alternator voltage regulator - debugging

[Boschi]

starting from a pre-regulator circuit for a bench PSU i slowly designed and built a voltage regulator for a motorbike. i did varius test on some perfboard, i semplified the circuit and i made my own pcb for it.

i attached the schematic. (its not very clear, but the ac input goes to the pins 2 and 3 of D1 and D4. i have two coils so i use two rectifier bridges, the coils are in phase, so i dont risk to just keep the scr conducting forever)

when it came in the mail i assembled it and it works... kinda. the input signal from the bike its a ac voltage that get up to 70v (140vpp, open voltage). to test it i used a trasformer (~60vpp and 50hz here) and it worked. i tested on the bike and it also worked, but just at low rpm. the alternator has an output frequency of 50hz up to 500hz, while revving the engine it just stop regulating and the output voltage start climbing.

after pulling my hair out for a while i realized that the problem occurs with just one of the coils, the "high" voltage one (it should be somewhere in the 12v 25VA region), but not with the "low" voltage one (6v 25VA). after probing around a bit i realized that the rectified voltage across the pins 1 and 4 of the gbu1510 it drops to 0 for a very short period of around 5us before rising again to a higher voltage so the scr cant turn itself off.  the hv coil has sharper rising/falling edge than the lv one, so its not that strange.

attached a screen of the rectifier output voltage.

im not really sure how to fix this really, andy advice?

[moffy]

I am sorry but the way it is drawn makes it difficult to follow, you need to show the terminals for the coil inputs, as well, I assume, the field coil output. I haven't spent any time decoding your control scheme except to say I am not familiar with it. Normally a PWM signal would be generated to control the field based upon the voltage output from the alternator which is simple enough to do. Perhaps if you also explained the operation of your circuit?

[T3sl4co1l]

Are you sure this is an alternator?  I see no control winding, looks more like a generator to me.  And the circuit is just, what, pulse skipping I guess?

10k isn't enough resistance to keep D7 on, anyway: I_H 5mA max.  You're lucky it's working at all; though I think logic type SCRs tend to hold down to fairly low currents anyway, but nothing that they'll guaranteed on the datasheet unfortunately.

Better to just use a pair of cross-connected NPNs as R-S flip-flop, plenty in the design already, no biggie.  With base bias (dividers), can be made to reset faster and more consistently near zero crossing, anyway.

What phase relationship do the two windings have?  If quadrature, the rectified output will never drop to zero.  If in phase, why not wire them in series (if capable of equal current) or parallel (if equal voltage)?  Depending on inductance, they may never share current this way, so the one is useless -- maybe unless as redundancy, I don't know.

Why the extra (loose) diodes?  They're doing the same thing as the bottom diodes in the FWBs.  I'm not sure that the extra negative rail does anything here.

Also no output indicated, but I'm guessing it's just the top and bottom, far-right corners.  Which, would that go to a battery?  That's a heck of a lot of leakage (the resistor divider(s)) on it, I guess only while the keyswitch is on, but... there are more efficient ways to do this, too (TLVH431, AP431S, etc.).  Well, as far as just getting something together, no worries, it can always be optimized down, something to think about later.

As far as efficiency, would it not be better to replace half the diodes with SCRs (if you want built-in latching behavior) or MOSFETs (sync rect), namely the low side ones, and then, let's see... everything can be common ground, the battery/output side can power the controls during interruptions, and a latch and some gain affords gate drive.  No need for zener clamps on gate signals, assuming the output is adequately limited of course (which, if not, it's not a big deal using them of course).

Let's see... the functionality needed, for pulse skipping as such, is a clocked flip-flop (master-slave type D), with data input from the TL431 (as a comparator, perhaps with a little hysteresis), and this way the output only turns on when requested, synchronized to the start of a cycle, and turns off at the end of one.

But really, it doesn't matter when it turns on or off, as long as EMI and switching loss is respected.  A plain old hysteretic control would seem to suffice, plus a little filtering/snubbing to deal with inductance and switching edges.  No logic required, basically, just a '431 or comparator with a little positive feedback wrapped around it.

Which actually looks a lot like a piece I made a little while ago: https://www.seventransistorlabs.com/Images/ShuntBallast.pdf
That's a good way to apply positive feedback to a TL431.  Now, that's a load, but you need a source, so add an inverter after Q2, and complement the driver.  So:
Move D2+Q4 forward:
Put an NPN under R9: emitter to GND, R9 to base, add 1k from base to GND.
Collector to pullup resistor (4.7k) to gate driver.
Invert the driver by using an NPN follower (C to +V, B to pullup, E to gate) instead for Q4, and flip D2.

And then Q1 and Q3 can be, pretty much the same devices I suppose, but YMMV on ratings of course, and use them instead of the low-side FWB diodes.  Can still use FWBs for the convenience of the high side diodes I guess, but also fine to use e.g. TO-220 common-cathode pairs instead.

Oh yeah, the 9.1V zener there, if the 12V output is reasonably stable and free of transients, then that can be used directly, maybe still with an RC to keep the worst nastiness away from this thing.  24V would be a bit much in my case so I needed to add that.  Something like 15V would be nice to use, in case of transients, so it's not drawing idle current as a shunt regulator, but still knocks down the peaks.  Obviously, reduce R1 to suit; or use a CCS (PNP in your favorite arrangement, or depletion MOS; alternately, can use DMOS+zener as a crude LDO).

HTH,

Tim

[Boschi]

first of all i attached a new schematic, it should be a little cleaner.
secondly, both of you are right, its not an alternator. italian to english translation (and not being formally trained on any of this) doesnt help with the technical jargon. i attached an image of the thing im working with  (the third coil is for ignition and it wont get tuched) , on top of that you have a spining flywheel with the magnets. i attached some images of the output of the coils. im using both coils simply because i can get more power this way, especially at low rpm.

@moffy: the idea here is to use the scr as a "memory cell", so once the voltage on the output caps drops below a certain value, the TL432 triggers the scr (Q2 stops pulling down on the scr gate) and Q3 starts conducting, making Q4 a CC source that charge the mosfets gate, thus charging the output cap through M1 and M2. since the scr keep conducting until the current across it its zero, it means that once turned on it wont stop charging the output capacitor until the coils voltage is low, low enough that the current is basically zero and i dont get any voltage spike out of them.

@T3sl4co1l: you are absolutely right about the holding current of the scr, but since this is a one off i decided to play a dangerous game and i measured the holding current of 5 of them that i had lying around. iirc the were all in the 100 -200 uA range, so i gave it a bit of margin and i used them. they have to hold at least until the voltage drops below 13V, with my resistor configuration there is a current of around 1mA at 11V, so it should be good enough. while tiping this i remembered that this kind of stuff change with temperature... yeah, ill change it in the next revision. you mean something like this? https://en.wikipedia.org/wiki/Flip-flop_%28electronics%29

in regards to the coils... they are in phase with each other, they have the same output power but with different voltages, 6 and 12v. keep in mind that i dont have theyr full caracterization, i say this based on the loads that they were driving. the bike is 40yo, the original wiring was ac and it drived the lights directly (lv coil for the main light at 6v 25W, the hv one that drived the position and stop light (in series) at something like 12v 25W). and i use both to get more power. since i have some losses (diodes and coils resistance) im trying to get as much power as possible, especially since at low rpm the output power drops.

D3 to D6 are necessary since thanks to the mosfet body diodes i have a negative voltage on the mosfet source. if i connect the negative output of D1 and D2 to the GND of my schematic (NOT Vgnd) and my output capacitor is charged i get a constant voltage across the scr, so it will never turn off.

i tought about using sync rectifier, but it wont get me that much more power and (with both coils) ill still have plenty, so the added complexity/BOM wont really make sense to me.
keep in mind that i need something like 1.3A to keep everything on, and the hv coil rectified and dumped in to a car battery  (~13v) gave me something like 1.8A, so with both coil i'll have more than enough.
also, another very important thing for me its the compactness of this thing, im trying to cram everything in a hammond 1590B to save space (its a vespa primavera, its a very compact bike whitout much space to add stuff to) and just the output capacitor banks use about a fourth of the space. so yeah, KISS wins for me in this case.

Let's see... the functionality needed, for pulse skipping as such, is a clocked flip-flop (master-slave type D), with data input from the TL431 (as a comparator, perhaps with a little hysteresis), and this way the output only turns on when requested, synchronized to the start of a cycle, and turns off at the end of one.

But really, it doesn't matter when it turns on or off, as long as EMI and switching loss is respected.  A plain old hysteretic control would seem to suffice, plus a little filtering/snubbing to deal with inductance and switching edges.  No logic required, basically, just a '431 or comparator with a little positive feedback wrapped around it.

Which actually looks a lot like a piece I made a little while ago: https://www.seventransistorlabs.com/Images/ShuntBallast.pdf
That's a good way to apply positive feedback to a TL431.  Now, that's a load, but you need a source, so add an inverter after Q2, and complement the driver.  So:
Move D2+Q4 forward:
Put an NPN under R9: emitter to GND, R9 to base, add 1k from base to GND.
Collector to pullup resistor (4.7k) to gate driver.
Invert the driver by using an NPN follower (C to +V, B to pullup, E to gate) instead for Q4, and flip D2.

And then Q1 and Q3 can be, pretty much the same devices I suppose, but YMMV on ratings of course, and use them instead of the low-side FWB diodes.  Can still use FWBs for the convenience of the high side diodes I guess, but also fine to use e.g. TO-220 common-cathode pairs instead.

Oh yeah, the 9.1V zener there, if the 12V output is reasonably stable and free of transients, then that can be used directly, maybe still with an RC to keep the worst nastiness away from this thing.  24V would be a bit much in my case so I needed to add that.  Something like 15V would be nice to use, in case of transients, so it's not drawing idle current as a shunt regulator, but still knocks down the peaks.  Obviously, reduce R1 to suit; or use a CCS (PNP in your favorite arrangement, or depletion MOS; alternately, can use DMOS+zener as a crude LDO).

im having some truble following you here. i see what you did there with r7 and the histeresis, you are suggesting to throw out the syncornization and just switch on/off as it please. im having trouble drawing your suggestion, but i get your point.

my main problem with this is that im not sure how to solve the voltage spikes, and even worse how high would they be. i currently dont have the tools/knowledge to model the inductors of the stator. i assume that slowing down the mosfet turn off could help with voltage spikes, but the mosfet could get hot fast.

what do you mean by FBS?

[T3sl4co1l]

first of all i attached a new schematic, it should be a little cleaner.
secondly, both of you are right, its not an alternator. italian to english translation (and not being formally trained on any of this) doesnt help with the technical jargon.

Ah, that would do it!

ill change it in the next revision. you mean something like this? https://en.wikipedia.org/wiki/Flip-flop_%28electronics%29

Yes, something like that.  The kind of stuff you're doing already, basically RTL (resistor-transistor logic), no problem, easy to wire up.  You can set a default power-up state by unbalancing the base resistor dividers (= one side turns on first), and a pulse to one side or the other will flip the state.  Can be done with just two transistors and some resistors, or four as a pair of proper RTL NORs, etc.

in regards to the coils... they are in phase with each other, they have the same output power but with different voltages, 6 and 12v. keep in mind that i dont have theyr full caracterization, i say this based on the loads that they were driving. the bike is 40yo, the original wiring was ac and it drived the lights directly (lv coil for the main light at 6v 25W, the hv one that drived the position and stop light (in series) at something like 12v 25W). and i use both to get more power. since i have some losses (diodes and coils resistance) im trying to get as much power as possible, especially since at low rpm the output power drops.

Hmm, I guess because the waveform sucks (non sinusoidal), there's enough peak to still get some power out of the "6V" winding, into 12VDC?  Every little bit I suppose.

Ideally you'd use a boost regulator to help out, or maybe even buck-boost since the max-power condition may be at other voltages (higher or lower) depending on RPM and demand.  And ideally, not just a DC value, but following the waveform itself, to get high power factor as well.

Anything less than that, you're throwing away some capacity.  FWB into a DC rail typically has a power factor around 0.5, or expect about half the ampacity for the same rating.

Expect less still, if the inductance is substantial (i.e. current is limited due to source impedance rather than ampacity).  You'd have to measure the coil inductance to find that out.  Which itself might depend on angle (is that one where there's magnets sticking out of the flywheel, and just air gap inbetween? or is it solid metal inbetween? or solid magnets all around?).

D3 to D6 are necessary since thanks to the mosfet body diodes i have a negative voltage on the mosfet source. if i connect the negative output of D1 and D2 to the GND of my schematic (NOT Vgnd) and my output capacitor is charged i get a constant voltage across the scr, so it will never turn off.

Ah, there is a capacitor now, and again I assume the battery connected on the far right side (a more confident assumption given the labels now).

im having some truble following you here. i see what you did there with r7 and the histeresis, you are suggesting to throw out the syncornization and just switch on/off as it please. im having trouble drawing your suggestion, but i get your point.

my main problem with this is that im not sure how to solve the voltage spikes, and even worse how high would they be. i currently dont have the tools/knowledge to model the inductors of the stator. i assume that slowing down the mosfet turn off could help with voltage spikes, but the mosfet could get hot fast.

what do you mean by FBS?

FBS?

Maybe FWB?  Full Wave Bridge [rectifier].

Spikes aren't a problem if they're snubbed.  At worst, dumping the inductive energy into a TVS diode (see above; the wirewound resistors have a bit of inductance, necessitating TVSs) dissipates that on turn-off.  Hopefully, that energy is a small part of the total, and/or switching is done relatively infrequently (every few cycles, say?), to the same end.  My circuit has enough hysteresis, and supply capacitance, to switch at some 100s to a couple kHz, depending on load.  At much lower currents, and say some 100s to 10s Hz, you won't even need as much capacitance; which you've more or less already hit on, as shown.  So, there you go, heh.

Could also FWB the whole thing into a relatively big cap, and buck regulate down to 12V.  Probably this sucks at high RPM where the peak voltage is massive, but would be doable.  And can be duplicated for each winding, without too much bother, to increase capacity.  SMPS take more work to get right, though, so I'm not necessarily recommending this, just putting it out there.

Tim

[moffy]

Thank you so much for the explanation and the new schematic, it makes everything much clearer. I assume the voltage waveforms that are shown are for the unloaded output of the alternator? If one winding was driving a 6v/25W lamp directly, the waveform should be much altered when loaded. Is there any chance of seeing that? It would give us an idea of the winding inductance which could be very usefull.
Your english is much better than my Italian, which consists in naming some pastas.

[Boschi]

Hmm, I guess because the waveform sucks (non sinusoidal), there's enough peak to still get some power out of the "6V" winding, into 12VDC?  Every little bit I suppose.

Ideally you'd use a boost regulator to help out, or maybe even buck-boost since the max-power condition may be at other voltages (higher or lower) depending on RPM and demand.  And ideally, not just a DC value, but following the waveform itself, to get high power factor as well.

Anything less than that, you're throwing away some capacity.  FWB into a DC rail typically has a power factor around 0.5, or expect about half the ampacity for the same rating.

Expect less still, if the inductance is substantial (i.e. current is limited due to source impedance rather than ampacity).  You'd have to measure the coil inductance to find that out.  Which itself might depend on angle (is that one where there's magnets sticking out of the flywheel, and just air gap inbetween? or is it solid metal inbetween? or solid magnets all around?).

the inductance SHOULD be around 300uH for the 6v coil and 120 uH for the 12v one. i measured it using a somewhat inaccurate tool that i build quite some years ago, but i measured the coil with my tool and i matched the value on the display by using a set of axial inductor of known value that i had lying around. far from perfect but it should at least get us close.

i attached an image of the flywheel. its not exacly like mine but very close.

Spikes aren't a problem if they're snubbed.  At worst, dumping the inductive energy into a TVS diode (see above; the wirewound resistors have a bit of inductance, necessitating TVSs) dissipates that on turn-off.  Hopefully, that energy is a small part of the total, and/or switching is done relatively infrequently (every few cycles, say?), to the same end.  My circuit has enough hysteresis, and supply capacitance, to switch at some 100s to a couple kHz, depending on load.  At much lower currents, and say some 100s to 10s Hz, you won't even need as much capacitance; which you've more or less already hit on, as shown.  So, there you go, heh.

Could also FWB the whole thing into a relatively big cap, and buck regulate down to 12V.  Probably this sucks at high RPM where the peak voltage is massive, but would be doable.  And can be duplicated for each winding, without too much bother, to increase capacity.  SMPS take more work to get right, though, so I'm not necessarily recommending this, just putting it out there.

i tried running some math here (feel free to correct me, im doing a lot of guesstimating) and assuming that only he 12v coil is working (ie the other one is doing negligible stuff), it should be able to pump out around 3A of peak current (its the short circuit current), assuming a inductance of 120uH it means that its storing 0.5*3^2*120*10^-6 J of energy, which is about 0.5mJ. assuming im dumping all that energy in a RC snubber, with a 1uF capacitor charged at 14V i have 0.1mJ of energy, i end up with 0.6mJ stored in the capacitor, which means that it gets charged at a voltage of around 35V, which sound feasible.

assuming on the output caps a load current of 1.5A, assuming that the coil is able to put out around 2Arms of current, assuming ~14mF of output capacitance (the same that i currently have) and assuming that i can tolerate a 0.5Vpp of output ripple, i get 14ms during the charge cycle and around 4.6ms during the discharge cycle. this means that i switch the mosfet off at around 55hz, which means that i should dissipate around 55hz*0.5mJ = 27mJ/s =27mW of continuos power in the damping resistor. plus the apparent current that flow in said resistor.

does any of this make any sense?

i too dislike SMPS, i have basically no experience with them and the huge input voltage range kinda scares me. also, in order to use both coil i'll have to use a voltage inverter, which seems even worse.

i like the idea to have an active discharge of the mos gate, but i'll change the D1 shunt regulator, at 40Vrms it will dissipate a bit too much power for my taste. instead of mos + zener as an ldo, is there any drawback in using a bjt instead? in this way i already have them.

also, the idea is to not have a battery so that it doesnt need any mainteniance (batteries gets old, and here during winter it gets cold enough that if i forget to bring it inside it could get frozen if its not fully charged)

Thank you so much for the explanation and the new schematic, it makes everything much clearer. I assume the voltage waveforms that are shown are for the unloaded output of the alternator? If one winding was driving a 6v/25W lamp directly, the waveform should be much altered when loaded. Is there any chance of seeing that? It would give us an idea of the winding inductance which could be very usefull.
Your english is much better than my Italian, which consists in naming some pastas.

exactly, the coils were unloaded. tomorrow morning ill measure the loaded coils an ill post the screens, i could do it this evening but the neighbours wouldnt be too happy!

[moffy]

I look forward to seeing the loaded waveforms. Since it uses permanent magnets the voltage must increase with RPM ( I know I am stating the obvious) so you might want to check it at different RPMs as well as loaded. I am a little confused that the lower voltage winding, 6V should have more inductance than the higher voltage winding, 12V, I would expect it to be the other way around.

P.S. Are the 6V and 12V coils fully isolated from one another, not the 6V winding just being a tapping of the 12V?

[T3sl4co1l]

the inductance SHOULD be around 300uH for the 6v coil and 120 uH for the 12v one. i measured it using a somewhat inaccurate tool that i build quite some years ago, but i measured the coil with my tool and i matched the value on the display by using a set of axial inductor of known value that i had lying around. far from perfect but it should at least get us close.

i attached an image of the flywheel. its not exacly like mine but very close.

Hm, is that magnets on the inside, and the coils face into there?

I guess it's metal all the way around, so that answers that, too.

The relevance being, if the coil is magnetically open, then its inductance will be higher between magnets; whereas, facing a lump of metal regardless of position, the metal reflects magnetic fields and inductance will be lower.

So it also matters if you measured the inductance in situ or not.  But also, if you measured in free air, we can take that as a worst-case figure.

The other thing is, the coil is likely constructed on laminated iron, which is quite lossy above some 100s of Hz.  Depending on what frequency you measured at, you may find less inductance than is really there*, and more resistance (low Q factor).  It may also depend on signal level.  Your average dollar store inductance tester doesn't tell you about all this, so yeh, could be whatever; but it's in the right ballpark at least, good enough to work off of.

*Compared to, whatever frequency the highest inductance is measured at.  Probably a low frequency, but not so low that wire DCR dominates.  There's a sweet spot between frequency extremes.  Which makes sense: no inductor is suitable for all frequencies, you have to use the stored magnetic energy before it bleeds away in DCR, but also not at such a fast rate that it's largely lost in the laminations and stuff.

i tried running some math here (feel free to correct me, im doing a lot of guesstimating) and assuming that only he 12v coil is working (ie the other one is doing negligible stuff), it should be able to pump out around 3A of peak current (its the short circuit current), assuming a inductance of 120uH it means that its storing 0.5*3^2*120*10^-6 J of energy, which is about 0.5mJ. assuming im dumping all that energy in a RC snubber, with a 1uF capacitor charged at 14V i have 0.1mJ of energy, i end up with 0.6mJ stored in the capacitor, which means that it gets charged at a voltage of around 35V, which sound feasible.

Nah, you're already there in the first part -- don't worry about the cap energy, it just follows along.  We want all the excess (magnetic) energy into the resistor, eventually; the L and C will ring down somewhat in the process, maybe taking some time to do it, but that only depends on the L/R time constant, and the energy balance is true for any R > 0.  For a single event, assuming it's not interrupted with overlapping events of course.  We merely want C large enough that little ringing, and a low enough peak voltage, occurs during this.

The C will take a little extra energy due to the voltage swing from the generated voltage, too, so we want to keep its value small.

Anyway, if we say we're doing 1000 such events per second, even if always at full peak current (somewhat more than 3A, probably?), that's power equals just so many times the energy -- or a half watt, or a few times more at peak current.

So it shouldn't be a big deal!

If we want a peak voltage (in excess of the generated voltage) of say 30V, at max peak current (say 6A), and it's the worst case figure 300uH, and say we switch at 100Hz (but always magically on the generated peaks), then we need to dissipate 0.5 (6A)^2 (300uH) (100Hz) = 0.54W, and the resistor will be (30V) / (6A) = 5 ohms.  4.7 is a common value, and maybe make it a metal-oxide 2W power resistor just in case.

To avoid ringing, we need C > L/R^2, or 13.6uF.  Pretty large, really; the resonant frequency is 2.5kHz.  We might tolerate a higher peak voltage, or use a TVS clamp instead (rated to avoid conduction at max RPM -- need peak voltage when you can get around to this).  We can also put this RC damping after the FWB, so it's not exposed to raw AC (which would dissipate a lot of power), and any excess charge just gets dumped into the load/bulk cap/battery the next time it turns on again.

And obviously, MOSFET needs to be rated for at least as much peak voltage as the TVS.

For best EMI, I wonder if maybe, like, 100nF + 22R (in series), from GND to each winding, and then a bigger one (but not the full 13uF, maybe just 1-4.7uF to help "take the edge off"?) and a TVS, on the rectified side.  Reason for the winding side is, the diodes themselves have some switching noise, so this just helps things stay a little quieter.

Speaking of EMI, maybe a "hash choke" on the output terminal could be helpful, guessing... well wait a minute, what is ignition on this thing anyway, if magneto it's probably not anything huh?  (Just keep wires away from the ignition cable itself!)  And I suppose if battery can be optional, you definitely won't have electronic ignition (even points!), heh...

...So yeah, that should cover for any source of interference you might need to worry about around this circuit.  I would suggest building it on perf board, mounted to the lid, using mounting screws for grounding.  If the lid seals to the box with a gasket, you may want a ground lug between them as well, just in case.  And the box or lid can be mounted in turn to the chassis, which will be common ground as usual, I'm guessing?  Keep the connections (especially EMI components) close to a ground connection; so, the R+Cs and the output bulk cap, basically.

If you want to get fancy, you can even mount the transistor, and resistor (get a TO-220 style one instead), to the enclosure -- another good reason to choose the lid, they can mount there while also soldering directly into the circuit board, without hassling with fly leads or awkward hardware.

(Note that you'll want to drill access holes through the board, to reach the hardware underneath!)

assuming on the output caps a load current of 1.5A, assuming that the coil is able to put out around 2Arms of current, assuming ~14mF of output capacitance (the same that i currently have) and assuming that i can tolerate a 0.5Vpp of output ripple, i get 14ms during the charge cycle and around 4.6ms during the discharge cycle. this means that i switch the mosfet off at around 55hz, which means that i should dissipate around 55hz*0.5mJ = 27mJ/s =27mW of continuos power in the damping resistor. plus the apparent current that flow in said resistor.

does any of this make any sense?

Careful -- RMS is not DC average!  Especially with a peaky waveform like that, it will be even higher (than the DC output).

Typical case for AC mains is input Irms being a bit over double the Idc.  (This is somewhat balanced by Vout also being 40% higher than Vrms -- so it definitely can't be equal currents, else you'd be getting extra power from somewhere!)  Which is to say, a power factor around 0.5.

Your PF might be closer to, Idunno, 0.3 or something?  Depends on RPM too, since you're clipping a ton of extra peak voltage at high RPM, or barely cutting through the peaks at low RPM.  So the conduction angle varies, and how much of the supplied voltage you're wasting, varies.

But yeah, your ripple calculation is correct, just needs the right inputs.  Exactly what the currents are, you'll have to figure out in testing, but if it's able to maintain output at all, it'll be somewhere around that time constant, and...some duty cycle (what duty, doesn't really matter!).

Like I said, you could get away with much less capacitance.  But, I think, not a whole lot less than I recommended; 1000uF for example, you'd be hard pressed to find enough ripple current ratings, especially if this thing is going to be block mounted... erm, well, nevermind that heh, it's probably air cooled right, searing hot?  Yeah, just find somewhere nice to stuff it, preferably away from heat...  But anyway, a couple of those in parallel, or most 2200 or 4700uF units, I'd think, will handle that ripple current fine.  With 14mF in there you should have no problem at all.  Can also set the threshold a bit narrower, perhaps, raise the frequency, reduce ripple further.

Tim

[moffy]

I have been playing around with an LTSpice model, based around your circuit but eliminating the SCR, which might not be necessary. T3sl4col1 has been emphasing the need for snubbing, because of the inductive energy, and is of course correct. Some snubbing as well as storage cap after the bridge rectifier has been added, mainly because without the SCR it doesn't switch at zero crossings, so inductive spikes will be larger. Please excuse component types as they are what I readily had in LTSpice, it is for illustration purposes.
If the voltage out of the alternator drops to 30v p-p the TL431 and MOSFETs go into a linear mode, not enough voltage for switching.

[TheMG]

This seems like a fairly atypical approach. Most small vehicles (motorbikes, mopeds, ATVs, snowmobiles, etc) typically have a permanent-magnet alternator. Usually the regulators in such vehicles are shunt regulators, not series. Essentially limiting the peak voltage of each pulse of unsmoothed DC by shorting the stator coil when a threshold is reached. It's been done that way for ages.

More modern units will rectify and smooth the output to create an unregulated DC bus (usually at a fairly high voltage) and follow that with a buck DC-DC regulator.

[Boschi]

I look forward to seeing the loaded waveforms. Since it uses permanent magnets the voltage must increase with RPM ( I know I am stating the obvious) so you might want to check it at different RPMs as well as loaded. I am a little confused that the lower voltage winding, 6V should have more inductance than the higher voltage winding, 12V, I would expect it to be the other way around.

P.S. Are the 6V and 12V coils fully isolated from one another, not the 6V winding just being a tapping of the 12V?

i attached them. if you multiply the frequency by 20 you obtain the rpm (just to have an idea). the first two screens are from the 6v coil with a 6v 25w lamp as a load, while the second two screens are from the 12v coil with a 12v 21w lamp attached. also to just give an idea, the 6v coil has a short circuit current of 8A peak and 5Arms, while the 12v one 1.7Arms and 3A peak. i also measured again the inductances, they are correct, by spinning the magnets i can get the 6v one up to 300uh, while the 12v up to 150uh. (by spinning i mean turning a few degree between each measurement).

the coils are two separate coils, and i have access to all the 4 wire out of them, they are not internally connected in series. the only thing is that one of the terminal of the 6v coil is connected to the body of the bike, but it shouldnt matter here and if necessary i could disconnect it.

Hm, is that magnets on the inside, and the coils face into there?

I guess it's metal all the way around, so that answers that, too.

The relevance being, if the coil is magnetically open, then its inductance will be higher between magnets; whereas, facing a lump of metal regardless of position, the metal reflects magnetic fields and inductance will be lower.

So it also matters if you measured the inductance in situ or not.  But also, if you measured in free air, we can take that as a worst-case figure.

The other thing is, the coil is likely constructed on laminated iron, which is quite lossy above some 100s of Hz.  Depending on what frequency you measured at, you may find less inductance than is really there*, and more resistance (low Q factor).  It may also depend on signal level.  Your average dollar store inductance tester doesn't tell you about all this, so yeh, could be whatever; but it's in the right ballpark at least, good enough to work off of.

*Compared to, whatever frequency the highest inductance is measured at.  Probably a low frequency, but not so low that wire DCR dominates.  There's a sweet spot between frequency extremes.  Which makes sense: no inductor is suitable for all frequencies, you have to use the stored magnetic energy before it bleeds away in DCR, but also not at such a fast rate that it's largely lost in the laminations and stuff.

 you are correct about the magnets and the coils! i measured them in place, if necessary i can disassemble the flywheel and measure them again, but i prefer to avoid it, setting up the phase of the ignition its a bit of a pain.
the coils are made out of laminated iron, and my tool is a colpitts oscillator that measure the frequence difference to know the inductance of the measured coil. i measured a frequency around 500Khz with a inductance of 220uH, so yeah, we are way above the 100s of hz.

If we want a peak voltage (in excess of the generated voltage) of say 30V, at max peak current (say 6A), and it's the worst case figure 300uH, and say we switch at 100Hz (but always magically on the generated peaks), then we need to dissipate 0.5 (6A)^2 (300uH) (100Hz) = 0.54W, and the resistor will be (30V) / (6A) = 5 ohms.  4.7 is a common value, and maybe make it a metal-oxide 2W power resistor just in case.

To avoid ringing, we need C > L/R^2, or 13.6uF.  Pretty large, really; the resonant frequency is 2.5kHz.  We might tolerate a higher peak voltage, or use a TVS clamp instead (rated to avoid conduction at max RPM -- need peak voltage when you can get around to this).  We can also put this RC damping after the FWB, so it's not exposed to raw AC (which would dissipate a lot of power), and any excess charge just gets dumped into the load/bulk cap/battery the next time it turns on again.

well, all my strange math with the cap energy was mainly to get an idea of the cap rated voltage / snubber capacitance / resistor power rating, the fact that somehow made sense means that i can count that as a win for me!
where does your capacitance formula came from? also, i like the idea of the capacitor after the FWR, mainly to reduce apparent current (and real power disspated on the snubber resistor).

...

Speaking of EMI, maybe a "hash choke" on the output terminal could be helpful, guessing... well wait a minute, what is ignition on this thing anyway, if magneto it's probably not anything huh?  (Just keep wires away from the ignition cable itself!)  And I suppose if battery can be optional, you definitely won't have electronic ignition (even points!), heh...

i have points! only the best of the 70s italian ignition tecnology went into this bike! its a lc resonator, the points open at a specific time and the third coil in the stator charge a 330nF cap, that somehow at a specific time start resonating with a hv trasformer that drive the spark plug. so yeah, we dont have any concern here for EMI, the ignition will be already blasting a lot of them around.

also, ideally ill try to keep everything insulated from the "real" ground of the bike, mainly since by doing so i dont have to make any invasive operation to the stator.

@moffy: i didnt simulate the circuit (yet), but are you sure R12 and C3 are necessary? also, ill probably add a resistor in series to C4, otherwise it could fry the mosfet if its low esr and the mos turn on is fast enough. tomorrow ill take a better look at it.

This seems like a fairly atypical approach. Most small vehicles (motorbikes, mopeds, ATVs, snowmobiles, etc) typically have a permanent-magnet alternator. Usually the regulators in such vehicles are shunt regulators, not series. Essentially limiting the peak voltage of each pulse of unsmoothed DC by shorting the stator coil when a threshold is reached. It's been done that way for ages.

More modern units will rectify and smooth the output to create an unregulated DC bus (usually at a fairly high voltage) and follow that with a buck DC-DC regulator.

you are absolutely right, but since this is a retrofitting i dont have much space/power to play with. since this thing will probably mounted inside the bike (and not near the engine) for space reason (there is not much space to begin with, but ill probably change the engine one day or another for a bit bigger one), i have to try to keep the power dissipation to a minimium, and doing a "normal" scr shunt regulator will probably dissipate something close to 7A rms at 1.5Vrms. i mean, 10w its still not much, but lower is better and for once doing a "complicated" real word design is fun for me. im studying computer engineering, so once in a while i love to flex my muscles a little and do some electronic stuff.

[moffy]

Quote: "the coils are made out of laminated iron, and my tool is a colpitts oscillator that measure the frequence difference to know the inductance of the measured coil. i measured a frequency around 500Khz with a inductance of 220uH, so yeah, we are way above the 100s of hz"

For laminated iron core 500kHz is way too high, you should measure the inductance at the frequency it is being used, even a 50Hz measurment would be fine.
The waveform pics are excellent! It looks like you are getting similar voltages out of the 6V and 12V windings with their respective loads. So the RPM was roughly 6000RPM and 12000RPM (aproximately 300Hz and 600Hz), revs pretty high.

[moffy]

Quote: "after pulling my hair out for a while i realized that the problem occurs with just one of the coils, the "high" voltage one (it should be somewhere in the 12v 25VA region), but not with the "low" voltage one (6v 25VA). after probing around a bit i realized that the rectified voltage across the pins 1 and 4 of the gbu1510 it drops to 0 for a very short period of around 5us before rising again to a higher voltage so the scr cant turn itself off.  the hv coil has sharper rising/falling edge than the lv one, so its not that strange."

Now that my understanding has caught up somewhat with your circuit and explanation  , there might be a simple solution. If you were to place a low value zener diode or even 2 or 3 ordinary diodes in series with the SCR you would effectively increase the deadband for the SCR to turn off. It would help guarantee that the SCR turns off each half cycle, at the low voltage point where charging is not going to happen anyway. I would also lower the resistances around the SCR gate as they are succeptible to being triggered by fast voltages across the A-K.

Also if you don't have LTSpice, download and play with it, it can help tune component values and also uncover some unexpected behaviours.

P.S. You have to get marks for how complete and responsive your answers have been.

[moffy]

A rough calculation of inductance, using the open circuit voltage, short circuit current and waveform frequency gives:

12V coil: 120V p-p @ 350Hz and 6A p-p short circuit current gives:  L12v = 60/(3*(2*PI*350)) = 9mH
6V coil: 60V p-p @ 350Hz and 16A p-p short circuit current gives:  L6v = 30/(8*(2*PI*350)) = 1.7mH

It is a very rough calculation, but the expected value of the 12V coil should be roughly 4 times and this gives 5, the value of the 6V coils inductance because on the same magnetic path inductance is proportional to turns squared. It is very rough but should give a ballpark figure. The actual inductance should be smaller because of the less than sinusoidal shape of the open circuit waveform, maybe 20-30 percent? but that is guess work.

[Boschi]

...
Now that my understanding has caught up somewhat with your circuit and explanation  , there might be a simple solution. If you were to place a low value zener diode or even 2 or 3 ordinary diodes in series with the SCR you would effectively increase the deadband for the SCR to turn off. It would help guarantee that the SCR turns off each half cycle, at the low voltage point where charging is not going to happen anyway. I would also lower the resistances around the SCR gate as they are succeptible to being triggered by fast voltages across the A-K.

the problem is that the voltage stay constant at 13v and then falls abruptly for 5us, since it doesnt have a slope i dont think placing diodes would work. im thinking about adding a capacitor on the coils terminal, that should add a little load, hopefully enough to cause a "real" zero crossing.

also, dont trust the counter in the top right corner, its not correct, in the screens the waveform was at 100 and 200 hz (2K rpm and 4K rpm).

about the inductance, yesterday i tried to measure it by putting 2 15v transformers in series and feeding the voltage to the coils while measuring voltage and current. i assume the combined impedance of the trasformers was a bit too high or whatnot, but i got out that the 12v coil has an inductance of 27 mH (which i assume is wrong).

i was tring to come up with a simple (and without any special tools) way to measure the inductance, since i have a scope and various resistors is there any obscure problem in measuring the inductance using it's time constant?

i gave it a go and i attached the screens. 13 and 14 are for the 6v coil (total resistance: 25,4 ohm), 23-24 for the 12v one (total resistance 26.8 ohm).
the cursor its at the 63% of the steady state voltage, so at exacly 1 time constant, after running the mat it came out that the first coil (6v) has an inductance of 800uH, while the second (12v) 11.5mH.

im kinda confused? does this metod make sense or im just getting garbage data? the steady state current its around 1A, so the core wasnt saturating i think.

[T3sl4co1l]

Show the circuit?

Tim

[moffy]

You see the calculations I did for the inductance? Lcoil = Voc/(2*PI*f*Isc) where Voc=Open circuit voltage at 'f', and Isc=Short circuit current at 'f' and f=frequency the measurements are made at. It assumes that the major impedance at f is inductive and that the resistance is negligable by comparison.
Your method isn't bad trying to use L/R but the results don't seem reasonable, and that might be because with a fast risetime the laminated core ceases to behave properly. I am talking more about the 6V coil measurement.

[Boschi]

Show the circuit?

Tim

the one attached.

Your method isn't bad trying to use L/R but the results don't seem reasonable, and that might be because with a fast risetime the laminated core ceases to behave properly. I am talking more about the 6V coil measurement.

i see. well, ill just use your values and then once i have a somewhat working prototype ill tune the snubber cap.

im sorry to you all, but for now i have to put this project on hold. im currently away form home and wont be able to pick up a soldering iron till september. hopefully ill still be able to try some simulations. until then, thanks to everyone! 

[T3sl4co1l]

Testing with lower voltage and resistance may prove helpful, slow down the time constant to get a better idea of the low frequency value.

Tim

[moffy]

An interesting little fact(I hope)/prediction about your alternators short circuit current is that it should remain relatively constant with frequency, assuming inductance is the limiting impedance. Since the output voltage is proportional to frequency and the impedence presented by inductance is also proportional to frequency the frequency dependence cancels one another out providing a relatively constant short circuit current.