H713, I'll bet it's oscillating. Do you have an oscilloscope to see what's happening? If not, put an AC voltmeter on the output to see if there are any spurious signals.
The NE5534 opamp is not guaranteed to work correctly if the input voltages get too close to the power supply rails. On the data sheet this is the Common-mode Input Voltage Range. In the latest design, the reference voltage applied to the inverting input will certainly go out of range. At 200 V out, the ratio of R13 to R14 will put the non-inverting input right on the edge as well. If you don't want to use a rail to rail opamp, you could make the NE5534 work by providing a negative power supply a couple of volts below GND.
This could be done simply by using four rectifier diodes in series with the 24 V supply circuit GND connection and the general GND connection, ie., rename the two GND nets in the 24 V supply as -2.4V and connect the cathode of the rectifier stack to it. U1 pin 4 would connect to -2.4V. U1 pin 4 should have a 0.1 uF to GND for AC bypassing.
I think R7 to R9 are too low to help share the load current much.
Yep, it was oscillating all right. Looking at the output on the scope, it probably a few hundred kilohertz. I should have guessed this by how twitchy the pot was- just
touching the pot shaft caused the output to drift.
I added a 470 pF from the output of the op amp (pin 6) to the inverting input (pin 2), and that killed the oscillation- mostly.
This thing does
not like capacitive loads. With a .082uF on the output, it draws almost 50 mA and the capacitor itself hums at about 3 kHz. As it turns out, poly film caps can become surprisingly effective transducers!
I replaced the 470pF cap with a 100nF cap, and it considerably improved the stability, but there is still a bit of oscillation into capacitive loads. As a side note, I can put a big electrolytic cap across the output and it doesn't care- it's the smaller film caps with very low ESR that make it go ballistic.
I tried switching the 5534 out for an OPA604 and then a TL071, and none of them were any better. I tried adding a negative rail, and it didn't seem to help much either.
If I can get it stable into capacitive loads, then I think I've got a solid design here. Load regulation is
excellent, at least for a simple design like this. Connecting and disconnecting a 250 mA load (at 250V) causes the output voltage to vary no more than .1V. Quite honestly, I had a hard time really measuring much as the most precise meter I have is 4.5 digits, and even that isn't really enough to see what's going on here. I'll call that passable. It is still worlds better than the old Eico 1030. I took apart and cleaned all the pots in the Eico, and replaced the two main filter caps (most of the smaller electrolytics are original and should still be replaced). It is now a fairly usable power supply, but its output varies as much as +/- 2V with a 150mA load.
So what about the FETs? The proof-of-concept has eased my concerns considerably. I'm testing with an IRFP450, simply because I have them in stock. I ruled out the IRFP450 from consideration early on due to its 500V limitation and the fact that it doesn't have a DC SOA in the datasheet, however even with some pretty awful things done during testing (intentionally trying to kill it), I was unable to do so. With transformer tap switching, I think the FETs will be running quite conservatively. As mentioned I will probably use the FQA8N90C-F109, though I may try a Fuji 2SK3675 as well. There is no SOA in the datasheet (

), but if I do some worst-case-scenario testing and they hold up I would feel comfortable using them. Regardless of what FET I use, this testing will be necessary.