-
4.7 and 6.6uF? That's way too little...
It seems you've discovered why!
Tim -
4.7 and 6.6uF? That's way too little...
Well like I say all designs are a compromise, economics, space, performance etc
It seems you've discovered why! Tim
Firstly this converter operates at 125Khz, there are no ripple components below that, secondly the load is purely resistive and there is no requirement for a low ripple output, thirdly the panels are a fairly low impedance source so the remaining ripple current produces a small ripple voltage that marginally reduces there efficiency however the ripple imposed by the mppt algorithm is much larger! I do not feel the need to fit excessive expensive space consuming capacitance just to be nearer some theoretical perfection, as it is the converter is cheap, small and reasonably efficient and insofar as its main capacitors electrolytic free!
In fact the main reason for any output capacitors is to control the rate of rise of output voltage should the output suddenly go open for some reason (fuse etc).
I had my own targets for the design and largely despite a few teething troubles they have been met
-
And the teething troubles continue this time caused by the contents of the junk box! Old GREY silpads OMG I hadn't realized how bad they are compared to mica TO220 insulators see here http://www.bergquistcompany.com/thermal_materials/graph_pdfs/COMPARE_SPVALUE.pdf (400 series) and here https://www.fairchildsemi.com/application-notes/AN/AN-4166.pdf for a mica comparison.
Anyway as I had some old grey silpads in my junkbox that's what I used and I am pretty sure that's what led to yet another mosfet failure today, I actually watched it dying slowly on the scope but didn't realize that's what I was watching at the time, I now have a mosfet graveyard in the corner of my workbench having lost about three in the last few days. At the time the heatsink temperature was around 40C and the mosfet dissipation could have been 20W so that means the tab could have reached 200C, should have spat on it to check it
Talk about unfit for purpose, whats the point of pad material that gives an interface of 8C/W under moderate pressure grrrrrrr rant rant grrrrrrrr
So I think more effort to be expended on a bench test setup to allow higher power testing under controlled conditions rather than have the sun let rip on me while I am dangling like a bat up some ladders lols
Just in case anybodies interested there is a 13 year long discussion on pad materials here http://www.diyaudio.com/forums/pass-labs/37262-mica-goop.html -
I know how much you like pictures so here is the bench test rig. There are a set of transformers capable of providing 110Vdc & 5A (limited by rectifier), the big blue variac can be used to change the voltage. This is connected in series with an ELECTRIC KETTLE as a ballast to represent the non-linear PV source. The output of the converter is connected to an electric fire with 1, 2 & 3Kw settings to represent the load. Ideally my source would be a CV/CC psu of around 1.5Kva but sadly I don't have one of those! So far the horrid grey silpads have been replaced with mica and gunk and the shorted mosfet replaced. I have been checking and verifying very carefully the mosfet switching conditions, as the rig cannot run at anything like full power I have been copying my observations at 200W into spice and then cranking up the power to 1100W in spice, once again not ideal but better then nothing I really cannot find anything wrong with it other than the suspect silpads, the mosfet dissipation extrapolates out to 16W at full power so it's certainly possible that was the cause.
There are some unexpectedly large current spikes at turn-on, either the boost inductor has quite a lot of capacitance, Qrr of boost diode or it's simply coupling EMI into the imperfect probes, in any case the mosfet rating is 65A (actually the soa is 130A@200V for 1uS) so it shouldn't be bothered. The turn on/off times are deliberately slowed to reduce EMI.
vdrive and vgate are the two sides of the gate resistor, the mosfet is an IRFB4227.
So today I will use the rig to verify some complex software around the buck to boost transition, much easier than waiting for the sun to do just the right thing!
-
Huh, gate driver appears to be around 5 ohms Rds(on), and the gate resistor around 15 ohms.
What's the yellow trace actually measuring? It doesn't seem to be carrying any load current at all, if that's what you had intended...
I hope the double-blip shape of the yellow trace isn't due to shoot-through or something. The gate waveform is quite slow, so this would be a concern in a synchronous converter.
If you're using a PN diode, yes, reverse recovery is a big deal. It can pay to increase (and snub) loop inductance, to reduce dI/dt and peak recovery current. There are silicon schottky diodes available in that size (10s of amps, 200V), though they aren't very good (Vf is comparable to PN diodes -- but with no recovery loss, so this alone can be worthwhile). (There are SiC schottky too, but these begin at 600V (and Vf ~ 2V), so they won't be worth it here.)
Ed: nevermind, I see MBR20200CT. Pretty reasonable. The schematic is very hard to read.
Tim -
Hello Tim, yes that's about right for the gate. The yellow trace is across the source resistor so its supposed to be source current, the main reason for including it in the scope shots is a common visible timing reference (I only have two channels). I suspect most of the actual trace is made up of probe artifacts and EMI as no attempt has been made at tip grounding. Ohh its an 0R05 resistor and were only running 200W so the normal run current is practically invisible, this resistor serves as an overcurrent protection not for current measurement as such TBH its almost superfluous except for inductor saturation that could occur due to a software fault.
I am sorry the schematic is very hard to read, it has to much crammed onto one sheet, just grew like topsy
Yup it's a schottky so Qrr isn't so bad although it is a fairly high voltage circuit. Some turn on spike will come from the snubber, some from the boost recovery and some from boost inductor capacitance, I don't really have the equipment to measure it all, as long as it is small enough not to represent a failure mode or generate to much EMI I am happy. I remain a little concerned about the double nature of that blip and may well spend some time trying to resolve the cause later. -
Today finished bench testing, as much as I can within the power available (a few hundred watts). Re-checked the weird double source pulse and found it changed when the scope ground was moved so pretty sure the double is actually spurious. Of course a single pulse in this location is perfectly normal Checked out the software moving from boost to buck and back again smoothly, takes a while to figure out how to use the test rig to simulate real PV conditions but got there in the end, basically if the load impedance is varied the MPPT acts to try and maintain the input power at it's peak which given a fixed input ballast resistor means it tries to maintain a constant input current that was visible on a test rig panel meter. I must admit I use the terms buck and pwm interchangeably, in one parlance it is bucking the input voltage but in another its simply pwm control of a heater Currently waiting for the conformal coating to dry prior to re-assembly then back outside (plus of course that kettle boiled) and I hope it knows I have no more mosfets left!! I did spend some time looking for something more robust but frankly the IRFB4227's are hard to beat without spending silly £££££ just hope the good old mica and gunk do the job.
-
Just a quick note to say I am too clever for my own boots, boost mosfet failed again yesterday and it wasn't even sunny! Stupidly I went out after re-installation so wasn't watching (overconfidence). So not thermal, not voltage, leaves current! Inductor could be saturating due to software fault and current limit lackadaisical. Started checking everything from scratch, here is inductor saturation measured on my tester, vertical is 4A/Cm so ~18A, design value was 13A. There is some overly clever software switching PIC PSMC frequency between boost and buck modes (to reduce switching losses in the latter) that may be causing runt extended pwm on times to the boost stage. Unfortunately the PIC PSMC is very poorly documented and so far I have been unable to find any timing parameters for it so even its reaction time to shutdown (overcurrent) is unknown so many more things to measure. Amusing this spell checker doesn't like inductor
Update......I found a software fault that could cause the PWM to drive the boost stage at 100% duty cycle for several milli-seconds occasionally, certainly enough to saturate the inductor. The remaining mystery is why the resulting overcurrent was apparently unable to shutdown the PWM!
Answer leading edge blanking cap was HUGE! giving 5.5uS delay, well long enough to fry a mosfet facing a saturated inductor! Somehow a gremlin crept in and it was 100nF not 1nF shown on schematic........
No more mosfets till Wednesday as RS my usual supplier have let themselves run out of stock of one of the most popular mosfets till mid August......RS are going down the tubes, getting less and less interested in discretes! Anybody have this problem, switch to Farnell, most MOQ's reasonable too, just don't work weekends.
-
Ewww this PIC PSMC thing is really nasty! firstly after several days on the mchip forum asking, no replies whatsoever regarding it's AC specifications. Then today I find a new glitch and it's caused by the way I handle it (dynamically changing steering) but I did find a way out, there is a synch bit that allows you to specify changing steering synchronously with the period. But another glitch caused by a similar problem with polarity doesn't have a synch option, however that glitch occurs at very low duty cycle and is not in the inductor (boost) drive so unlikely to have any bad effect. Just beware anybody using this thing seriously you need kid gloves and a lot of time and patience pictures are some glitches.
Mosfets arrived today, fitted and bench tested, nearly flooded out yesterday/last night/this morning so a bit damp for refitting outside today.
Have a nice day
-
Re-installed 0830 today, horrible grey skies all day but at 1600 have gotten 1Kwh, not bad for a 1Kw array, ran in pwm mode all day, power typically 75-200W (peaked at 465W). Tank thermostat just cut off at 64degC (148degF) and a few hours daylight left so very happy.
Next project is to use the surplus energy without batteries!
BUT first of all see if it can hang together when the sun comes out again (maybe a week or so we seem to have been plunged into uk clag for a while). -
Well we have had some brief interludes of sun, no blown mosfets but a very aggressive overcurrent protector, every time it gets up between 550-600W it shuts down! I have proved this is the boost stage overcurrent protection that uses a 0R05 source resistor with a 100R/180pF filter into a comparator who's reference is a silicon diode (~600mV), the output of this comparator internally shuts down the PIC PSMC (Programmable Switch Mode Controller) module. OK so it could be the leading edge spike except the PSMC has LEB (Leading Edge Blanking) programmed and I have tried up to 1uS but the problem remains. Sadly I cannot access the comparator output with a scope as the output pin for it is in use for an ADC input with a whopping 680nF filter cap!
So it could be LEB doesn't work on this chip, or I am misunderstanding how to use/program it, or something else. This problem didn't used to happen when the filter on this pin was mistakenly 100nF, then it just used to evaporate mosfets instead!
It could be noise entering the pin at some other time than the leading edge although I have seen no evidence of it. So for now I just have my thinking cap on!
Scope shot is driver output (pre gate resistor) and comparator input after LC filter.
2nd scope shot is worst case noise at comparator input BUT should be covered by LEB
I am thinking of extending LEB even more than 1uS or slowing down the comparator but I am a little chicken to loose yet another mosfet even though I think I may have found the cause of that (occasional crazy on times).
Update..... managed to conceive a way of getting a realtime output from the PSMC when it's shutdown to trigger the scope this proves beyond doubt is is the current comparator and that LEB is not working!
3rd shot shows shutdown together with filtered overcurrent signal, 4th shot shows shutdown together with PIC boost drive signal (LEB starts at its leading edge). LEB here is set to 500nS!
-
Well at last a work-around for the unexpected shutdowns from the boost overcurrent! The thing that fixed it in the end was setting the comparator to SLOW MODE! I don't actually know why this works
Either LEB doesn't work at all (well I have not seen any reports of anyone trying to use it), or there is a glitch at some other time than the leading edge, well there is no scope evidence for that and there is also no high power activity other than just this boost converter, so no asynchronous noise source except the APU but it's only <100mW when the fan is not running (it wasn't) and it's on a separate PCB.
I think my conclusion is LEB doesn't work at least on this chip and from the C1 input. That's what caused me to increase the filter capacitor so dramatically and enter the sorry downward spiral of mosfet failures The only way to prove this would be an extensive breadboarding exercise that frankly I don't have time for.
Well thank you Microchip, you have cost me ££££ in failed mosfets and probably a month finding out your chip doesn't work, don't suggest I ask a question on your forum nor ask for tech support, I am still waiting after 2 weeks for someone to even answer the simple question of what is the Tpd from an asynch shutdown input to the PSMC outputs changing state! Time to move on to a better specified chip methinks for the future.
When you consider this PSMC is also used in the flagship DsPIC it's astonishing how poorly supported and documented it is!! -
It is a poor designer who blames an MCU for hardware failures... just sayin'.
I find that a borish and stupid comment to say the least, I didn't think this site was infected with stupid trolls, perhaps I was wrong!!
Tim
It is a matter of fact that I have found several silicon faults in a long career and as author of several patents and past member of IEEE do not consider myself a poor designer, perhaps you should examine you mirror more closely, and please do not reply as I have no further interest in your comments.
-
Well I did set up a test rig specifically to test leading edge blanking on this chip and indeed it doesn't work, duly documented and sent to Mchip. Fortunately there is a work around of sorts to simultaneously program trailing edge blanking though the downsides of this are application dependent.
I also measured the undocumented PSMC in to out delay and got a creditable 20nS.
So I have some fixes to install outside, this will allow the power to ramp up fully on hot days (not to many of those around at the moment)
1st scope shot shows PSMC in-out delay, 2nd shows a leading edge pulse successfully ignored by the blanking feature using the workaround.
PS I should mention the path from the comparator to the output pin is different from the internal path from the comparator to the PSMC so these measurements are only approximations to the conditions internal to the chip (if that makes sense)
PPS I had to set up a specific test rig as on the actual target the vital (to this problem) C1OUT pin is in use for something else that could not be dispensed with.
-
Been chugging away outside for almost a week, some sun so been up to 1100W, no more blown mosfets
As for Mchip, they have ways of keeping there errata small, ensure you have a business case before investigating any bugs
I seem to remember someone on this site comparing micro-controllers according to the size of there errata, well that doesn't work if the manufacturer doesn't update the errata with known bugs 
EDIT: Just added a plot of tank temperature and watt hours for the last week or so, on the whole the weather has been poor but it's still enough to keep the water temperature up.
-
You are quite right, the scope shot shows what happens in low light conditions, the MPPT is switching the duty cycle between nothing and something small but the PWM EMC swamps the current signal, the scope probe connection is not good so some EMC is entering that way.
Can you have the A/D sample in sync with the PWM so it does not do a conversion in the noise area? That's common with motor control.
-
Can you have the A/D sample in sync with the PWM so it does not do a conversion in the noise area? That's common with motor control.
Yes, it took me a long time to figure how to do it on that chip and it did make a dramatic improvement but then the whole thing has been a case of incremental improvement, like kelvin connections on the current sense amp also suggested here, oversampling and averaging the ADC, slowing down switching speeds and adding a flux strap to the inductor, all have contributed It's been a learning curve, very different from a simple SMPS! Many years ago I used to work on servo-motor systems with accurately controlled velocity profiles BUT the power stage was linear (big heatsinks) so no switching noise!!
Ohh I should also mention the "noise area" is somewhat extended by relatively slow op-amps AND is integrated by any low pass filtering prior to the adc further compounding the problem......so no magic solution just suppressed enough for the system to work
I should imagine the problems recovering low level current sense signals in high power motor inverters are a complete nightmare.
-
I have been working on MPPT charge controllers as a hobby for a long time now.
My latest MPPT prototype oversamples the ADC pins 60 times and averages, 30 times a second. The MPPT duty cycle is adjusted about 15 times a second.
To sample current I use a MAX4080S that reads from a 1mOhm sense resistor, the output of that is tied to a 10k resistor and 0.1uF ceramic cap to ground before going into the PIC ADC pin. The PIC is an older model 8 bit job with a 10bit ADC. I can get accurate 13bit readings out of the thing by oversampling (I don't divide the 60 readings by 60, generally by roughly 10).
My MPPT tracks very well, though I have sweep and active tracking algorithms (full sweep occurs every half an hour in bulk mode - adjustable - takes about 5 seconds) and active tracking is constant in Bulk, Absorb and Float modes. The sweep action is mostly redundant though as the active tracking is always spot on.
My advice therefore is to oversample a lot. DC conversion is noisy, no matter how much component level filtering you do, digital oversampling is necessary. Also make absolutely sure your PIC voltage rail is very stable. I use a MAX5033 powered from the battery side along with a combination of 7 capacitors to keep the thing rock solid. -
Hi Redsky, thank you for your comments I use binary modulo's for multiply/divide so I can use shift operations, sometimes it's a bit limiting but so much faster. My PV voltage is much higher so I use low side sensing, any op-amp with a common mode extending below zero will do, I used LTC1637 because I happened to have one I actually round my current resolution to 8 bit before MPPT processing, with an FSD of 10Amps this still gives a resolution of 39mA that I find more than enough.
I thought the PIC ADC source impedance spec was somewhere around <=2K but I may be wrong, in any case I use just 470R/10nF to kill the real nasties prior to the ADC, I found a lower Fc causes noise integration within the lpf leading to bigger errors.
Of course the whole thing is highly dependent upon the particular noise environment your working with so every case is individual
I agree on the ADC reference voltage or PIC supply rail, I use a LP2950-33 linear to generate 3V3 for the PIC from a ~12V source so the 3V3 is stable and nice and quiet.
My MPPT runs at around 16hZ that I find fast enough to track clouds, fortunately I don't have any tree shading to worry about, I don't bother with the additional complexity/power loss of sweeping because as you say I always find the MPPT spot on after extensive testing although that did take some effort at very low insolations (I use an adaptive step size).
It's good to hear what choices other people make, always interesting
-
I should point out the 10k resistor is in parallel with the cap to ground, not series and you are right, it is fairly specific to using the MAX4080 chip as the current sense to get very accurate and stable readings.
My prototype is designed to work with up to 50 amps output. I process my current readings so they end up as the amperage value to 2 decimal places. For example 1234 in a variable would represent 12.34 amps. I use that value in my MPPT algorithm and for display/data logging and other regulator functionality. The duty cycle is changed in 0.2% increments. It results in tracking that keeps fairly close, say +/- 0.2V to the actual MPP at most times, and doesn't hunt that much, which in my case is fine. For example, The MPP may still hunt up a few volts within 10 seconds if the sun came out behind a cloud, which to me is fast enough and leads to maximum extraction of power in a day. I don't do adaptive step size, for me that was unnecessary but I have read plenty of whitepapers that it's one of the best methods. To deal with low isolation, I have very low amperage resolution (10mA) which helps, though i'll admit if the power is down to within a few amps output level the active tracking may get stuck if it wasn't for the full sweep now and then, I could of course increase the duty cycle increment on the active tracking to deal with that if I wanted but It's all a compromise at the end of the day, there is no perfect method.
I generally agree with your comments on the sweep vs active algorithm, however the sweep algorithm can be quite impressive. I do a full solar open circuit to battery voltage sweep in less than 5 seconds, and for half of that 5 seconds its running a fine sweep very close to the MPP thus losing very little available power. It also finds the MPP with an accuracy of 0.1V
If you do all the maths it works out for two sweeps an hour I lose on average less than 0.1% of the available energy from the sweeps.
I believe the Morningstar Tristar MPPT uses even faster sweeps as it's primary way of tracking, so does the Outback MPPT and the Midnite Classic, though some of these have configurable algorithm options.
For me that small percentage loss from more than a couple of sweeps per hour is unacceptable. But I guess these big companies have to allow for all the different customer systems and forms of shading that might occur and can see that sweeping would be more reliable in some specific cases.
Having a sweep algorithm is also very handy to check your own active tracking algorithm. I can initiate a sweep instantly anytime which then provides the results on the display for confirmation that the active tracking algorithm is accurate. It's also useful to give the active tracking algorithm a head start when the MPPT controller is first turned on.
I too think reading how you are going about doing your design is interesting. Keep up the good work and posting about it. -
My prototype is designed to work with up to 50 amps output. I process my current readings so they end up as the amperage value to 2 decimal places. For example 1234 in a variable would represent 12.34 amps. I use that value in my MPPT algorithm and for display/data logging and other regulator functionality.
Ahh potentially we are talking about different things here I measure my PV current for the MPPT algorithm! Measuring my output current would be no good as it is not directly related to PV current and would therefore give a false MPP for the panels. As I am measuring PV current in a single series string IFSD is 10A as that exceeds ISC of any panel I have encountered and certainly those I use (Axitec AC-260P/156-60S).
-
different load yes, still maximum power point tracking
-
This is the best thread I have read here. Thank you for detail and all the contributions. This is the level I hope to get to and would be stoked if I can pick your brain sometime fourtytwo42. Thanks for the knowledge.
-
Thats very kind of you to say that and thank you, I often wonder if my ramblings are any help to anybody, if nothing else I hope they either make people smile or inspire them to have a go themselves. Anybody is welcome to pick whatever is in my head anytime they like