Zener Based Power Supply

Please define <...improved...>?

This schematic does not make much sense to me for multiple reasons.

First, the diodes you reference are not Zener diodes. They are standard rectifiers.

Second, the fuse is not just underrated for interrupt voltage, it must be able to interrupt a high DC voltage of at least 680 VDC. The single fuse will also not interrupt a phase to phase or diode breakdown fault current.

Third, three-phase 480 VAC power usually provides a significant amount of power to something. The tiny current rating of the initial fuse value should prevent this circuit from drawing a significant amount of power to anything after the fuse. So why bother to take a small amount of power from an enormous, under-protected power source?

One improvement would be to make the drawing actually readable.

Ok, here's a close-up of the schematic: https://i.imgur.com/SwuLn7j.png

I'm considering to use varistors between the line to line input voltage and remove the single varistor at the DC side. Also, I think using 3 independent fuses at each line would be better. The cost could be higher, but the reliability is increased.

Thanks for a readable diagram.

You need 3 phase because a fault might lose one phase, if the same phase as power to earth leakage gismo then it will not work.

I endorse the posts already made about the lack of fuse protection at the 480 VAC supply input. Three phase is significant power - think 20 kA short circuit at building meter for 100 amp incomer fuse - you need a fuse able to break that at 480V 60Hz. Think 20 kA arc welder for what it will do to the PCB & anything near.

A/ The LM317 circuit gives some concern...

1. The ADJ terminal draws 100 microamp max, typical 50μA [minimum not specified], but a 12V zener will be specified at 5 mA typical. Zener impedance at 0.1 mA may be > 1000 ohms. Zener voltage will be less than 12V by 0.1- 0.2 V, but not very significant compared to +/- 5% voltage tolerance. The regulation & ripple rejection will not be as good as data sheet suggests. A capacitor bypass from ADJ terminal to common would be an improvement if your 13.25V load is noise sensitive.
2. Minimum output current for regulation is 3.5 mA typical, 10 mA max for LM317. The 3k resistor only ensures ~ 4 mA & 13.25V load minimum is uncertain [> 6 mA required for 10 mA].
3. FET M1 along with D8 and R2/R3 acts as a current limit circuit. Supposing a FET gate-source voltage of 3V [tolerance may be +/- 1 volt at 10 mA without effect of temperature] suggests only about 6 volts across R2/3 or 12 mA current limit [plus up to 1 mA thro' R1]. That is a problem if your detection circuit draws more than 12 mA - 4mA =8 mA N.B. R5 is drawing 4 mA. Along with item 2 above, it seems your load must be in a "window" between 6 mA to regulate & 8 mA not to hit FET current limit.
4. If your detector circuit requires +/- 13V, the current available at -13V is limited to a few mA by current through D11 zener, due to R5 & +13V load. That is OK if -13V load is less than +13V load.

B/ Presumably SCR U4 is triggered on by your leakage detection. When U4 fires, it kills the LM317 supply. That is OK if the L4 relay coil inductance is low enough for U4 current to reach its holding current before the trigger current ends.

Assuming holding current 10 mA & 10μs pulse length with 480V, I get max inductance 0.5 Henry, resistance neglected. However, if you want function when line voltage has gone to 48V in a fault, that would be 0.5/10 H & L/R maybe no longer ignorable.

C/ The RL1294 spec suggests maximum 100V - how does this lie with 480*√2 volts suddenly applied across two in series via two diodes & 22nF?? KMP1200 gives no limiting voltage at all....

D/ Voltage/type not given for D13/14/15 - do they do anything not provided by V3 varistor & D12 across relay coil L4??

67model

Thanks for your comment. I really agree with you regarding the operation of the circuit. Some points:

1. D13, D14, D15 are TVS diodes. This is the datasheet: 1.5KE300A

I believe these diodes are to protect the drain-source. Are they really needed?

2. The 3k resistor was only to simulate the load in Ltspice. It's not in the real circuit. The 13.25V output voltage feed some opamps used to process the signal from the CT.

3. This is what I know from the inductor: Resistance: 290 ohm +/- 10 ohms at 25C.

Force: Must provide a min. of 2.8N. at 0.059 striking distance at a supply voltage of 115 Vdc. Winding: 3750 turns of 40 AWG

About some improvements:

1. I will add the capacitor you said from ADJ to common.

2. I will replace the input inductors. I will use the 1200LRS I posted above. However, I might lower the inductance to 470 or 330 to increase the current. (maybe ?)

3. Do you agree adding the 3 fuses at the input side and maybe a PTC resettable at the DC side?

4. The varistor is okay like that? Or would it better to use 3 varistors between lines at the input side?

Adding three fuses and three varistors will certainly increase the cost. But we want higher reliability..

5. At last, I was thinking of changing the FET. This is the FET originally used: IRFBG30. It has a little heat sink attached. What do you think of this FET and use the PCB as heatsink?

Also, as the original fuse is 125 mA/250V, I will change it to 500mA/1A depending on the availability. I think 125 mA is low and would break at some normal transient.

Another possible modification, I think, is to remove the three input inductors, and use Pulse Withstanding Fusible Flameproof Metal Film Resistors.

Fusible resistor OK if they are safe at really high fault currents/voltages - data sheet only claims 240V operation - you would have to consult maker if two in series are safe at 480V. Not such a problem if 400/230V 4 wire supply with earthed neutral is application.

Seems a good idea, applying 6 KV step to 680μH gives 9 amps/microsecond di/dt - saturation must soon occur [I assume ferrite core] - goodbye inductance.

You reminded me of another possible problem component...

R1, 700k, has 680V DC across it. Even common 3 watt resistors are not rated over 500V continuous. TT electronics VRW type comes to mind - 1600V rating.

A basic point: your original post wrote that input could be up to 480 VAC. That seems like "USA 60Hz Practice" 3 wire supply nominal - really one should allow 480V +/- 10% for variations. Your diagram shows 392 [volts?] at input terminals, looks like 380V N.B. the european tolerance is +/- 10%.

Replying to your "Some Points"...

Item 1 - The datasheet gives 1.5KE300A stand-off voltage as 256V x 3 diodes = 768V; that is greater than 480V*1.1*√2 = 747V, which is OK. Breakdown at 1 mA is 315V x 3 = 945V, which is less than FET breakdown, but current shunted off at 1000V is unclear, 414V x 3 @ 3.7 amps is above FET tolerance. Question is - how much current/time do you need to shunt? N.B. At this voltage, the behaviour of Zinc Oxide varistors is good; at 1 amp voltage is only 1.2 x voltage at 1 mA. At 1000V the FET [datasheet fig.8]can sink > 2 amps for 100 μsec. At 2 amps 350 ohms coil drops 700V. The real unknown is the shunt capacity of the inductor [nF] & supply impedance.

Item 3 - Best to make a measurement across 5 ohm resistor in cathode lead of SCR [anode direct to inductor/coil] with storage oscilloscope when SCR is triggered.

Replying to "Some improvements"....

Item 2 - Beware of assuming improvement - your proposed replacement has only 7 mm between its terminals compared to 18mm on original. Internally I expect clearance is much less. What do you think that does to its impulse voltage withstand? The new item does not specify a voltage limit - that means it is low, not infinite!!

Item 3 - you certainly need fuses at 480V, but user ought to have fuses to protect wires from distribution to leakage detector & detector. Distribution HBC fuses are usually 2 amp minimum, but 32mm x 6.3 mm fuses can have mA ratings, blow in < 5ms @ 10x rated current & 30kA breaking capacity. It matters a lot how long cable runs are - 10m run of 1 mm² Cu @ 480V two wire is 0.34 ohms, 480V/0.34 ohm = 1411 amps. But if your relay is in distribution board incomer 20kA or more. Also, resistance of 680 μH inductors is 2 x 1.6 ohm = 3.2 ohm - that is 480/3.2 = 150 amps short circuit - provided the inductors do not short before a fuse blows. By my reckoning 40 AWG wire has I²t of ~0.3 which is same as really fast blow 800 mA fuse like SIBA 7017240.0. It is not clear what happens to Inductor L4 at 2 amps/480V for any time - presume energising L4 cuts supply in milliseconds & there is a flag or target released to tell it was earth leakage caused that - a PTC seems unnecessary.

Item 4 - Varistors at input better than DC side because the diodes do not have to carry surge current & may fail. It is not clear what surge voltage/energy you want to survive - 600V rated equipment at near to power line [Category III] must live with 6 kV impulses to EN61010 standard (UL3111 similar). This helps...

http://www.ni.com/product-documentation/2827/en/

Item 5 - As pointed out, R2 [5.1 ohm] & FET set a current limit of 1.2 amp, while 290 ohms of load @ 480V rectified is 2.2 amp. Conclusion is FET sits at 300V & 1.2 amp [360 watts] until it overheats & fails [rating 125W @ 25'C case] unless something trips supply quickly. The heat sink thermal mass may be needed for FET to survive long enough.

N .B. 3 phase rectify of 480V gives minimum 85% of peak voltage through cycle, while single phase dips to zero volts without reservoir capacitor. Two more diodes are a lot less cost & more reliable than reservoir cap.

Thanks again. Your comments really make sense.

My intention is not to completely alter this design, rather I was trying to see possible ways of making it more robust. Now, there's something I haven't told you. The reason why I'm trying to improve the circuit is because in the real PCB, the original fuse often blows without a convincing reason. That is why I thought the main problem was the underrated fuse 125 mA/250V.

So in my conclusion, possible improvements are:

- Increase fuse rating at the DC side, or use 3 600 VAC fuses at the AC side.

- As in my previous post, the fuse resistors could be an option instead of the inductors at the input side. I will search for a higher rating.

- Varistors at the input side are ok, but having varistors both at the AC side and the DC side would increase the cost.

- This FET seems good. It's SMD but as I said, I could use the PCB as heatsink. Also it's rated at 1200V VDS.

I will go over R1 as you pointed out.

Found this fuse resistors for >600 VAC http://www.vishay.com/docs/28915/pr02fs.pdf

I think they can replace the input line inductors.

The PR02FS data sheet gives it a 600V impulse rating [1.2/50μs], but unlike TT EMC2, it does not say it is safe if you connect it across 240VAC.

I was wondering about the purpose of the fuse - it may be to avoid burnout of solenoid & FET if the supply is not cut [or user tries to test without inconvenience of tripping off his load].

Looking at a glass tube fuse data @ 10 x rated current [1.25 amp for 125mA rating], blow time is 5 to 70 ms - this may be too short at 5ms to trip supply. Current for 10 ms minimum time was 5.5 x rated [use 250mA fuse, but maximum too long for solenoid/FET] - but is even 10ms enough to trip??

I suppose the trip time, by MCB/MCCB shunt trip, of customer supply is outside your control.

The question "How does customer know it was Earth Leakage relay ELR, not something else?? remains unanswered.

If the supply is tripped OK, including the ELR, what is the ELR stress - if the load is a motor, or part motor, will the inductive voltage surge damage ELR? I well remember making the mistake of opening the isolator to stop a 110 VDC motor at 25 amps without disconnecting Digital meter reading motor volts first. The meter gave out smoke - internally the + circuit had flashed over to a negative PCB track, which itself was burnt away over 30 mm - meter read low & non-linear after track repair, kaput. That meter was Cat II 600V rated - 4 kV transient to IEC1010.

I looked at the IXTA3N120 SMD FET spec - the SMD one has 1/3 the metal in it compared to TO247. As I noted before, the FET can sit at 1.2 amp 300V = 360W while SMD is 125W max at 25 'C case. The case temperature will rise 20'C per 10 ms, based on its thermal mass [56 cubic millimetres] - heat sink with 0.5'C/watt temperature rise due to thermal resistance case-sink (180'C rise) makes little difference.

Increasing current, so FET is saturated cuts its loss to ~15watts, but the solenoid coil will overheat in 75ms & fail.

A solution might be a PTC thermistor in place of fuse, as you suggest, or changing U4 to a FET & driving by a 50 ms pulse.

I was thinking that in this design there's no significant inrush current as I don't have much capacitance after the bridge. This means the fuse resistors are not necessary. I think normal fuses instead of the inductors should be better.

You must think why the design had the inductors.

This device is 3 phase & likely intended to trip a main incomer. It seems main incomer trip cuts the detector supply, along with loads. Any transients due to trip hits the diodes. Also devices at incomer see worse transients than normal sub-circuits.

By just fitting fuses you have a problem with the breaking capacity of the those fuses at 20kA or more fault level - the PCB & fuses may turn into a fireball. Of course, suitable 32 mm fuses are available but are not "normal", probably need holders & are quite expensive

You are the one who knows which parts have failed most often - is it bridge diodes? If bridge not failure problem, removing inductors may make one.

I think the fusible resistors may be equally good as inductors - only you can consult resistor makers as to if two in series are safe applying 480V. It will not be expensive to burn a few pairs up in your own tests on PCB.

My opinion is that for 480V there ought to be 480V rms HBC fuses of minimum current rating able to break fault level existing at monitored source & protect the short wires to detector according to code.

Just as an example 20 kA rms @ 480V is 24 mΩ resistance; 1 metre of 2.5 mm² wire from busbar is 7 mΩ - which reduces fault level significantly, but it is still high. If at an incomer, these could be the fuses [same circuit] that protect incomer voltmeter & switch. I would expect these to be available at 1 or 2 amp rating minimum with a certain cut-off current & I²t let-through spec & that your device will not catch fire or throw out dangerous debris when failing with that fault level. Other fuses may not be necessary, if you do not require the board to be repairable after a heavy fault. I would expect your installation instructions to mandate suitable HBC distribution fuses on the circuit external to your device.

If diodes have been failing, this may be due to high voltage spikes at CAT III/IV location 4/8kV. VDRs would help in this case. In the days of mechanical consumers' energy meters, these had VDRs to protect windings and nearly every one of the millions of meters used rolled on for 40 years, these VDRs also protected installation wiring from overvoltage. You may find it instructive to look inside a modern electronic industrial 3 ph energy meter to see what it takes to survive the incomer environment for a lifetime & not be dangerous if they fail. This includes sufficient leakage path lengths & clearances on simple items like terminals & PCB tracks.

I don't recall the diodes fail. I think they have never failed. The part which fails the most in the original design is the fuse. The FET has not failed so far, but I believe the behavior is too close to the top of the SOA when the trip coil turns on. Your calculation of 300W sounds right. I endorse your advice about selecting a higher current FET. I was looking inside some 3 ph meter designs, and all of them use fuse resistors at the input. Some of them don't even use varistors at the input line to line voltage. I have to check space constraints to see if I can actually place three varistors.

It would appear fuse was intended to blow before solenoid coil - unfortunately, tolerance on normal fuses is such that it might blow before solenoid has actuated - or after it has actuated, but before breaker trip - putting device out of action until fuse is replaced. A PTC might have a repeatable enough spread of time to do job.

I suspect the designer might have looked at maximum fusing time curve rather than min-max curve & made wrong assumptions about what minimum time could be. A lot of "wiring code" fuse curves are concerned with maximum clearance times for safety and do not show the minimum times.

I looked at...

https://www.mouser.co.uk/datasheet/2/281/bk-e0937f-ptgl-sas-51v-e-1518535.pdf

the 6r8 value might replace your 5R1 in FET source - but data sheet gives no idea of how long it takes PTC to heat up when current exceeds trip value, nor tolerance on resistance/current curve.

Personally, I would try to get a far more "definite time" drive to the solenoid by putting a capacitor C across D8 and a resistor R from U4 anode to FET gate [with diode between U4 anode and R2/R3 junction]. Firing U4 would drop FET bias with RC time constant [maybe SOA problem with time-voltage problem]. Crude values R = 5k, C = 10 μF tantalum gives RC = 50ms N.B. tantalum have better μF tolerance & leakage than aluminium. I do not know the worst case trip time for the solenoid.

There would be problems with U4 "off" current affecting FET bias & R1 current varying with line voltage [supply volts*R/R1 gives a residual voltage biasing FET on] , but tolerances may be much better than 5ms to 75ms tolerance for fuse at a given current, let alone fuse current variations.

A source-drain resistor on FET would assure a hold-on current for SCR with FET off & take some of the watts off the FET in not-trip normal state.

Maybe there is a spare op-amp in the analog section to give a pulse & U4 can be changed to FET as I suggested before.

It may be better to just add a CMOS gate IC, R-C and swap SCR to FET, rather than struggle with all the tolerances of extra components on SCR circuit. For example, driven by a positive step from leakage detector, a 2 input AND gate with one input direct to step & the other with resistor to GND and capacitor to step - or better 2 NAND gates with second as inverter - to give square positive pulse of 50? milliseconds.

The military design guides suggest that it is better to combine a set of proven blocks with clear functions than use less components, but have a circuit which is difficult for understanding & fault-finding & susceptible to component tolerances.

My old company had a concept NRFT - "Not Right First Time". Experience showed that development time & cost over-runs were usually due to NRFT, and if NRFT got into production, really expensive. So mandatory effort was applied to identifying risky items not based on proven tech , researching the problems, building demonstrators to prove solutions etc.

I'm ordering some components for testing. I've selected this PTC PTCTL7MR100SBE instead of the fuse F1. It has a trip current of 400 mA, and it's rated at 600 Vmax. The max trip time is 7s if 1A is flowing through it.

I'm still finding another FET with higher current for the trip coil. The STP8N120K5 looks good, but it's quite expensive compared to the old one.

Regarding to replacing the SCR with a FET, it sounds good. But the SCR and the leakage detector (built with opamps, comparators, inverters) are in board separated from the power board. I'm not paying much attention to the control board, as it has never failed and it seems to work very precisely. I think the weaknesses are in the power board, specially the fuse. The control board draws very little current, that's why the current limiter circuit in the power board provides no more than 12-13 mA. When the SCR is activated, the 510 ohm resistor in the power board gets bypassed and that's the point where the trip coil draws more current until the three phase supply is quickly cut off.

If there were a mechanical issue where the trip coil gets mechanically stuck, and the power supply is not cut off quickly, the coil is going to draw much current and the fuse probably blows. The PTC would offer more room I think and I don't have to change the fuse every time it blows.

A/ There are some fundamentals to applying a PTC. Supposing PTC & coil are similar cold resistance, the PTC must have a switch temperature Ts and thermal capacity J/K matched to coil protected. If PTC has higher Ts than max insulation temp of coil or greater J/K then coil will overheat first.

B/ Your coil, with 40 AWG wire; 0.0031 inch diameter has 0.00487 mm² area.

The adiabatic withstand time for copper with K = 115 - a temperature rise from 70 to 160 Celsius, is given by the equation [from wiring regulations short-circuit withstand time].....

t = K²s²/I² seconds - s in mm², I amps

Hence t = 0.145 seconds for K=115, s = 0.00487, I = 1.47 amps

N.B. Looking at IRFBG30 FET data Vgs threshold is 2V minimum. Fig.3 indicates that at 150'C, Vgs for 1.5 amp is 0.5 volts more than threshold volts at 25'C. Taking Vgs = 2.5V and D8 voltage of 10V, that leaves 7.5V across R3 of 510 ohms - hence I = 7.5/5.1 = 1.47 amps maximum when the FET is current limiting. Taking 1.5 amp in Fig.3 at 25'C one gets 4.5V which gives minimum regulated current as (10-4.5)volt/5.1 ohm > 1.08 amp [N.B. Fig.3; 4V at 25'C is <0.1A, compare Max threshold spec. 4.0V]

One must suppose that, after detecting excessive leakage, the main breaker is tripped within 0.145 seconds (solenoid operate time + breaker opening time). Also that solenoid trip current is much less than 1.47A.

C/ One should not forget that that 1.47 amp means the FET sees supply 648V less (290 ohm x 1.47 amp = 426 volts) = 222 volts @ 1.47 amp = 326 watts, far more than continuous withstand and in excess of SOA safe operating area 200V/1.1 amp for 10 ms in Fig.8 of its data sheet. The FET would fail short circuit so trip breaker.

D/ Given that 40 AWG copper is 17.24 mOhm/m for 1 mm² cross section; 0.00487 mm² is 17.24/0.00487 = 3540 mOhm/m from which 290 ohms for coil is 290/3.54 = 82 metres. Hence the coil has 82x 10³ x 0.00487 = 399 mm³ of copper ≅ 0.4 cm³ weighing 0.4 x 8.9 = 3.56 grammes; with specific heat 0.385 J/gK that is heat capacity of 1.37 Joules/Kelvin.

E/ Viewing EPCOS/TDK PTC data sheet https://www.mouser.co.uk/datasheet/2/400/PTC_ICL_OC_Leaded_260V_1000V-1220369.pdf one finds that the 500 ohm C773 version [N.B. minimum spec. for 500R is 320R @ 25'C] has 120'C operate temperature and J/K = 0.6. Hence two in parallel would be 250 ohm, 1.2 J/K, thermally equivalent to 290 ohm with 1.2*290/250 = 1.39 J/K. Note that final current would be 5 mA @ 650V, see diagram in data sheet. Maybe TDK have similar PTC with smaller diameter & lower J/K.

F/ I wonder if it is worth considering driving the coil direct from the SCR via a PTC, but that depends on the required current/maximum time delay for the trip coil. Note that with a minimum current from FET of 1.1 amp [Due to threshold tolerance, see B/ above] the coil could get only 1.1 amp x 290 ohm =320V or half of 648V 3 phase rectified 480V rms.

67model

The resistor R1, 500k 3W is ROX075500KJNEL. However, I want to replace it with this one ‎RR03J430KTB. I think that going a bit lower from 500k to 430k is not an issue. What do you think? ROX is expensive. Also, I think it's a good idea to replace R2 or R3 with a PTC to ensure the load doesn't draw more than ~12mA so not to hit the FET current limit. By the way, I haven't found a proper FET yet. Still looking.

P.D. Just in case you don't have the schematic, this is the close up: https://i.imgur.com/u1wOLOM.png

The FET needs to be at least 1.2kV, >3A. This one looks good, but it's expensive. STP8N120K5

Comparing STP8N120K5 [A]with IRFPG50 [B], I find...

1. Thermal resistance, junction to case, A = 0.96'C/W: B = 0.65'C/W, B wins.
2. Mass of metal in case (Joules/kelvin heat capacity) A TO220 vs B TO247, B wins.
3. Safe Operating Area SOA 10ms: A = 0.3A @ 1kV: B = 0.5A @ 1 kV; B wins.
4. SOA 10ms @ 200V; A = 1.5A; B = 2.5A; B wins. FET current limit/coil 290 ohm puts DC operate point at about 220V & 1.5 amp.
5. Vgs threshold: A= 3 to 5V; B = 2 to 4V: "A" smaller range but R2, R3, D8 increase to get small improvement.
6. B lower cost; B wins.

With regard to point 4, considering that the coil resistance will increase about 50% at its max. temperature, the FET sees very little final Vds volts at 1.47 amp max.; 50V @ 1.37 amp typical or 150V @ 1.14 amp min. - which may help considerably with Tj & SOA perils for FET, but it is all a matter of trip timing vs thermal resistance, heat sink J/K & SOA. N.B. 480V rms gives 648V mean DC supply used for calculation.

Possibly FETs have survived in service, because your stock items have always been typical to low threshold gate volts, keeping the volts & Tj down.

I must clarify that the FET used in the actual circuit is IRFBG30, not IRFPG50. I will do the comparison anyways.

I think you wrote ....30 and I used its data before, but circuit diagram has IRFPG50!

SOA 10ms for IRFBG30 appears 1.1 A @ 200V, so ...50 definitely tougher.

These VR68, 7 kV rms max, are fraction of price of ROX....

https://uk.rs-online.com/web/p/through-hole-fixed-resistors/6835304/

Even two in series for watt de-rate cheaper than ROX. RR03 seems adequate

Small reduction of resistance of R1 is not a problem, however RR03 spec. lists E24 values so 510k or 470k should be available, 500k is an unusual value.

PTC have poor tolerance usually. In any case, max FET current is given by Vgs threshold minimum= 2V, which gives (10V - 2V) across R2 & R3 i.e. 8V/515 ohm = 16 mA giving FET 9.5 watts for 480V rms supply. Threshold maximum spec. 4V gives 11.5 mA. I do not think spread with PTC could be that tight

If you want 12 mA max, change R3 value and use min. threshold Vgs at working Tj of FET.

Although DC safe area is not in data sheet (check with Vishay), I do not think 20 mA is problem @ 650V or 1000V. The IRFPG50 FET seems to be a problem for failure due to excess Tj if trip time after leak detection is too long. Very approximate I estimate case temp rise as 25'C/50ms @ 330W but this is nothing compared to RthJC 0.5'C/watt x 330W = 165'C rise possible 50 ms after SCR fires [see Fig.11 of datasheet].

I am surprised the FET has not been a failure problem - you have never specified the time to trip & cut leak-detector 480V supply after SCR trigger.

Possibly inductance & operate current/delay of coil are such that FET is saturated, rather than current-limiting, until supply is cut. In that case, why not connect SCR direct to coil??

About the trip times, they can be adjusted up to 1 sec. It may be 10 mS, 50mS, 100 mS, 0.5 sec, and 1 sec at worst.

I'm planning to use this circuit with 600V (rms) input voltage. I have found this FET IXFP6N120P which looks promising as it includes a Forward SOA. I will check if I just need a little heat sink. Ideally, I would like no heat sink, but I have to be realistic. In addition, what do you think of adding a resistor in series with the coil to reduce stress over VDS? That resistor should be greater compared to the DC resistance of the coil. Maybe >10k could be okay. This can also help remove the three TVS diodes in series originally used to protect the VDS.

I guess that "up to 1 sec" is time from detecting leak to firing SCR, which is not relevant to FET stress. Calculating for existing design....

1. With 480V rms in = 648V mean, FET is continuously carrying 16mA max. at (648 - 33V) = 9.84W - call it 10W, since that "480V" will be more than that most of time.
2. IRFBG30 thermal data is 1'C/W junction-case + 0.5'C/W case to sink. Small extruded heat sink 38 x 22 mm is 5'C/W. That makes 6.5'C/W total x 10W = 65'C rise above ambient.
3. Ambient 50'C makes junction temp 115'C before SCR fires.
4. Taking typical Vgs of 3.5V & D8 10 volt, there is 6.5V across R2, so Id is 6.5V/5.1 ohm = 1.274 amp initially when SCR fires.
5. The voltage across the coil will be 324 ohms [@ 50'C, for 290 ohm @ 20'C] x 1.274 = 413 volts, leaving 648 - 413 = 235V across FET which makes 235 x 1.274 = 300 watts lost in FET initially after SCR fires.
6. Per item 3, with Tj max = 150'C, you have only 35'C left, requiring Rthj-c to be 35/300 = 0.12 'C/W. Referring to Fig.11 of data sheet, that only applies for < 1 ms.

This leaves out the effect of coil inductance [depends on air-gap/iron area & length, no data] and heating of coil, however my feeling is that these are "small beer" compared to above & the design has only survived because all the production FET were less than typical Vgs.

It is quite possible you got 100 FET in a box, which came from the same silicon wafer and are all similar & better than typical, but the next box could be worst case high Vgs!

Now the important issue of longest delay after SCR fires before breaker trips + cuts supply to zero volts, removing heating from FET &/or coil.

Breakers may be "3 cycle" or more, say 60 ms plus...

Add to that operate time of coil/mechanism. What happens to Leak detector if breaker does not Trip? "well its burnt out innit - your relay failed!".

Note that, in a previous post, I estimated that coil would not overheat for 75ms if the SCR drove it directly [ I left out the heating of coil increasing its resistance, so estimate is pessimistic].

You could add a resistor in series with coil for 600V rms e.g. 290 ohm x (600 - 480)/480 = 73 ohms, but 10 kilohms is "Lucy in the sky with diamonds [Beatles]".

Regarding TVS & protection of FET against voltage surge from supply or turn-off surge, have you considered, say, Littelfuse VDR V510LA80CP? That is 902V @ 1 mA nominal & 440J energy withstand [far more than TVS]. When IR made them, the characteristic was claimed to be 20% voltage increase per 1000:1 current increase . They are used to protect transistors driving auto electronic ignition coils. One dodge would be to add the 73 ohm on high side of L4 coil - its end to end capacitance would be much less than the coil, reducing spike coupling to FET.

67model

I think I'll let the TVS diodes there. But I will use just two higher rating diodes instead of three. Let's say I will use two 485V diodes instead of three 300V diodes.

Could you explain again why 73 ohms would be an option in series with coil?

When you say: "What happens to Leak detector if breaker does not Trip? "well its burnt out innit - your relay failed!". Let's imagine there's a mechanical issue that prevents the coil from tripping. In that case the coil will continue drawing current, I think that's one of the reasons the fuse blew in the original design. I'm considering to remove the fuse, but I have to consider that overcurrent scenario.

Lastly, as for the FET, what's your final suggestion? I'm thinking of doing some bench tests with different FETs if needed. The original FET has this heat sink attached. If I have to use a more expensive FET, but get better results, then I would accept the deal. Of course, once I make sure I get reasonable power calculations. According to your calculations, which looks good, I should change the FET.

I forgot to add that I have come across the case where the coil is found to be open, but not the fuse. 40 AWG has really limited current rating.

40AWG copper has a fusing current of ~1.5 amp - but that is bare wire at melting point in glass tube. Current rating as insulated wire, open to air, is only tens of mA.

My textbook on protection relays recommend not less than 0.004 inch diameter wire, great care with winding to avoid kinks and securing of wire end to lead-out. Encapsulation & kilovolt induced voltage test to detect manufacturing winding faults were done. Also connection of one end of coil to negative supply to avoid electrolytic corrosion [your coil is positive].

I would expect higher volt diodes to have (clamp volt/working volt) higher and temperature coefficient [+ve] higher. This means initial volts on surge are higher & increase due to tempco during pulse is more.

Resistance of 73 ohm gives same current with total 290 + 73 ohm & 810V as 290 ohm with 648V. Fundamentally,

A 648V source with 290 ohm can deliver 362 watts max.

810V with 290Ω can deliver 566 watts into the FET. Calculating for FET as current sink gets 536 to 564 watts for currents 1.078 to 1.47 amp and ~400V. Clearly, operation is near the "matched power" point & the coil will get nearly the same power as FET.

For both coil & FET the power is increased over 50% from 648V level by using 810V . Obviously the time profile depends on the temperature rise & resistance increase of the coil, which I described in previous post. Both coil & FET will overheat quickly.

After failing to find a much more powerfull FET without big increase of cost, I looked at IGBT in TO247 like NGTB40N120FL2WG...

https://www.mouser.co.uk/ProductDetail/ON-Semiconductor/NGTB40N120FL2WG?qs=%2Fha2pyFaduhIpzaYIlPYHXbgbo9F7Tl9vv921voBqQCwRfFL8GAPNQ%3D%3D

They have lower thermal resistance & higher watt/amp capability, but for 810V case still looks too much - also Figure 17 of data sht [S.O.A.] has note "Curves must be derated linearly with increase in temperature" which needs to be understood.

My security programme refused to access your heat sink link "old, insecure protocol", but it is straightforward to calculate sustained FET temperature before coil pulse.

67model

I previously thought of an IGBT, but I was skeptic about it working in the linear region rather than switching. I'll review the one you found. I would use it if I wouldn't have to use a heat sink.

This is the heat sink I have for a TO-220 package: https://www.digikey.com/product-detail/en/aavid-thermal-division-of-boyd-corporation/7139DG/7139DG-ND/1625513

OK, the catalog link worked & graph shows 100'C rise at 3.8 watts for 7139 type.

Working through your circuit diagram - D8 is 10 volts, so the voltage across [R2 +R3] is 10V minus 4V (max Vgs threshold on IRFBG30 data) = 6V.

Hence the minimum FET current @ 25'C is set to 6V/515 ohm = 11.6 mA & similarly max is 8V/515 = 15.5 mA, ignoring resistance variations & Vgs fall with temperature rise.

D7 is 43V, so source of FET is nominally 40V and drain is at 645V for 480V rms supply making 605V across FET. That makes watts lost in FET 605V x 15.5 mA = 9.4 watts. Call it 10W, since Vgs falls ~0.5 V @ 150'C Tj.

So temperature rise above ambient would be 100'C *10W/3.8W = 263 + 15 degrees junction to case/case to sink. = 278 + 25 ambient = 303 Celsius.

Tj max = 150 Celsius Absolute Maximum Rating!

It won't be quite that bad, heat loss rate W/ΔT increases with ΔT, I seem to remember 250'C as maximum at which solders began to melt in metal power transistors.....

I do not think you can operate without a heatsink at 480V rms, let alone 600V, the one you have is inadequate. That these detectors have passed bench test must be due to the FET manufacturer's margins for long life & reliability or Perhaps the R3 value was increased & only used at 380V rms.

The question comes up, how did the FET survive - already overheated when SCR fired - the answer must be that the coil resistance of 290 ohm limits power to about 300W - for 10ms that is 3 joules, while the silicon & copper in the FET itself have heat capacity sufficient the extra temperature rise was not destructive before supply was cut.

IGBTs are hardly more non-linear than BJT or FET, they are all approximately exponential law devices. Figure 17 of NGTB.... data sheet does give a "DC safe area"; which many FETs do not, but most high voltage FETs are described as for "switching". The NGTB.... data is bad in that it is at 25'C, rather than Tj max, which many applications will be near.

Some additional details. This circuit is used in a three pole earth leakage protection module. That's why three phase input. I'm not sure if it could be implemented just using a phase voltage instead of the three phases. That way I could use a smaller single bridge rectifier. The main idea behind this circuit is that when the coil L4 is fully ON, it will cut off the main power. The control circuit (I'm not showing) will determine this condition.

<...What do you think of these changes and what else would you improve?...>

It sounds as though no further comment is necessary.