After going through various suggestions, and whatever I could try under situatiuation, I think there are 2 parts to be addressed:

1. Problem itself appears to be noise generation. Clean electrical environment. To keep it clean, RC or zerocrossing driving of inductive load, etc. If problem can be killed at source, it does not spread. But all these are always not under your control and you are not been able to verify level at which it is reduced. So the second part.

2. Even if problem enters your system, try fighting agains it. Like try to over rate powersupply, better decoupling and reset circuits and recovery from hang using watchdog.

Some how I feel, watchdog may come in too late after damage is done, recovery may not be always trust worthy for running parameters, or variables like averaging stack. But when best security is needed, everything has to be done, as we can not see killer with limited resources and time.

Vimal Amin