Wind mppt attempt

[Janne]

Hello all,

With the snow piling up outside, and hampering other projects I've been drawn back to the electronics lab. I've had this idea on the back of my head for a while, but only now I've been getting started with it. So basically, it's a buck topology switching mode converter, that will take the input from the 3.2m axial flux generator, and convert it to the 12V battery bank. This should make the load maching much better, and increase energy yield especially in the average wind region. Currently the machine is, like the usual way, connected directly to the battery after the rectifier, which makes the loading curve awful.

The problem with the buck topology, that I've went with is, that the full power of the turbine always has to pass through the converter. The easier option controller wise would be to make the stator for a high cut in speed, and instead make the mppt contoller with boost topology, so it would boost the voltage proportional to the optimal loading curve, before the "cut in" point. After the cut in point, the current from the turbine would go directly to the battery, bypassing the converter. I think some forum members here have actually tried and succeeded in that arrangement, but can't remember at the moment who they were.

Anyways, as I wanted to avoid making a new stator, and as the furling currently is set at a quite low current (35-40A peak, continous current less than that) I went ahead with a buck style converter.

At first i started by designing the components for the power swither part. At the start i was pretty clueless, but it turned out once again in the internet there was more than enought resources available to design the components of the switcher. For the switching frequency I settled for a reasonable 100kHz, that i think was a good balance between passive component sizing & layout problems / switching losses. The attached schetcing shows the basic blocks of the converter, and the power components of the business end;

 

So, the system works by reading the speed from the turbine frequency (detail of this part is still unknown, it is one of the will be solved once we get there issues ), and then comparing the speed to a value in a lookup table loaded to the memory of the controller. The value in the lookup table will represent a power value for the current speed, and from that, based on the battery voltage, the desired output current is then calculated. The desired current is then, along with the current feedback, is then fed into the tl494 pwm controller, which in turn will control the pulse width going to the switching transistors. The fet gates are driven with an isolating gate transformer, which should provide some noise rejection to the contoller board, i hope.

Currently the power part of the controller is quite finished, but i've yet to test it at full load, so far I've only teased it at mild 10A loads... For the full load test i think i have to go with the mig welder .  The controller board is also quite nice in progress, but the speed measurement part is still totally untested, and the tl494 + it's accessories need to be tested further for stability. But if all goes well, i think it will be ready for real life testing once the snow will start to thaw away =). And even if it turns out to be dud in the end, at least i had fun while playing with it.

[scottsAI]

Janne,

You're on the right track, RPM to load is great way to control.

Buck controller required to handle low wattage to high watt may be difficult the way its shown.

One inductor and MosFET can handle 50w to 500w (not exactly easy)

Adding slightly more active control. Using several inductors and MosFETS may simplify the controller allowing wider power range to be controlled.

Have fun,

Scott.

[altosack]

Hello Janne,

You state that with a buck converter that all of the current has to pass through the primary switch. Well, I think there are two scenarios that make this a bit less relevant.

The first is that, while the current may not be passing through the primary switch in a boost converter, it is passing through a diode, and if everything is well-designed, the MOSFET shouldn't be less reliable than a diode if both are loaded similarly relative to their maximum ratings.

I believe that it is desirable to use (2) MOSFETs instead of a MOSFET and a diode (in either a buck or boost converter) to increase the efficiency (i.e., a synchronous converter). In this case, there is no advantage to the boost converter as far as power handling is concerned.

The second is that often by the time you have higher winds, your batteries are already full from the build-up, and you never really need full power from the converter. If you connect a PWM-controlled dump load in parallel with the buck converter which is responsible for taking the excess power the batteries don't usually need, you can rate the buck converter for considerably less current and make it easier and cheaper to build.

I would do all this with a microcontroller, but I think it's also possible to do it the way you are doing it.

Any way you do it, I'm interested in following your project !

Best Regards,

David Voss

[Flux]

Keep going the rewards are worth the effort. You look to be most of the way there.

You are braver than me to tackle 100kHz , I settled for 30 but it may have been ok to go higher.

Mine produces over 30A into 24v with an input voltage going up to about 80.

I am sure the real route is by a 3 phase interleaved buck converter but the controllers are all surface mount and I have reached the age where I would need to build the boards under a microscope and I can't be bothered. Beyond about 2kW you would probably have to go 3 phase or improve the layout or drop the frequency.

To me the issue has always been the converter and as you say there is a lot of information on the internet but they withhold much of the vital stuff about layout , snubbers etc but a few manufacturers data sheets give some advice particularly IR.

You young ones are well capable of dealing with the digital control but my simple analogue scheme works absolutely fine. I measure prop speed, square it and track the output current to speed squared. When you add in the inevitable losses it comes out very close to speed cubed and you can fiddle the track by altering the gain of the tracking loop.

For 12v a synchronus buck converter would be best but at 24v the gain is minimal and I didn't bother. One % extra in converter efficiency is swamped by the 100% or more gain in prop power from correct matching.

Look forward to your results.

Flux

[joestue]

Is that a mix 26 iron powder core? They conduct electricity too well, perfect for dampening harmonics as filter inductors but not so good for buck/boost regulators and transformers, Don't neglect the capacitive input filter requirements, the textbook formulas don't work when the 35 volts at the fet drops to 25 at turn off.

Other than that first thing I would do is get an oscope and take a look at Vgs during the switching transitions.

Fets are cheap, wind turbines are not...

[Janne]

Hi all, and thanks for the comments. Some replies to the comments come here;

The inductor cores (2 on top of each other) are as far as i know Ni-Zn ferrite material. The relative permeability of them is ~60. They're ripped off from and old UPS supply. The input filter was selected more based on the ripple current rating of the capacitors, rather than capacity. I did scope the Vgs while running the tests (making a gate transformer would be pretty hard without a scope ), and there seemed to be small amount of ripple on top of the driving square wave. Rise and fall times were both 350-400ns, with the gate voltage amplitude of 15V. Duty cycles from 0-90% seem to work ok.

At the moment the diodes don't have any sort of snubbler circuit across them, and I haven't yet looked into the diode voltage waveforms. I also did consider synchronous buck converter, but I wanted to try to avoid the added complexity.

Interleaved 3-phase converter indeed would be ideal, especially when considering the amount of ripple current the input capacitor is going to endure.

when it comes to 100kHz switching frequency, I think "unaware of the pitfalls" would be more appropriate description than brave

[bob golding]

hi,

there is plenty of info that would be relevant to  what you trying to do on the solid state tesla coiling groups. they have done a lot of research into gate drives into big igbt bricks. have a look at the 4hv forum or steve wards site. i was going down this path until my turbine destroyed itself,nothing to do with the inverter, just  too much wind  in too short a space of time. rebuilding the turbine  has taken priority over mppt, but now  the turbine seems to have survived a cornish winter in one piece i will  go back to it. i need to change the bearings again first. no more cheap chinese ones skf this time.

[Ungrounded Lightning Rod]

When doing MPPT on a mill keep this in mind:

 - Energy in wind goes up with the cube of the wind speed.

 - Peak power RPM of a mill goes up with the speed.

 - Voltage of a permanent-magnet alternator goes with the RPM and current with torque, so:

 - Current goes up with the square of the wind speed.

 - And heating goes up with the square of the current.  So:

 - Current goes up with the FOURTH POWER of wind speed on a MPPT-controlled PMA mill.

Fourth power means "double the wind speed, multiply the heating by a factor of 16".  Or "It climbs REALLY FAST once it gets going."

So you MUST get your furling working right and/or have your MPPT controller go into current-limiting mode at high currents.  Otherwise the first sustained high wind will fry your mill.

[Flux]

A lot to think about here.

You are right that with wind power you have to do something to limit power by furling or other means but I don't think the mppt case is as bad as other forms of loading as far as stator heating is concerned.

Without mppt you have the choice of loosing low wind performance and having a good power in high winds or you can go for low wind cut in and deal with the consequences in other ways.

Most here go for very low cut in and you have 2 options, you can let it stall into a powerful alternator and have very little increase in output with wind speed and you might get away with pretty ineffective furling ( and accept abysmal performance)

You can reduce the overall electrical efficiency and keep a reasonable prop performance in the higher winds. The cheapest method is to under build the alternator to incorporate this inefficiency, this works well from cut in to perhaps mid 20mph wind but you need very effective furling to save the thing in higher winds.

You can build a very efficient ( big and expensive) alternator and add the loss mostly outside the stator, this is a good way especially if you can use the heat for something useful and gives a lot more latitude with the furling.

With mppt you are in a similar situation to the last condition where alternator voltage is rising with wind speed but you convert the heat losses to a great deal more power into the battery. Ultimately you have to furl but with a lot more in hand than with a small alternator not run is stall but matched reasonably.

You can current limit the converter but it's  not without problems, once you do this the input voltage to the converter rises rapidly and you have a challenge in dealing with high input voltages. From the mill point of view it is fine, it just behaves like the old reactance limited iron cored alternators and sheds prop load and lets the blades scream away and become partly self limiting from their own inefficiency ( it frightens me and I can't stand the noise).

You also have the option of phasing the converter forward to pull the blades of the peak point and into stall. For best results with mppt you need the alternator as efficient as possible and you may be able to stall hard enough to not need furling but you have to be absolutely certain you can't get through stall in any wind. It does give a lot more latitude for furling by going this route and you can also use it when batteries are full to drastically reduce the demand on any charge controller.

Your figures are right for a machine with near perfect overall efficiency but in reality I find that to match input power cubed I can only get power squared or slightly better into the battery. I am tracking current squared with wind speed at the battery end so alternator current is not rising as quickly as your perfect example predicts.

Good thoughts all the same but my belief is that stator heating is less of an issue with mppt than with direct connection.

Flux

[Dave B]

Way to go Janne,

  It's great to read of and actually see electronic control work done here. You obviously have the background to offer your current work and progress with your project(s)here on the board and I for one certainly appreciate it.

  I'm all for discussion, theory and suggestions etc. but when push comes to shove and the work gets done it's people like yourself who make things happen and help many along the way by sharing not just talk but current hands of your project as well.

  I look forward to reading up on your progress of your projects. Thank you for posting real nuts and bolts work here, it cuts through any fluff like a knife.

  Dave B.

[Janne]

I've been running more tests with the switcher. The main problem right now is the ringing on the phase node when the fets are switching. I calculated the leakage inductance to be around 50nH. I then calculated a suitable snubber to be 20ohm / 2nF. That didn't help much, but 2.5ohm / 4.7nF helped a lot. It is still ringing, but not as bad as before.

The other problem, which i think is related a lot to the phase node ringing, is running the switcher at low duty cycles. In low duty cycles the ringing in the gate signal is so high in amplitude, that it will cause the fet to start oscillating as well. This already caused the other FET to blow up, while i tried a current limited startup agains a short circuit. Duty cycles down to about 5-10% work ok, though they are messy.

Anyways, with 35VDC input, and output at 35A, I measured the efficiency of the converter to be around 0.85, so it looks it might be good enough, If i can solve the ringing in the fet gate.

[bob golding]

have you got a stopper resistor on the fet gate? something between 20 and  100 ohms right on the pin is supposed to help.

[Janne]

Hi,

Thanks for the tip. I tried 15 ohms & 47 ohms in series with the fet gate, instead of the 3.3 ohm that was in there before. 15 ohms didn't do much anything, and 47 ohms just made the fet switch slower, but the ringing in the phase node is still the same.

I should also add, that the gate signal is a perfect square without any ringing, until power is applied to the switcher.

I also tried a schottky diode in place of the current diodes(3x mur2020rg's), and that helped also some. The bad thing about schottkys is, that are connected just the opposite way compared to the mur2020rg diodes currently in there, so I would need to rework the busses if i wanted to change the diodes.

[GWatPE]

Hi Janne,

a useful addition is a small ferrite tube, placed over the leg of the FET.  This will reduce the RFI pickup in the FET gate.

BTW; I prefer to switch the negative side, and have a common +ve.  The mill is usually isolated from electrical earth, similar to a solar panel.  I use opto coupled battery voltage sensing, and a micro.  I also just measure the unloaded input and also the output voltage, to calculate the duty cycle.  My Buck cct is an open loop control design, and is rated to 1kW.  100kHz does create high peak currents as the inductor tends to be very low resistance.  18-20kHz is more manageable.  There should be a simple way of modding a solar MPPT.  At least the power electronic part would have already been worked out.  Splicing a new brain into say an Outback or similar solar MPPT should be doable.  Efficiency of 97-99% should be possible.  AERL quote a maximum 99.7% efficiency on a coolmax buck type MPPT.  No heatsinks, and all runs cold.  

I believe that high efficiency is achieved when all the digital waveforms are close to perfectly square, and there is no ringing, or slope to the switching signal edges.  I looked at the waveforms on the AERL, and the square waves consisted of a series of horizontal lines.  The rise times on the gate, and the output switching were nS.  

Good luck with your testing as you get a lot of satisfaction when the end product works well to match your mill.

Gordon.

[Janne]

Hi Gordon,

Thanks for the comments. Indeed, ferrite bead in the gate lead helped some when I tried it. Also, using a ready solar mppt power stage would have made life easier, as would have lower switching frequency, but I'm refusing to give up at this stage .

I went with high side switch & isolated gate drive, because the common ground made some stuff easier, and isolated gate drive didn't seem too hard at that point. When I think about the situation now, it might have been easier to go with a low side switch, and driving the fet without the signal transformer.

Anyways, with added heatsinks I was able to test output power of 35A, 11V into a dump load. With the input voltage being 35VDC, I measured the efficieny of the controller to be about 0.84. With lower duty cycle (and power), the efficiency was slightly less, about 0,83 which I think is because of the higher freewheeling diode losses. I think I can accept this level of efficiency, considering the level of my expertice about swithed mode power supplies before this thing.

As the next test I'm going to test the signal again with a different kind of oscilloscope lead, for some reason the measurement probe I currently have is heating up.. I wonder if it has something to do with the constant swinging of the "ground" potential (Fet source).

I've also discussed this thing in the picaxeforum, if you want to follow the discussion there, it's part of this thread:

http://www.picaxeforum.co.uk/showthread.php?t=10986&page=2

[GWatPE]

Hi Janne,

Sounds like you have a ground loop problem with the CRO, if the lead is heating up.  Could also be the cause of low efficiency, if it was connected during those measurements.  You should use a DC electrically isolated CRO input, for high side gate measurements.

12V systems are a pain with diode losses, so a synchronous rectifier may still be the way to go.  If you get it wrong, though, pfff.

Gordon.

[Janne]

It's been a while but I'm back at working the mppt system.
The problems I had earlier with the gate drive were not too difficult to solve, I did some more fine tuning with thr RC snubber across the switch node. Also, I added one turn to the gate driving transformers secondary, and let the zener diode clip the voltage at the gate to 15V. That cleaned up the drive signal really nicely, and now it works stable even to the lowest possible duty cycles, around 1%.

I assembled the thing in a steel box, that i picked from the scrapyard. The only thing missing now is some wind to test it out, but it's been really cold and calm here for the last couple of days. Hoping for better test weather for the next weekend.






Higher resolution pics can be found at:

http://pics.ww.com/v/Janne/Electronics/P1060284.JPG.html
http://pics.ww.com/v/Janne/Electronics/P1060288.JPG.html

[Janne]

Not entirely on topic, but I guess it's okay to mess up a little in a diary post.  I also recorded power and wind data from the turbine, as to get some sort of reference to compare performance with the MPPT controller. So these graphs are from the 3.2m diameter axial flux turbine, connected directly to the 12V battery bank via rectifiers.

I recorded the averaged values from 1 minute, and from those I averaged the values and generated the graphs. The coefficient graph is just the recorded power divided by the kinetic power of the wind available for this 3.2m diameter rotor.

Blue graph is the coefficient of power, and red is tha actual power recorded, both in relation to the windspeed in m/s. During the time data was logged, wind was very calm, but it's clearly seen how the efficiency quickly starts to drop after peak power, as the rotor starts to stall.



Bigger and more clear iimage:
http://pics.ww.com/d/431397-1/hyotysuhde.gif

[ghurd]

Nice work!
G-

[TomW]

Not entirely on topic, but I guess it's okay to mess up a little in a diary post. 

Especially when it is YOUR diary

[Janne]

Well it's not been smooth sailing with commissioning of the unit. Last weekend I made some initial test runs, with the power section bypassed directly to battery, so I could test the controller part of it alone. The current measurement was not working, as the controller board was drawing more current than I had anticipated, which caused offset voltage to the quite small suplly wires. This offset voltage then created a problem, as the current is measured with a 0.1V / 50A shunt, so only currents above of 15 amps were being measured. This problem was rather easily fixed by changing the supply wires to thicker and shorter ones, hence reducing the offset voltage down enough not to be meaningful.

Today I started to test the whole unit, and removed the bypass wire. First I was filled with joy with current flowing to the batteries again,  but it soon turned to frustration as I noticed the voltage on the primary side was just the same as the battery voltage! Something fishy going on. It was soon obvious the switching transistor had died, shorted internaly. RIP another FET  . No wonder the controller board was drawing excess current, trying to drive a shorted FET gate...

What's puzzling me, is what caused the transistor to die? The last time I tested it on the test bench, it worked ok, and before this day it was just being hooked on the battery, being bypassed. Only thing I can think of is that when I hooked up the unit to the battery bank, the sparking somehow caused a voltage spike to fry the FET. I had put the bypassing wire in there earlier, to prevent the primary capacitor bank charging through the FET body diode, and possibly frying it, so that is not a likely cause.

The annoying thing is, that it needs to be taken almost completely apart to change the damned transistor. After that is done, I'm going to first try and charge the capacitors in the controller with a resistor, and only after that connect the cables directly.

[joestue]

do you have photos of the switching waveforms?

It is unlikely that the buck inductor is pushing the input capacitors up to 50 volts higher than the battery when you plug it in, the body diode of the fet could certainly stand the current however.

[Janne]

Joestue,

I don't have pictures of the waveforms at the moment, but to describe, the gate drive signal is very clean 0V - 15V "square" wave, with slow ramps about 300ns both falling and rising. The switch node waveform is more messy, with ringing that maxes out at 80V peak at maximum tested power. Freewheeling diodes can withstand 200V, so at least those are not endangered by it.

I will take a scope with me, when I will install the controller again. Yesterday I changed the FET's, and at least to the outside the old fried one (other had fried due to gate drive problems in a test earlier, could be that the other one was damaged as well..)looked perfectly ok, but all pins are shorted together. I'm going to try to charge the capacitor bank with a resistor this time, instead of directly hooking it up to the battery.

[Janne]

Today, armed with fresh set of MOSFETs I decided to give the controller another go. This time everything went smoothly, and the controller survived the hookup to the battery ok.

So far everything seems to be working, at least somewhat. The controller board seems to make an audible whining at currents above 20A to the battery, so I think the current feedback loop is somehow oscillating. Other than that, everything seems smooth, FET and diode's are running cool. Next time I'll take a battery powered scope with me, so I can scope the waveforms on the FET gate and on the switch node, to see that there are hopefully no problems in there. And also I'm hoping to shed light to the instability of the control loop.

Next, if there're no new problems with the controller, I'm going to hook up the datalogger onto it again. I can then use the data to see, how well the turbine is tracking the optimal TSR with the current load curve. After that, it should be easy to do the adjustments required to the power curve to further improve the performance.







(Larger versions of the pictures can be found at: http://pics.ww.com/v/Janne/Electronics/?g2_page=5)

[Janne]

Well it's been on the test for a couple of weeks so far and everything has worked without a hitch. It survived a couple of windy days, pulling ~45A at best before furling. Furling still works too, I was quite worried about that aspect based on previous experience of non-furling cases.
I've nearly finished the new datalogger, that will log data directly to a µSD card. So soon I'll be able to tap into some real data, instead of just thinking and guessing