Zener failed short.

[davelectronic]

Sorry for the confusion. Your schematic is accurate.

[David Hess]

Darlington transistors can be finicky in that circuit so their gain needs to be carefully controlled.  They really are not needed but if you insist ...

At a minimum, the 100 ohm input resistor should be significantly reduced in value so that full output current is reached when the regulator also reaches its full output current.  The base voltage of the Darlingtons, when combined with their emitter ballast resistance, is about 2 volts so there should be 2 volts across the input series resistance as the 7812 approaches full output to maximize control over the transistors, which yields about 1 ohm.

A better configuration is shown below, although not for the Darlington configuration.  Two diodes are placed in series at the input to compensate for the 2 x Vbe voltage drop of the transistors and then the ratio between the input resistors controls the current sharing.

[Ian.M]

@davelectronic: Thank you.

@David Hess: I definitely prefer your suggestion with a Sziklai pair pass transistor .

@all,
There's no easy way to tell how or why it initially lost regulation or whether the Zener cooked off just sitting there because its knee voltage was too close to the output voltage, but once the Zener failed short, the LM7812 regulator, with about 20V across it,  would have rapidly gone into limiting, so the base drive to the pass transistors would have become fairly weak and they probably didn't have enough drive to pass enough current to blow the fuse.  They then sat there cooking for long enough to blow them.  At 20V input to output differential, a TIP147's SOA curve says it will die if it has to pass much over 6A for very long, even on a perfect heatsink.  That's not enough for two of them to blow a 15A fuse before the transistors blow.

How to fix it while still using Darlingtons?  Well if you want the regulator's current limit to protect the pass transistors, the current through the transistors must be proportional to that through the pass transistors and must limit before the pass transistors pass excessive current.  This wont be practical as the peak output current of a LM7812 can briefly reach 2.5A, but its only good for max. 1A continuous, so you'd need a *LOT* more transistors to let you derate them to only 40% of their SOA limit so they could handle the short-circuit surge.  Therefore you've got to come up with an alternative protection circuit.   One approach would be to monitor the drop across the emitter resistor of one of the pass transistors, and use it to turn on another transistor to rob them of base current when it exceeds about 0.6V

Also its advisable to let the LM7812 current increase in proportion to the pass transistor current so its under similar thermal stress.  Then if all of them are on the same heatsink the LM7812's thermal limiting will also protect the pass transistors. To do that you need to compensate for the Vbe drop of the Darlington pass transistors, by feeding the regulator input through a couple of diodes, and make the currents proportional by feeding it through a resistor of N times the emitter resistors, where N sets the current ratio.  The proportionality breaks down at low currents so to prevent the pass transistors turning on due to the regulator's quiescent current + minimum load current, you need another resistor across the diodes chosen to pass the required minimum current with approx 1V across it.

Finally to crow-bar this safely, you need to sense the output but crowbar the input. To get rid of the excess output voltage without blowing the regulator when you crowbar the input, you need a really beefy Schottky diode across the whole regulator + pass transistor assembly, reverse biased in normal operation.  Put the fuse between the reservoir caps and the rest of the circuit and use a big enough SCR or TRIAC to handle the I²t to blow the fuse, and everything except the fuse should survive a crowbar event.  I prefer a precision crowbar circuit - see a TL431 datasheet for details.

I've drawn up a possible arrangement in LTspice.  Sim attached.

Edit: design FUBAR - I under-estimated the leakage current leading to uncontrolled output voltage at max temperature with no load.  Change R3 to 15 ohms to fix.  That makes the LM7812 do more of the work at low load currents, but that's no bad thing.

 

[davelectronic]

Thank you for all the help, i'm probably going to have to go over your theory a couple of times to see what is going on, as i said my maths is quite poor, ohms law not to much trouble, but multiple equations i find really tuff. I did try another transistor i have the MJH11021G PNP Darlington but it also showed exactly the same as the TIP147 output of almost twice the voltage.

[David Hess]

@David Hess: I definitely prefer your suggestion with a Sziklai pair pass transistor.

That is from a very old National application note.  (1) National recommended the Sziklai configuration because NPN power transistors were more common then but that hardly applies now.  They did *not* discuss using PNP Darlingtons but I described how it can be done by using two diodes in series instead of the single diode which is shown in the example.  The low value parallel resistance across the base-emitter junctions prevents the transistor from turning on at very low output current.  The advantage of including the diodes is better control of the current sharing.

Personally I have only used big PNP transistors but with good results.

(1) National Semiconductor Voltage Regulator Handbook.

[Ian.M]

Thank you for all the help, i'm probably going to have to go over your theory a couple of times to see what is going on, as i said my maths is quite poor, ohms law not to much trouble, but multiple equations i find really tuff. I did try another transistor i have the MJH11021G PNP Darlington but it also showed exactly the same as the TIP147 output of almost twice the voltage.

Its well worth wrestling with learning the LTspice simulator.  Its Free (unless you are in the semiconductor manufacturing business), and its professional grade as Linear Technology and their successor Analog Devices use it extensively in-house.  The UI can be a little clunky and it doesn't have dumbed down animations and on-screen 'multimeters' etc. - if you want that with a pro-grade simulation engine, expect your wallet to rapidly go on a crash diet!

Once you've got the hang of LTspice, it does nearly all the circuit maths for you. and spits out the results as voltages, currents, power dissipations etc.  available at a mouse click, + pretty graphs - waveforms with respect to time or curves to represent sweeping a parameter value as I did above for load current.  You rarely have to do anything more complex than back-of-an-envelope Ohms Law calculations to choose a resistor to start with, or maybe calculate a RC time constant to help you choose a capacitor.

Now back to the excess output voltage question - any regulator can deliver too much voltage if you don't meet its minimum load current requirement.   Lets look at the Fairchild LM78xx series datasheet.  It says in the LM7812 table on page 8  parameter 'regload' (Load Regulation) is specified from 5mA to 1,5A as typ. 11mV, max. 240mV.  The mV numbers aren't important but that 5mA lower limit of the specified range *IS*.  Below that load current all bets are off.  That's why I included R6 the 2K2 resistor in my design.  Its got nom. 12V across it so it draws just under 5.5mA, to satisfy the minimum load current and have a bit left over to mop up leakage current through the pass transistors.  Decreasing it to 1K to draw 12mA would help if they are notably leaky when hot.  Its also a good place to add a LED power on indicator, as long as you reduce the resistor to allow for the LED voltage drop.

On the same page we can see the max quiescent current is 6mA.  That + the R6 min. load current must come through the regulator's input pin and R3 has the job of keeping the pass transistors Vbe low enough so they don't turn on significantly with no load on the circuit's output.   

Setting up a quick test jig in LTspice, to bias a transistor using the OnSemi TIP147 model [TIP147 datasheet] I provided with a voltage swept from 0 to 1V and measuring its collector current, with the effective temperature set to 150 deg C (Tj_max) leads to the conclusion we *ABSOLUTELY* *CANNOT* tolerate 1V Vbe at min.load current when hot as the pass transistors would leak 14mA each and that 2K2 resistor has only got 0.5mA to spare after meeting the LM7812 minimum load requirement. 

You could do a similar experiment on your bench with a 1.5V battery (use an alkaline AA, not a lower capacity battery, so the voltage holds up for the duration of the test) and a 1K pot across it to provide the base bias.  Hint: include a suitable resistor in series with the transistor collector so you don't pop your multimeter's low current range fuse if you turn up the base voltage too high!  Also its going to be more of a PITA in real life heating the transistor close to Tj_max - you'll have to put it on a heatsink with a thermocouple temperature probe with a smear of thermal paste in a close-fitting drilled hole right next to it and heat the finned side of the assembly with a hot air gun as close as you dare to Tj_max.

Back to the LTspice results.  Double clicking the curve title (labelled 'Ix(Q1:C)') and sliding the resulting cursor around tells me the leakage current becomes excessive  (over the 0.25mA per transistor we can tolerate) with 420mV Vbe.  The voltage across R3 will exceed that with only 4.2mA through it which means I've FUBARed the design in my previous post - if you switch off the load when the transistors are hot the output voltage will climb uncontrollably, and trip your crowbar.  The fix is to decrease R3.    It needs to pass 11.5mA without exceeding 420mV across it. Ohms Law says it cant be over 36.5 Ohms.   That would initially lead you to a 33 ohm resistor, but so far this is just a sim-w@nk and I'd be happier with 15 ohms for a much greater safety margin in real life, as the pass transistors h_FE may be significantly greater than the 10K typical value the model uses, or its internal base pulldown resistors may be higher in value.

TIP147 test jig sim attached.

[davelectronic]

Thank you for the explanation, this is really interesting. As a novice it's quite heavy reading, but I do get it, although I will go over it more than once just to be 100% that I've got it. From first reading, it looks like the configuration I used doesn't meet the requirements for the darlington transistors, like the minimum requirements are not being met to bios them in the on state . I might not have that quite right though. Thank you for the software simulation information, I'm primarily a Linux user, but can knock up a Windows machine for this software use. I'm going to do some more reading as well, but thanks again, your post is greatly informative.

[Ian.M]

LTspice is tested to work under WINE, so no need for a windows PC or VM, unless you are really unlucky with your distro's WINE support.

[davelectronic]

Usually use Mint, but have used others. Wine can be hit or miss, I've had mixed results using it. But thanks, I will certainly give it a go.