Crowbars

Remove the '.':

If you design the current dump carefully, it is very unlikely to fail. The stuff I put into production only failed if the semiconductor manufacturer sold us parts that didn't meet their specs. The only example I can remember was a single Hewlett-Parkard opto-isolator that was much slower than it should have been - when the service engineer found out which part was creating the problem he just replaced it. It was very unexpected and the manager got one of the other engineers to check my worst case propagation numbers, which were fine. Many years earlier I'd run into a TI digital latch which was fine at room temperature but got too slow when you closed the cabinet and the ambient got warmer. We found that with a hot air gun and a part swap cured the problem - but that was on a development machine and management didn't get into the act.

Adding redundancy is rarely cost-effective or sensible.,

So 37.5 to 62.5 W which is quite a bit. Turning off the charger would be cheaper and easier.

But dissipating that heat costs money. Heat sinks aren't free and neither is the ventilated space they need to do their job.

Which neither detailed nor helpful.

And you haven't told us anything like enough. Dumb newbies never do.

3A to 5A MAX when the batteries are empty. Less than 1A at 80%. mA's at 90%.

They are warm to be touched. No heat sink necessary.

The motor draws around 20A to 30A peak out of 24KWhr battery. So, just a couple of A's from each portable. Perhaps a bit more when fast charging an empty one. But so far, the 3A fuses are holding up. The rotary switch is capable of 3A and the wires are at least 5A.

Dumb readers don't read everything either.

PPS: I guess I still need position sensor to make sure it stop at position #1 for 168V and 192V. No energy need to keep in position. Namely, bi-stable or tr-stable relays.

Actually if memory serves me right it's the negative resistance aspect of the diac which makes it ideal for driving a triac. Hopefully someone with experience of these beasts will adjudicate the matter and provide some clarification.

35V is the absolute maximum according to the datasheet so with a breakover of 31 or 32V I'd feel justiified in assuming there's enough headroom there to kill the supply rail before it gets anywhere close to damaging the RF trannies.

Is this an RF amp?

The RF guys are weird. An RF transistor might peak at 2x the supply voltage if it's driving a tank. Sometimes the data sheets specify the actual abs max, and sometimes they actually specify the max supply voltage and assume the 2x effect. They never say.

I test them to see where they actually start to draw current.

Well, a zener clamp is fairly easy, but you might want also to slew-limit the voltage rise on the input; a PNP pass transistor with a little emitter-base capacitance will filter out fast-rise edges, give a hard clamp a millisecond to operate. With a few microseconds leeway, TL431 can implement a clamp fairly effectively.

Well now you've made an assumption there which is false. The PSU was badly designed in the first place and is notorious for failing in this particular piece of communications gear. In fact, rather than attempt to repair the unit and replace the underspec components with heavier duty ones, many repairers rip it out altogether and fit a low-noise switcher in its stead. The orginal PSU is actually very inefficient anyway and positively belches out heat when worked hard. Up-rating the cooling fans is another task faced by those who choose to repair over replacing the entire unit with a switcher. I haven't yet decided which way to jump with this decision and just wanted a quick and dirty but effective way to shut the thing off it it failed again, until such time as I make my mind up.

It's the RF final output amplifier which kills the supply due to the supply having a number of design defects. The item itself is one of these:

Yes, they can be confusing with all that reactance to take into account.

This is where a decent VNA and a curve tracer prove invaluable. In fact a curve tracer is one of the few pieces of test equipment I don't possess. I really must get around to ordering one.

The RF folks should give up all that silly s-parameter and load pull nonsense and furnish Spice models for their parts, and publish sensible DC curves.

They usually say "adjust the gate voltage until it works" or something equally scientific.

I use RF parts in time domain and have to measure things. Like capacitances, breakdown voltages, Rds-on, pulse behavior.