst place into which to put a microprocessor,
and I think the best results come from an overall systems-level
approach - which of course involves 20 or 30 or 60 aspects.

That being said, let me relate some of the most effective methods
in my experience:

1. EMI and spike filters at the hiVAC power input - series EMI filter,
   MOVs, etc.
2. EMI/spike filtering on the power supplies - "transient voltage
   suppressor" diodes [eg, Microsemi 1.5KE series] on the P/S input side,
   downstream of the AC transformer, if there is one.
3. Bypass caps, of course, at all the usual points.
4. Of course, diodes across the relay coils and snubbers across relay
   contacts.
   NOTE - some loads produce significantly worse switching transients
   than others, so the same snubber may not be universally effective.
   When looking across the contacts, you will see hi-speed spikes [bad]
   riding on a slower wave [normal] in cases where the snubber is not
   effective enough. You might want to use an isolation xfrmr to do
   this measurement.
   NOTE - one thing you can do is sniff around with a scope probe,
   ungrounded, to measure presence of EMI, and determine what it is
   correlated to.
5. On the low-level signal lines, RC filters and transorbs [transient
   voltage suppressor diodes, which are essentially low-inductance
   zeners] are effective. Note - we ended up putting transorbs on
   practically ever signal line - and regular zeners are too slow here.
6. Layout issues - physically separate hi/low V, and signal/switching
   circuitry. Separate power busses and gnd planes [if possible]. Don't
   run different lines near each other, and don't cross them.
7. Use watch dog timer in uC s.w. Use debounce routines on signal
   measurements.
8. Use digital bus crowbar ckt in cases where chip latchup is an
   especially bad problem. [Note - the watchdog will not fix latchup].

There are probably a few others which I forget.
==============

>
>The whole point of this long-winded explanation is to ask if
>filling the "no trace zone" with a ground plane would do any
>good. If so, should it be earth ground or circuit ground? Is
>there any way to hack a current board (adding foil or
>something) to simulate the ground plane that would give
>meaningful test results before we redo the board? The board
>is two-sided, through hole.
>

Regarding gnds, the digital, analog, switching, and switched
should be as physically and electrically separate as possible,
and should be connected at "one point" as close to the power
entry point of the board as possible. The more current, the larger
the trace/wire.

Regarding hacking the test bd, try using aluminum duct tape
grounded back to the hiV side [looks like relay power in this case].
In essence, you are putting a guard ring around the hiV side which
is intended to contain the internal fields. Likewise, you can
put a gnd guard ring around each other subsystem, connected to
the gnd of that subsystem.

Hope this helps,
- Dan Michaels
Oricom Technologies
http://www.sni.net/~oricom
==========================