Notice. New forum software under development. It's going to miss a few functions and look a bit ugly for a while, but I'm working on it full time now as the old forum was too unstable. Couple days, all good. If you notice any issues, please contact me.
Ralph2k6 Senior Member Joined: 24/09/2017 Location: AustraliaPosts: 129
Posted: 04:27am 10 Jul 2018
Copy link to clipboard
Print this post
It's not just off grid here Rodger. Alot of the projects here are for various pursuits or hobbies that relate to r.e. in general, grid or otherwise.Ralph
zaphod Regular Member Joined: 03/06/2018 Location: United KingdomPosts: 93
Posted: 03:42pm 18 Jul 2018
Copy link to clipboard
Print this post
Well guess I should continue in case anybody is ever building something like this :)
Unfortunatly not long after my last post during EMC testing the H-bridge failed again, remembering this now contains 1200V IGBT's!! and this time it was one IGBT that failed with a punched through gate (same way the mosfets used to fail). The fundimental problem I have in this area is lack of test equipment, I simply do not posess high voltage scope probes and although I have made some high voltage DVM probes they are not suitable for use near high voltage switching circuits as they are unsheilded.
To cut a long story short I listed some possible causes and fired up the simulator to try and find what was happening, my initial thought was the transformer leakage inductance was much bigger than I thought and/or the secondary snubbers were not doing there job, well neither of those proved to be a problem (I re-measured leakage at under 700nH). As I progressed further along the high voltage chain I inserted a simple switch where the H-bridge was in the output to examine what happened when the H-bridge unloaded the smps and WHOAAA I found the elephant in the room (see attachment except I cannot load png's and its only 43Kb ?)!! OK I will describe it instead, the open end of the smps output inductor rings at +/-5Kv!! Very dramatic in a picture, as you can imagine the H-bridge will avalanch but depending on conditions may not be able to absorb all the energy hence device/s destruction, the whole thing being random and unpredictable..............
Now in utopia this is not supposed to happen because the H-bridge will only switch off at or near zero cross when the smps is ramping down to zero power BUT as I have explained before I live in a rural area with really nasty stuff on the grid and I already have a lot of trouble tracking zero-cross in software, noise often occours near zero-cross as at that point the noise immunity of its detector is at its worst. To protect the H-bridge from catastrophic software failure there is a hardware path from the zero-cross detector to the H-bridge forcing the H-bridge open near zero cross. Unfortunatly this also provides a path for the H-bridge to open should there be a spike from the zero-cross detector that because spikes are ignored by the software will not cause the smps to shutdown, in other words in real life it is possible for the H-bridge to unload the smps whilst it is still working!
Of course this is only the latest in a series of hypothesis but at least it fits the circumstances so is worth a shot :) Meanwhile the distributor doubled the price of the IGBT's claiming they were hard to get hold of so I found them cheaper on Fleabay instead!
Edited by zaphod 2018-07-20Cheers Roger 1Kwp DIY PV + Woodburner + Rainwater scavanger :)
Solar Mike Guru Joined: 08/02/2015 Location: New ZealandPosts: 1138
Posted: 09:07pm 18 Jul 2018
Copy link to clipboard
Print this post
Can you make a large LC + Transorb mains filter for your grid feed, get rid of any spikes so your software can work.
Cheers Mike
wiseguy Guru Joined: 21/06/2018 Location: AustraliaPosts: 1156
Posted: 04:02am 19 Jul 2018
Copy link to clipboard
Print this post
Dont you just hate it when you spend a lot of effort in an answer and then you go back a page to check something and presto you return to a blank "post reply" form ?
The following comments are made without the benefit of a circuit & may be wrongly deduced.
Perhaps it would be helpful to determine zero current flow (in the inductive path as this caused the huge spike), which may not be at the precise zero crossing point? Or and combine zero/minimum current with zero crossing. Can you make a software phase locked loop to more accurately give you the zero crossings - even if they aren't the whole answer.
Can you use an active circuit ie transorb or Mosfet/IGBT with zeners to form an active 1KV clamp to swallow the spike? Depending on topology perhaps using an existing IGBT ? In my experience I prefer to actively swallow a spike than to rely on avalanching.
Are you able to tell whether the gates failed Collector/Gate or Gate/Emitter ? If the latter, is there a gate zener/clamp used ? I havent really worked much with IGBT's, using collector & emitter in the same sentence as gate just sounds wrong..... Edited by wiseguy 2018-07-20If at first you dont succeed, I suggest you avoid sky diving.... Cheers Mike
zaphod Regular Member Joined: 03/06/2018 Location: United KingdomPosts: 93
Posted: 10:15am 19 Jul 2018
Copy link to clipboard
Print this post
Hi Mike thank you for your reply, I would like to make it work without any special needs but also this will increase the grid impedance seen by the CSI and given it is also a pretty hefty noise source may be self defeating. Also an EMC filter for the whole house would be expensive even without a surge arrestor!Cheers Roger 1Kwp DIY PV + Woodburner + Rainwater scavanger :)
zaphod Regular Member Joined: 03/06/2018 Location: United KingdomPosts: 93
Posted: 10:34am 19 Jul 2018
Copy link to clipboard
Print this post
Thanks for the reply I would post the topology except the image posting doesnt seem to work anymore :( Ahh the inductor is on the output of a 25Khz smps, not grid frequency so it's current passes through zero at 25Khz during normal operation, the problem is if you disconnect it from the load (the grid) then it flies back like any self respecting inductor would :) Of course I was not thinking of such a circumstance when designing the circuit.......
The software contains a phase locked loop and that is why it is not sensative to noise to such a degree but consiquentially it can be slow to respond hence the hardware override.
Yes in simulation at least I have found a moderate RC snubber will swallow most of the energy and keep the voltage below 1Kv (just) I would post a pic but......
I have not yet removed the old IGBT as I dont want the paste to dry so waiting for the new one but it also has only 0.7R from collector to emitter so pretty much totally dead.
It is an IXGP20N120B3 there are no zeners mentioned on the data sheet but the driver supply is 15V only. There is a potential failure mode due to parasitic turn-on when the driver supply is missing there only being a 10K Rge, however a 50hZ sine wave is not known for high di/dt :) but I am still sniffing around there, anyway to the best of my knowlage the failure occoured during normal operation when the driver power was fully available.
The problem with random intermittent faults is the chance of capturing anything on a scope are about nill, so all you have to go on is intuition :)
P.S. Sorry about the lost post and thank you for your persistance in re-writing it!Edited by zaphod 2018-07-20Cheers Roger 1Kwp DIY PV + Woodburner + Rainwater scavanger :)
wiseguy Guru Joined: 21/06/2018 Location: AustraliaPosts: 1156
Posted: 10:55am 19 Jul 2018
Copy link to clipboard
Print this post
My thoughts about high di/dt were were related to the spike condition and coupling back to the gate from the high di/dt event on the collector. Maybe not relevant in your configuration. If the current flow, even though sourced from 50Hz, is turned off fast enough there Will usually be a fast di/dt event.
Can you synchronise the turn off (disconnect from H bridge) at the end of a 25kHz off period just prior to next turn on - should correspond to a minimum current, worst case is having to wait an extra ~ 40uSecs ?
I am following your thread with interest Good Luck !Edited by wiseguy 2018-07-20If at first you dont succeed, I suggest you avoid sky diving.... Cheers Mike
zaphod Regular Member Joined: 03/06/2018 Location: United KingdomPosts: 93
Posted: 05:50pm 19 Jul 2018
Copy link to clipboard
Print this post
Yes you are quite right but when not operating there is a substantial LC filter between the grid and the H-bridge as well as a standard EMC filter / surge arrestor. I have drawn a topology diagram I just hope the site is nice enough to let me load it :)
Unfortunatly the 25Khz is completely asynchronous to the grid and would not be feasable to phase lock, Exactly as you say the phase of the interrupt strongly affects the amount of energy stored in the inductor and would be one reason for the apparent randomness of the failure! However as you also mentioned if I can snub the energy in an RC then there is far less if any to be dealt with by IGBT avalanche.
I hope the topology diagram is visable, I think it is potentially unique (position of H-bridge in LCL) and solves many other nasty problems.
As it loaded that png I tried the inductor ring again but nogo, maybe its the density of the plot........
Edited by zaphod 2018-07-21Cheers Roger 1Kwp DIY PV + Woodburner + Rainwater scavanger :)
Warpspeed Guru Joined: 09/08/2007 Location: AustraliaPosts: 4406
Posted: 09:24pm 19 Jul 2018
Copy link to clipboard
Print this post
Its certainly unique, and current drive has rather a lot going for it.
Direct mains connection is a bit of a worry though, as there is no way of specifying how high voltage spikes may go. Some years a go we connected a Dranetz mains analyser up, and (microsecond) voltage spikes of several Kv were surprisingly common.
Not a lot of energy there as the pulses were so brief, but still a concern.Edited by Warpspeed 2018-07-21Cheers, Tony.
wiseguy Guru Joined: 21/06/2018 Location: AustraliaPosts: 1156
Posted: 03:04am 20 Jul 2018
Copy link to clipboard
Print this post
A picture is worth a heap of words. The last grid feed inverter I worked on used a self commutating FET bridge (no logic or uC was used) a 500V FET died...no surprise.
One aspect I liked though was that there was a small dead-band at the zero crossings so shoot-through was not possible. From memory it had a front end push pull converter and it monitored the mains differentially and used that as a reference signal to modulate current from the push pull converter on the primary side.
The 5kV spike effect you saw, if I understand correctly, by open-circuiting between the current source and bridge stage is probably not truly representative of the failures you are having, as current interruption should be occurring near the zero crossings at low/no current ideally. With the series switch, it could be openened much further from zero with higher currents flowing & much more spectacular results.
If the IGBT bridge is controlled by software is it allowable (near the mains zero crossings) to stop the front end current source and then commutate one side of the IGBT bridge off and then commutate the other side on and then re-enable the current feed. Alternatively if that cant happen could an IGBT be placed directly across the output of the 25khz bridge rectifier (a current shunt) but can only be activated for the time around the zero crossing, when commutating from one leg to the other of the IGBT bridge.
I know that last bit sounds drastic but the current & voltage from the 25kHz current feed should be at a minimum then so little power is actually passing through the current shunt. Although there may be a small increase in signal distortion from your grid feed near zero crossings, in the scheme of things it is of little consequence?
What happens if the mains feed to your property open circuits due to a line fuse blowing in the middle of high current flow through your bridge ? Could that also be an issue? Too many questions...sorry If at first you dont succeed, I suggest you avoid sky diving.... Cheers Mike
zaphod Regular Member Joined: 03/06/2018 Location: United KingdomPosts: 93
Posted: 08:39am 21 Jul 2018
Copy link to clipboard
Print this post
I agree even the action of plugging and unplugging something with a reactive element can do that. Apart from the EMC filter that provides a bit of damping the VDR's in this design are there to swallow the energy in those Kv transients hopefully :)Cheers Roger 1Kwp DIY PV + Woodburner + Rainwater scavanger :)
zaphod Regular Member Joined: 03/06/2018 Location: United KingdomPosts: 93
Posted: 08:58am 21 Jul 2018
Copy link to clipboard
Print this post
Yup well first up there is a dead band enforced by the zero-cross detector that consists of two detectors one for +half cycles the other for - that cannot overlap, on top of that there is an output from the pll that disables the bridge when it thinks it is zero cross and the pll directly controls the smps ensuring it ramps towards shutdown and back up again around zero-cross so all the mechinisms are there BUT for some reason something occasionally goes wrong (and yes it could also be crap software I am renowned for not being the worlds best coder) :(
Yes I do like your IGBT shunt idea that would certainly be belt-n-braces but nothing wrong with that however if its a pll glitch and the pll also controls the shunt IGBT.......
The line fuse could happen and in fact we often get power losses during storms but usually the re-closers succeed in sorting it out! Of course this whole inverter impliments anti-islanding so it drops out on the first outage, not gonna worry me as not much sun around usually during a storm hahaha
I agree it is unlikely the H-bridge should open during high power output BUT I am doing the usual clutching at straws because ATM I cannot think of any other reason for a 1200V IGBT to blow :)
I might try increasing the deadband around zero-cross as there are already lots of waveform problems around there and like you say adding a little bit more isnt going to be noticible, with these particular IGBT's RBSOA=FBSOA so turning the wrong bridge legs on is not an issue and it is impossible through hardware to turn on more than one diagonal at a time or even the wrong diagonal......in theory :)
One other posibility is the inteferance causes the bridge to bounce (multiple on-off transitions) and this is doing the damage, however to add a latch that will work under all possible power situations is not easy so I will leave that one as an as yet unclutched straw!
In the meantime when the parts eventually drift along to my remote location I will try the snubber first :)
P.S. I once saw a half drawn schematic of that self-operating H-bridge so completed and simulated it, I think I concluded that no matter how you tuned the components it was a wish-and-prayer job and not something I wanted to pursue further, I believe it is used in certain far eastern products :)Edited by zaphod 2018-07-22Cheers Roger 1Kwp DIY PV + Woodburner + Rainwater scavanger :)
zaphod Regular Member Joined: 03/06/2018 Location: United KingdomPosts: 93
Posted: 04:10pm 30 Jul 2018
Copy link to clipboard
Print this post
Well put that theory to bed! After replaceing the failed IGBT and fitting the new snubber it failed after an hours soak testing on the bench, nothing untoward anywhere except blown IGBT's again!! Fortunatly I am off on vacation now for some well deserved rest so I shall just ignore it till after I return and get down the bottom of immediate jobs to do again.
I also discovered my OWON VDS1022I scope self triggers every 5-30 seconds from some internal noise that is not visable on the trace at any timebase with probes shorted or removed so I cannot even leave that in single shot mode to capture any events, what a life!!Cheers Roger 1Kwp DIY PV + Woodburner + Rainwater scavanger :)
zaphod Regular Member Joined: 03/06/2018 Location: United KingdomPosts: 93
Posted: 06:05pm 09 Aug 2018
Copy link to clipboard
Print this post
Being fedup with buying new parts everytime to test a theory I replaced the failed IGBT in one quadrent of the H-bridge with an 800V mosfet but added 3 off 170V TVS diodes in series across it. For some reason this is the quadrent that always fails first! Anyway so far today it has run a record 5 hours without failure. This is not nececeraly proof of something good as the grid and it's noise is highly variable but as the design is not perfectly balanced impedance wise it is possible this quadrent receives more than it's fair share of impulse noise, weirder things have happened before. Of course nobody likes unproven failure analysis but without very expensive equipment this is simply not possible. Hopefully my luck will remain and I can progress to more intensive testing such as breaking and remaking the grid at full power and increase the power level.
I am still unsure if this is a voltage surge problem or a cross conduction problem, I can certainly re-create the latter in simulation by adding in very heavy (maybe unrealisticly so) 25Khz switching noise into the zero-cross detector and an assumption that the processor is capable of missing interrupts (it's all a nasty high impedance opto-coupler circuit so not impossible).
1008 Ran 7 hours yesterday,still alive this morning, tried repeated grid fails, DC-bus fails and Aux supply fails and still works!! Emperically the fix seems to be the TVS's across the sensative quadrant!
1208 Well it's accumalated another day's worth of testing up to the limits of the bench psu ~200W, I added more hardware cross conduction protection and shuffled the mpu pins to allow direct connection to 5V tolerant pins of the zero-cross signals to avoid high value series resistors used previously, all these things are maybies but with a random high voltage fault and a scope that self triggers there is little else I can do beyond speculation :)
Just finalising documentation, conformal coating and bit's n peices before another feild trial attempt...... Edited by zaphod 2018-08-13Cheers Roger 1Kwp DIY PV + Woodburner + Rainwater scavanger :)
zaphod Regular Member Joined: 03/06/2018 Location: United KingdomPosts: 93
Posted: 05:51am 15 Aug 2018
Copy link to clipboard
Print this post
Did a field trial yesterday, electrically everything ran fine BUT the EMC was so bad it completely knocked out the DSL even when the latter was restarted to try and work around the issue so the trial had to be abandoned after 30mins or so :(
The problem is the EMC in the actual installation is very very much worse than it is on the bench I think in part due to the complexity of the installation and wires radiating everywhere including to the water tank heater that is some 15Metres away in the centre of the house quite near the DSL modem that itself is located there to give good WiFi coverage.
The other GTI I use is a belt-n-braces German one with a huge EMC filter on both AC & DC sides and does not affect the DSL at all, so I have now had the bright idea for trial of gutting the German one with the exception of the EMC filters and fitting mine within it's case, that will at least tell me if my electronics are particuarly noisy (I dont think they are) or my filters are simply not man enough (that I think may be the problem).
Whilst all this is going on I lose a bit of energy production as the solar is only used for water heating and the surpluss goes entirely unnused. Cheers Roger 1Kwp DIY PV + Woodburner + Rainwater scavanger :)
zaphod Regular Member Joined: 03/06/2018 Location: United KingdomPosts: 93
Posted: 06:36pm 15 Aug 2018
Copy link to clipboard
Print this post
Worked like a charm that just goes to show EMC is a black magic art! Now just have a crop of issues concerning the shared DC-link voltage to resolve, the booster generates the voltage according to both MPPT and export power trying to maintain 180V. The current source inverter increases power delivery till the buss voltage falls to 174V. The booster has an overvoltage trip around 195V & the CSI an undervoltage trip around 150V.
Well due to the dynamics and reaction times there are a few hiccups to fix, here is a typical overvoltage trip (booster locks out for ~7 seconds):
Here is a CSI UVLO event as it tries to restart (at 5 second intervals) followed by a sucsesful start :
This is an interesting wierd sequence I have yet to fully quantify:
Hope you find this lot entertaining, one major change compared with running the German job is the DC-link reservoir is ~1/10th the value so the dynamics need some tuning :)
zaphod Regular Member Joined: 03/06/2018 Location: United KingdomPosts: 93
Posted: 12:56pm 23 Aug 2018
Copy link to clipboard
Print this post
I didn't have time to do debugging as busy with other things but after a day or so noticed solar system not working, diagnosed to CSI shorting out auxillery 12V supply so removed and replaced with standby unit again. On bench CSI dissasembled and proves to have blown track of pcb where DClink+ feeds drain of half-bridge mosfet pair who'm on inspection are both s/c all 3 terminals so sadly dead! This unit has worked forever on the bench so suspect death caused by persistant overvoltages, that being the bus hitting 190V+ so I should have fixed that problem instead of leaving it installed and unfixed!! The mosfets are IRFB4227's so working pretty much up at the top of there voltage range. May have taken out the IRS2110 driver too as there's definetly a resistive short on its supply rail.
Probaby a moral along the lines of don't start something you dont have time to finish properly, or don't leave something half done etc etc. Backside hurting from own kicking hahaha
PS this also illustrates the problems of trying to use TO220's for high voltage and power, you run out of copper area due to tight clearances and in this case not using the topside due to the heatsink being flush with the pcb.Edited by zaphod 2018-08-24Cheers Roger 1Kwp DIY PV + Woodburner + Rainwater scavanger :)
zaphod Regular Member Joined: 03/06/2018 Location: United KingdomPosts: 93
Posted: 08:36pm 11 Mar 2019
Copy link to clipboard
Print this post
I do apologize I seem to have started another thread (letterbox converter) instead of updating this one! Well the above blow up proved to be very expensive and time consuming indeed as not only the fets were blown but worst of all the IRS2110 driver was destroyed in such a way the high voltage ended up connected to it's vcc that took out all sorts of other ancillary components connected to the same vcc rail (12V) :(
After much investigation I discovered the overcurrent detection was not working as it should combined with occasional sub-harmonic oscillation causing the transformer to saturate hence the severe overcurrent condition.
I was warned about all this by a strange ticking noise coming from the transformer during operation but chose to ignore it on the basis it was handwound so what did I expect! After fixing the sub-harmonic problem it runs soundlessly......Cheers Roger 1Kwp DIY PV + Woodburner + Rainwater scavanger :)
wiseguy Guru Joined: 21/06/2018 Location: AustraliaPosts: 1156
Posted: 03:10pm 12 Mar 2019
Copy link to clipboard
Print this post
Hi Roger, I was recently looking into a strange effect in my saturation tester which uses the IR2113 (same as 2110 but 600v not 500V). I noticed a weird effect as I was increasing the pulse duration (driving a large choke inductor) where suddenly heavy current was drawn from the supply voltage.
The output high side driver was driving the upper Fet on continuously - essentially un triggered - the input was low at this point. On the HO pin the Cro trace was continuously high except an ultra narrow low glitch appeared at the in high transition (Hin pin when it was driving the HO pin on (despite that it was already on)), pulse train was ~ 50% duty at the time. I found an IR application note AN978, top of page 9 describes the issue and cause - VS pin momentarily exceeding 5V below Com pin from inductive effects of load, layout, trace lengths etc.
In my case I cured it by installing a 4R7 between the bootstrap capacitor/VS node and the actual VS node of the Fets and added a UF diode between ChipVS and chip Com (anode to com) to clamp negative transitions at ~ 0.7V, the 4R7 was to minimise peak current through the added resistor. I know this is effectively added gate resistance in series with the FET gate drive circuit but didnt hurt operation - it already turned off too fast.
Maybe not the cause of the ticking you heard but good to be aware of - totally unexpected behaviour. If inductor saturation was your issue it could have caused this? Hope this makes sense - I should be asleep already.....
My circuit is a half bridge flyback circuit where both upper and lower Fets are driven on together with a winding also in series, but for a conventional half bridge converter where the upper Fet is latched on, the implication for driving the lower Fet on sounds scary.
Maybe also explains some inverter failures here, as the IR2110 is used a lot and sometimes connected to the power stage FETs through ribbon cables !Edited by wiseguy 2019-03-14If at first you dont succeed, I suggest you avoid sky diving.... Cheers Mike
zaphod Regular Member Joined: 03/06/2018 Location: United KingdomPosts: 93
Posted: 07:42pm 12 Mar 2019
Copy link to clipboard
Print this post
Hi Mike, Ouch that was a nasty problem you had there! I was aware of it's ability to do that having read the same app note, it also likes to UVLO the top driver too :) I have so many scope shots but I did actually trace the trafo sat to sub-harmonic oscillation and having fixed that the trafo no longer saturates, see the scope shot for the unequal conduction periods, this being a half bridge the results are bad!!.
I am astonished peeps would connect fast fet drivers and there fets with ribbon cable (that implies to me at least significant length) OMG! My layout is very tight as shown in the pic below, driver (U2) and fets (Q1/Q2) within an inch or so!
I think maybe my peak currents are lower than yours so I have not as yet suffered the com/vs problem but it's a very nasty gotcha to watch for. I hope everything is sorted in your saturation tester, blow-ups can be expensive and very time consuming to do the repairs let alone work out what went wrong!
Edited by zaphod 2019-03-14Cheers Roger 1Kwp DIY PV + Woodburner + Rainwater scavanger :)
Notice. New forum software under development. It's going to miss a few functions and look a bit ugly for a while, but I'm working on it full time now as the old forum was too unstable. Couple days, all good. If you notice any issues, please contact me.
Print this post
