High input voltage (flyback) converter - any ideas for the transformer?

[Kremmen]

I have been tasked to design an AC/DC converter for harvesting power for a telemetry device from the supply voltage of an associated induction motor.
What makes the routine design a bit challenging is that the input voltage will be at least 400VAC and up to ~600VAC. There is no neutral line as that is not commonly wired to an induction motor.

Everything else is more or less business as usual, but my go to distributors, i.e. Mouser, Digikey, Farnell are a bit helpless in the transformer department. Nobody seems to stock anything usable and since this is a fully commercial product, DIY is out of the question due to safety of life and the associated approvals.

Basically what i need is a simple flyback transformer with turns ratio appropriate for 400-600 VAC primary, roughly 24V secondary (this can be tuned in many ways so not critical, and a primary side feedback coil. The key is sufficient isolation between pri and sec; at least 2.5kV but preferably more. Power requirements are modest, just a few VA to run an ARM MCU, a small battery charger and a couple of simplish analog front end circuits.
Because i need to build a number of prototypes before committing to production, it is not possible to start with an order of 10k items. So i need to find a small volume option for the proto phase.

I don't expect you to do the design for me, but any suggestions where to even look for the above will be appreciated. Meanwhile, the search continues...

P.S. Basically i don't care how exactly the power is harvested. If there are better ideas than a traditional flyback, i will listen

[jbb]

Sounds like a job for a custom transformer. Or you could try deploying two transformers with the primaries in series (forces current sharing) and the secondaries in parallel (forces voltage sharing between two primaries).

If you need isolation, the flyback topology is a favourite for a reason. They work over a wide input range and don’t need a lot of components. What are you using for the primary switch?

[chris_leyson]

Try Midcom, now owned by Wurth, they will design a custom flyback transformer for you and you will get samples in a week or so. However, given the 600VAC input voltage requirement they won't be cheap, not even in 10k quantities.

[Zero999]

A flyback transformer is not suitable for this task.

Is the motor running off a variable frequency drive?

If not, just use an ordinary control panel transformer with a 600V primary and 24VAC secondary.

See the link below for an example. You want 9070T50D166, page 23.
http://www2.schneider-electric.com/resources/sites/SCHNEIDER_ELECTRIC/content/live/FAQS/268000/FA268443/en_US/9070CT9901.pdf

If it is, then it's a bit more tricky. No doubt it's possible to buy a switched mode PSU which can do this but you probably need one with power factor correction, otherwise the VFD won't be happy.

[T3sl4co1l]

Doing it direct sounds suboptimal.  The switch will have an impedance of, let's see...

Say you're doing max 5W, max 900VDC supply, lowish duty cycle say 25% so the peak voltage is only 1200V (plus leakage inductance peak, plus high-line and transient conditions -- use a >= 1600V transistor here).  5W at 900VDC is 5.6mA average, or 44.4mA peak at turn-off.  The switching impedance is therefore 1200V / 0.0444A = 27kohms.  Very high indeed!

If it were closer to 100 ohms, it would be easy enough to deal with the parasitics -- but with Zsw so high, your problem will be all the capacitances of the switch and transformer winding.  Expect terrible efficiency!  A ZVS design would be helpful here, but that might not be practical either (cost and size??).

Instead, a quicker start is this: throw a conventional iron-cored transformer in front of it.  Reduce it to 240VAC max, and use an off the shelf supply.  Bulky, kind of expensive (medium voltage transformers aren't a common item), but it's The Right Way to do it.

Tim

[jbb]

Is the motor running off a variable frequency drive?

Good point!  Don't try to run something like this off a Variable Frequency Drive (VFD) output.  Their outputs come straight from large power transistors and have very high dV/dt rates.  This will result in huge current spikes into your DC link filter cap.

Also it's nearly impossible to measure power on a VFD output, because the high frequency content of the voltage waveform (i.e. sharp square waves) is very hard to sample properly.

Edit: ah hah! I knew I'd seen an add recently!  Mornsun makes wide input range AC-DC converter modules.
3W encapsulated, safety certs pending...
3W open, needs external DC link caps, has safety certs
5W open, needs external DC link caps, has safety certs

Remember to check your peak power consumption for stuff like radio modules. (Especially cellular modems with 2G fallback. Those draw huge peaks.)

[Zero999]

How about the Eaton PSG60F24RM?
320V to 600VAC in, 24V out, 60W and is available from Farnell.
http://uk.farnell.com/eaton/psg60f24rm/power-supply-ac-dc-24v-2-5a/dp/2546047
http://www.farnell.com/datasheets/2309125.pdf?_ga=2.56380331.1664337938.1526987896-1612659360.1413284049

Is the motor running off a variable frequency drive?

Good point!  Don't try to run something like this off a Variable Frequency Drive (VFD) output.  Their outputs come straight from large power transistors and have very high dV/dt rates.  This will result in huge current spikes into your DC link filter cap.

If it has a power factor corrected input stage, then you stand more of a chance. A big inductive filter will also help, but may prove bulky and expensive.

[Kremmen]

Thanks for the very good responses. Some issues i did think of already, others give food for thought.

The 2/3G radio for instance will have a Li-ion battery to supply the needed transmission current. This is a familiar issue from earlier designs.
I had also thought of just using an ordinary iron transformer but not as a drop-down front end for a regular AC/DC module. That is worth looking into.
What i neglected to say earlier is that the PSU does not have to supply the entire load all of the time (the battery is there to help). In some cases the motor is DOL, sometimes fed from a VFD and then there will be an undervoltage cut-off as well as a filter to stop the worst noise from getting through.

Some responses to specific comments:
I have already started queries for manufacturing custom flyback transformers. We'll see.
The Eaton power supply is technically right on, but infeasible as far as size and price are concerned. That would be the way to go but alas, no. It has far more wattage than required for this application and that drags the price to an undoable level.
The DC caps need protection against inrush current, a good point. Possibly a simple PFC or even an inductor could suffice (because low output power).
Measuring the voltage of a VFD output? Impossible - no. Difficult-ish, yes. Not a trivial task by any means but for that i have the help of a platoon of PhD's. That part will be done. OTOH, those guys are helpless in designing the power supply part...
At this time i am not too worried about the happiness or otherwise, of the VFD. The motors are relatively big (at least several kVA up to 100s of kVA). The VFD will hardly notice a couple of VA whatever the PF.

The Mornsun modules are real interesting. Actually, just the thing i was looking for. Now to find a distributor... Any info on that?

Next: Once this is done, there is a line of 6 kV motors waiting in line for similar add-ons. Now that will be a challenge. Maybe i need to reinstate my IEEE membership to get access to some relevant papers...

Thanks for the replies, all of them were valuable.

[jbb]

How about an alternative plan: steal some power off the current sense transformers?  If you don't need to use the radio often, you can design your measurement equipment to have low operating current and provide just a bit of charge current to the LiPo.

What i neglected to say earlier is that the PSU does not have to supply the entire load all of the time (the battery is there to help). In some cases the motor is DOL, sometimes fed from a VFD and then there will be an undervoltage cut-off as well as a filter to stop the worst noise from getting through.
...
At this time i am not too worried about the happiness or otherwise, of the VFD. The motors are relatively big (at least several kVA up to 100s of kVA). The VFD will hardly notice a couple of VA whatever the PF.

It is certainly possible to do this, but there are a number of issues you'll need to look at.

Undervoltage cutoff doesn't mean anything in the context of a VFD.  If the VFD is operating at all, it is spitting out square waves* of the DC link voltage (approx. 600V).  To get a lower equivalent voltage, the pulses just get narrower.

If we look at the data sheet for the 3W open frame module, the recommended deployment for EMI compatibility (Figure 2) has a 0.1uF X cap and 470pF Y caps.  When subject to (round numbers) a 10 kV / us output from the VFD, the X cap would try to pass 1000A.  The common mode choke will of course restrict this, but you could be looking at some ugly current stresses.

These current spikes could be measured by the VFD and make it a) do slightly strange control stuff or b) trip into fault protection.

There will also be high frequency current flow through the Y caps. The common mode choke will try to stop the current, but that means it will have a lot of flux going on (and at the switching frequency of the drive, not just 50Hz) and might get very hot.

In addition to X and Y cap issues, VFDs with longer cables are known for having reflected voltage wave issues.  That is to say, the VFD generates sharp square-wave pulses which go down the cables like a transmission line.  When they reach the terminals of the motor, there is an impedance mismatch and some of that wave reflects back.  Depending on where you measurement box is mounted, it could be subjected to a 600V pulse (the DC link voltage of the VFD) followed by a step up to a higher level (maybe 1.5X or even 2X if the motor is very small).  These voltage levels will want to rush into your device and may blow up your DC link.  You can add a filter inductor to take  the edges of, but then you have to check for possible resonances (note VFDs operate over a wide variety of switching frequencies from 400 Hz to 8 kHz or more).

Measuring the voltage of a VFD output? Impossible - no. Difficult-ish, yes. Not a trivial task by any means but for that i have the help of a platoon of PhD's. That part will be done.

Ah, that will help.  Also, have you considered the possibility of a) getting information out of the VFD via Modbus or similar or b) measuring the line side of the VFD?

The Mornsun modules are real interesting. Actually, just the thing i was looking for. Now to find a distributor... Any info on that?

Mornsun offer some kind of sample request on their website.  And they have a list of European distributors.  I haven't used them before, but their datasheets look like they give a damn.

Once this is done, there is a line of 6 kV motors waiting in line for similar add-ons. Now that will be a challenge. Maybe i need to reinstate my IEEE membership to get access to some relevant papers...

!!!
That will be interesting.  Please be aware that once you get into medium voltage (>= 1kV AC), a lot of things change: insulation doesn't insulate (creepage, ageing, partial discharge), applicable standards change, electrical safety rules change, over voltage classes become a thing...

* Yes, OK, there are multilevel drives out there which spit out more interesting waveforms than straight-up square waves.  But they are unusual in Low Voltage (LV) (< 1kV AC) settings.  We may see more 3 level drives in the LV market soon because the cost of semiconductors is dropping and 3 level drive allows for improved efficiency (reduced iron losses in motor).

[Zero999]

I agree that taking some power from the current transformer is a possibility, but you dont' want to affect the measurement too much.

I've just found some application notes, which might be helpful.
http://www.ti.com/lit/ug/tidu412a/tidu412a.pdf
https://www.monolithicpower.com/pub/media/document/AN084_r1.0.pdf

[Marco]

What's wrong with say the WE 7508110151?

[Kremmen]

The CT option is interesting. In principle it can be done because none of the measurements need to be absolutely continuous. It would be possible to measure current say 10% of the time and use the rest to harvest energy. I need to study the numbers a bit to see if this is feasible in practice since the CT output is scaled to the ADC, i.e. it will be <3.3Vp under all conditions. I hesitate to plan disconnecting the burden resistor to get more voltage because any mishaps there will result in fireworks. However it might be possible to use an extra series burden resistor that is shorted for measurement and there for energy harvesting. I will have to look into the papers Hero999 linked - thanks.

Modbus was mentioned and yes, that will be used in all cases where it is available. Not in this particular case however, because of reasons i hesitate to be too open about. Suffice it to say that modbus will not reach the site where this particular unit is installed.

VFDs are a familiar beast. For background, i used to work at Stromberg ( now part of ABB ) that had and still has a world class power electronics design division. The daddy of all modern VFDs, original SAMI (Stromberg Asynchronous Motor Inverter) was developed there and quickly pushed DC drives out of the market. My last jobs for the company were supervising the commissioning of AC drives for 2 big Valmet paper machines delivered to UK and USA and equipped with Stromberg VFD induction motor drives. It couldn't be done they said but they were wrong . Today it is all routine of course. ( A brief glipse into the origins of VFDs for those who really want to know: The audio is Finnish but there are English subtitles).

So.
While the VFD output voltage is a constantly varying PWM modulation, it is possible to detect the RMS value by several means. One straightforward way is to massively oversample the waveform with say 1 MHz sample rate, possibly decimate it in real time a bit and just detect the pwm edges and zero crossings. There will be enough computation power to calculate the RMS voltage from that. Additionally, an analog LP filter front end could take out the worst spikiness. Or something else. That is all up to the design cycle of the actual measurements.

Voltage excursions due to cable reflection need to be limited of course. Plan A is to use inductors followed by limiting components such as TVS diodes or suchlike. But we will see once the design goes forward.
Regarding X and Y caps - the installation circumstances are a bit special (for starters, there is no neutral anywhere so no X caps from L to N). I'll have to see what if anything we want to use there and if the issues mentioned then raise their head.

Again, thanks for the points and suggestions. And i managed to find a local distributor for the Mornsun product line so that one will progress and could prove to be the basis for the supply solution. Special thanks for pointing that one out!

re the 6 kV motors - tell me about it! That will be a whole different ballpark to play in...

[T3sl4co1l]

No need to oversample, RMS is not a reconstruction process and therefore the Nyquist sampling theorem does not apply.

Riddles: what sample rate do you need to measure the RMS value of a random noise signal?  What is the expected value of a single sample?

Tim

[jbb]

Glad to hear you have some VFD experience.

While the VFD output voltage is a constantly varying PWM modulation, it is possible to detect the RMS value by several means. One straightforward way is to massively oversample the waveform with say 1 MHz sample rate, possibly decimate it in real time a bit and just detect the pwm edges and zero crossings. There will be enough computation power to calculate the RMS voltage from that. Additionally, an analog LP filter front end could take out the worst spikiness. Or something else. That is all up to the design cycle of the actual measurements.

You can sic some PhDs on it.  Please note that measuring U RMS and I RMS and multiplying probably isn't the best way to measure true average power.

Also, for extra value to the customer, you should think about future options to get more value by condition monitoring: phase balance / phase sequence, power ripple, estimated torque ripple and general power quality events (dip, sag etc.). Maybe even characterise motor starts (for DOL motors) to look for problems in the rotor.

[Kremmen]

No need to oversample, RMS is not a reconstruction process and therefore the Nyquist sampling theorem does not apply.

Riddles: what sample rate do you need to measure the RMS value of a random noise signal?  What is the expected value of a single sample?

Tim

OK, possibly not. But as i noted, that was just the first thing that came to mind without really thinking carefully about that part. But i need to detect the pwm pulse distribution somehow to reconstruct the RMS value. If i manage to reliably detect the rising and falling edges that will do it, but remains to be seen if it can be made to work well enough. One challenge is that you cannot assume anything about the VFD (if there is one). That device is not part of the managed package whereas the motor is.

True random noise? RMS is 0 if it is unbiased . No need to sample to find the RMS. If you wish to know something else though...

[Kremmen]

...
You can sic some PhDs on it.  Please note that measuring U RMS and I RMS and multiplying probably isn't the best way to measure true average power.

Also, for extra value to the customer, you should think about future options to get more value by condition monitoring: phase balance / phase sequence, power ripple, estimated torque ripple and general power quality events (dip, sag etc.). Maybe even characterise motor starts (for DOL motors) to look for problems in the rotor.

Actually just multiplying the RMS values will disregard the cosine phi and produce an erroneous result so that will not work. Fortunately that is familiar ground and additionally i have the help of a motor specialist PhD for modeling the state and operating point of the motor - that being a core part of the whole effort. Actually, the motor current will be analyzed for a massive number of factors relating to the motor itself, and the load it is driving. But at this point i seem to hit an NDA so i'll leave it at that.

Mornsun power modules on the way  Waiting with interest.

[PartialDischarge]

Go with the morson psu. Relatively cheap and isolated but you will need the 1500V rated one they have ( and also add the external rectifiers).

About the 6kV motor this is a totally different play. Anything connected to a Medium Voltage like this needs to be tested for continous 7.2kV and a 60kV lighting impulse. Currently intrument transformers are used for this task, but think BIG, 20-25kg in weight and 500€ or more each, and this for one phase only. I can however offer you another solution, ill PM you.

[T3sl4co1l]

OK, possibly not. But as i noted, that was just the first thing that came to mind without really thinking carefully about that part. But i need to detect the pwm pulse distribution somehow to reconstruct the RMS value. If i manage to reliably detect the rising and falling edges that will do it, but remains to be seen if it can be made to work well enough. One challenge is that you cannot assume anything about the VFD (if there is one). That device is not part of the managed package whereas the motor is.

RMS current will be more important, presumably, as it's filtered by winding inductance.  In any case, easy enough to sample voltage at the same time, do the product, and get the correct result (P = avg(V*I)).  Doesn't need much [sampling] bandwidth either.

True random noise? RMS is 0 if it is unbiased . No need to sample to find the RMS. If you wish to know something else though...

I'm glad I asked!  Start with normal distribution, sigma = 1, mu = 0.  Show proof.

Tim