Transformer shot! (was scope SMPS/ capacitor venting)

OK, well I can easily establish that safely by popping V1812 out of circuit temporarily so sweeping the frequency adjustment pot won't have any effect.

Indeed. That seems to be the key point I have to observe.

Well I could.... But that's spot on resonance. I was under the impression that they're not supposed to run actually directly at resonance?

Yes, no problem. It seems the key regulated output is the 12.7V one and if that's correct, the rest should follow. There's a trimmer for 12.7V on the underside of the board.

Thanks again, Dimitrij. I'll report back tomorrow...

As long as you haven't been fooling with psu trimpots, you can forget about the resistor.

If you have been fooling with trim pots, then you'll have to follow the manual adjustment procedure first...assuming similarity to PM3262. (check pot rotation effects expected between models).

Forget about the resistor while you're doing this.

Continue to forget about this resistor until (and if ever) V1811 dies again.

RL

[...]

I am happy to assure you that I have not touched a single trimmer so far and neither will I be dismissing the heating of the power resistor, which is clearly indicating that something is still amiss here.

I've just noticed that towards the bottom of (true) page 106, it states the following:-

"The oscillator frequency is approximately 25kHz, determined by network C1811, R1823 and is adjustable by means of R1824."

It then goes on to specify the duty cycle. Two things stand out as requiring further investigation here. Clearly, the 25kHz clock frequency mentioned is *miles* away from what my clock is running at - and the frequency adjustment is made with R1827, not R1824 (which is fixed anyway). I'm guessing the reference to R1824 is just a typo, but can we say the same for 25kHz?? Since this *completely* changes our former assumptions, I'm going to confine myself to just replacing the flaky polyester caps for the time being. Be interested to hear how you think I should proceed now in the light of this...

And before anyone suggests it: I've frequency swept the primary circuit just in case there's a second resonance peak at around 25kHz and there isn't one.

Hi

News indeed. Something must be amiss, and quite heavily amiss, that's for sure.

It looks like the resonant circuit considerably out of tune.

Driving a 17 kHz LC with 25 kHz would not make sense to me, and it looks like the circuit does not like it too much either, since it overheats.

If it was indeed tuned for 25 kHz, then the currently set 22 (or 20) kHz pulse rate setting could at least make some sense. It would be slightly below resonance, but probably not too far away.

I was assuming that the resonant circuit was ok-ish and the frequency somewhat matched, but this turns out, now, not to be the case.

Now I've looked in the TDA1060 controller datasheet again, and checked the adjustment range (with the trimpot set from 0 to 10 kOhm and the 11 kOhm fixed resistor and 3.3 nF timing capacitor) and the calculated resulting range of frequencies happens to be from 17.3 kHz to 33 kHz.

With a 17.3 to 33 kHz range, that would put 25 kHz quite well in the middle. A 17 kHz resonance frequency does not fit anywhere though. It can't even reliably be adjusted (even if it were correct, which it surely isn't) because it's simply out of the trimpot's range.

But it would not make any sense to pulse a 17 kHz LC at 25 kHz either. The LC will be heavily capacitive, the power factor will be down in the ditch, and that will overload (-heat) the driver (just as it happens).

Therefore I can only think that 17 kHz is wrong. The LC is out of resonance. If it were OK, it should never have a 17 kHz SRF.

That leaves either a measurement error (now rather less likely) or a heavy stray capacitance somewhere, that brings the resonance down.

I can only think of C1806 and C1807. They are in series. If one of them has an isolation problem, that would leave the other one alone in the circuit - and therefore double the capacitance.

Also there are the resistors in parallel - R1817 and R1818. They should be 10 Megaohm. But if one of them is either shorted or improperly replaced (maybe it formed an isolation breakdown from over-voltage or someone put in a wrong value like zero Ohm instead), then that would also short out the corresponding capacitor, and have the same effect.

Usually resistors are reliable, but sometimes, some old ones of the "carbon composition" variety, do form a "hot channel" and break down.

Doubling the capacitance would almost halve the resonance frequency. Actually it won't *exactly* halve it, because there is still a third capacitor in parallel, namely the stray winding capacitance.

This looks quite enough to be realistic. If one resonance cap is shot, the LC frequency will go down by a great deal, and (assuming it was initially slightly higher than the pulse frequency), a change from some

25 or 27 kHz to the 17 kHz that you are now seeing, could happen.

It could also explain the overload on the resistor - a heavily capacitive out-of-resonance load would easily do that.

Can you get these two caps out altogether, connect a known good 15 nF (or two of 30 nF in series) instead and sweep again?

Also, to avoid unforeseen measurement errors, can you do that out of circuit? Just the transformer and the capacitor(s) on the primary. This would also nicely avoid the resistors too, they too may be questionable.

You won't need a high voltage cap for sweeping - any good 15 nF one will do. But don't power it up with "any 15 nF". To run at full power it needs something like a FKP1 or MKP-4C with proper ratings (see my earlier post, there are some suggested models that can work there).

Regards Dimitrij

You are refering to manual component numbers from the 3262 manual and schematic. The 3264 does not have the same schematic or part numbers.

The schematic is functionally similar, but uses a different control IC and different components are present/selected to set the IC's function. Part type for the main transformer/size and pinout, the size of resonant/snubbing components, along the actual supply power ratings may vary with model number, as well. One example is the different resonant cap size used.

3262 PSU adjustment begins with para 3.4.4 on 3262 manual page 115. (do not ignore preceding setup instructions re scope settings) You can assume the sequencing and intention of these instructions to mirror those needed for 3264, however numbers and test conditions that reflect operating power and typical frequency can be expected to vary. This includes chip reference pin voltages - TDA1060 internal reference is 3V62.

The converter runs at a fixed frequency, above resonance. The frequency only needs adjustment if the output voltages lose full power low-line voltage regulation. This is also the protective power limit.

The two model control circuits do not limit in a similar manner. The

3262 manual describes a latching power limit that occurs after repeated continuous overload. This does not appear to be present in in 3264, as it is chip-based feature.

RL

P.S. There may be another failure mode that I did not consider right away, that can make it a little harder for you to test the caps.

These high voltage impulse-rated capacitors are usually made with three metal layers and two layers of isolation internally. Mains "X1" and "X2" rated capacitors are also made in this way. They are like two capacitors that are connected in series inside. When one isolator breaks down, the whole capacitor won't be destroyed catastrophically because there's still the other internal half in series.

But it can happen (depending on the construction of the capacitor, if it has a continuous metal layer between the isolators) that the capacitance will "double itself" instead.

If one of your 30 nF caps has failed in this way, it will "become 60 nF" instead of becoming short-circuit. So you won't see it on an Ohmmeter.

But that would also be reason enough to detune the resonance a lot.

So, my advice would be, do not trust them, and do not trust their parallel resistors too much either. Take a known working 15 nF cap and sweep the transformer with it. If you get anything significantly above

17 kHz with a new cap, look for the main problem in this direction.

Dimitrij

P.P.S. If you decide to sweep the LC part in circuit instead of out of circuit (because the transformer is difficult to solder out etc...), you can use a small voltage source (a 9 V block battery) connected to the power supply's "AC" input. This will precharge the circuit enough to get the parasitics down. It won't start the power supply controller, but it will reverse-bias various diodes and also the base-collector junction of the main switching transistor. That will make these semiconductor parts non-conducting (at small signal levels) and prevent them from rectifying the test signal from your sweep generator, and messing it up in various ways through leakage paths. It's an easier alternative to removing the switching transistor from the circuit.

Dimitrij

OMG you're right. I've had this issue before when looking at manuals from .pdf files on a screen rather than hard-copy. Well that's just dandy I must say. Now I'm *really* confused. Once again I'm really tempted to just scrap this thing as it stands, salvage the transformer and start afresh in a year's time with a conventional non-resonant PWM design using a MOSFET instead of a BJT and a more up-to-date controller. :(

Sorry, Dimitrij; as you were. Legg has spotted I was mistakenly referring to the wrong diagram in fact. They look very similar and when you don't have the physical hard copy manual in front of you then errors like this are far more easily made, I regret to say. Sigh. I've had enough for today, I'll take yet another look at it again tomorrow. :(

Well, I've replaced all the flaking capacitors and still no improvement. A number of people have been suggesting I remove the two resonant caps (the 30n ones) from the primary circuit and test them. I didn't have any expectation that this would achieve anything since they tested good in- circuit, but as we're running out of ideas now I did remove them this afternoon and they both tested at 31nF a piece and no signs of any physical damage. I then subbed a couple of common-or-garden mylars of the same value in their places and re-swept for changes in resonance. Result was no change in resonance - but a slightly better Q(!!) Also checked the two 10Meg resistors whilst I was at it and they were fine, too. So I can only think of making up Dimitrij's winding tester and looking for signs of any turns shorting in the main transformer. I'm coming to the end of the amount of time I'm prepared to spend on this psu as it stands. I'm more and more tempted to mothball the key parts of it til next year then rebuild it as a conventional non-res converter to a fresh design. TBH, I'm not prepared to still be testing this thing after the end of this week, so if anyone has any last-ditch ideas, now's the time to toss 'em into the mix. Speak now or forever hold your peace. Thanks, all.

Waste of breath.

RL

I have no idea what you now consider to be wrong with the PSU. Apart from a resistor that in your opinion runs too hot even though its well within its rating I seem to recall amidst your ramblings that the voltage rails are correct?

It seems to me that the only fault was the diode which was pretty obvious from the beginning. (Always check for previous repairs).

You persist in not using the correct manual, refuse to test it with an appropriate load and won't put it in the scope to see if the scope actually functions.

If you just want to discuss switching supply design go buy some control chips or an evaluation kit and build some.

Yes, they're fine. It's not *my* opinion that this is a poor design! I posted the schematic to s.e.d and the designers there told me that. I've said all along I know nothing about this type of PSU so I defer to those better qualified. Having said that, the widespread evidence of charring and soot residue that appear even worse in real life than in the photos would seem to indicated that that is not a happy board. The rest of the scope's boards are still pristine showing no sign of repairs at all.

Nothing obvious about it! - apart from the rather poor replacement technique. The problem with that replacement was very subtle inasmuch as it wasn't recovering quickly enough at 20kHz. Nevertheless it was still capable of functioning as a viable rectifier right up to 500kHz.

All the recent tests have been carried out using the actual scope itself as a load - can't get any better than that. I won't put it in the scope until that resistor is running cool. It's proximity to other temperature sensitive components like diodes and its siting deep within the scope uncooled lead me to believe that it is running far hotter than it should be - and will be even worse in situ. That *might* just be a design flaw, but I doubt it.

Yeah, I know what you're thinking - your suggestion still hasn't been adopted. I've been giving it more thought and I've come up with a brainwave.

This is a 2W resistor. If it's trying to dissipate more than 2W as I strongly believe, something's definitely wrong. The problem up until now has been measuring the dissipation, because we can't use I^2*R or variations thereof because of the highly noisy/irregular waveform. So... Here's the clever bit: Measure exactly how long it takes at present for the resistor to reach say 50'C. I believe it's around 1 minute, but get the exact time. Then let it cool completely back to the room ambient temperature. Remove from circuit. Attach to bench power supply and by means of trial and error, set the voltage across the resistor to raise it's temperature to 50'C in

1 minute (will obviously require several attempts, but no matter). Read off the voltage level which produces this outcome, then just do V^2/R to find W and see if it exceeds 2W. I'll do it first thing tomorrow!

[snip]

Finally it gets *really* interesting. >:-}

Good test. It would be more straight to plug the right voltage to make it d issipate 2W, for 20 ohm that would be 6,3V and check what temp results in 1 minute. You could use another identical resistor for the test if you have one lying around to avoid unsoldering.

There's no need to unsolder, neither to actively avoid unsoldering. That resistor is connected through a diode on the board. Just apply the proper polarity signal, and the diode will take care of the isolation.