99% vw 100% PWM, does it make a difference?

[HighHopes]

there's the current sense right there on the left side, looks like two parallel through hole resistors right above the big DC link cap. you should show pictures of the full board with better lighting and top & bottom. i can zoom in/out as needed.

that layout has to be the worse i've seen. ever. but for 1kw or less it's probably ok

 

[zombiess]

there's the current sense right there on the left side, looks like two parallel through hole resistors right above the big DC link cap. you should show pictures of the full board with better lighting and top & bottom. i can zoom in/out as needed.

that layout has to be the worse i've seen. ever. but for 1kw or less it's probably ok

I can post up better photos. You are looking at the nichrome battery current sense shunts. I can't figure out why the phase appears to limit so much more predictably on these. 1kw... ha... where did you learn how to value engineer. These 12 FET controllers do 6kw 100v @ 60a battery without sweat in the hands of the endless sphere crew.

 

[HighHopes]

i've actually never even heard of the term "value engineering" until recently. lol. all of my designs are .. or begin with .. the intent to be operational for 20years without interruption or need of replacement. this is the requirement of aerospace. furthermore, much of my experience has been with "mission critical" design which means that it is more severe than "life safety" because it matters not if the human is alive, the mission must still succeed. so it takes incredible amount of thought into failure modes and to under stand PCB stress, component stress and aging. i know that china controller can "do 6kw" but i look at it from 20years point of view and say no.. it can not, but it might do 1kW for 20 years (if the components used are certified authentic). so maybe now you can better understand where i am coming from.

i also recognize that my design philosophy does not match the desire of endless-sphere. but i can easily argue that the desires of endless-sphere does not match the design you are currently using (pushing china cheap controllers beyond their capability) and results in one sad customer blaming china controller when it is not their fault because their design is clearly not capable.

the power pass design we worked on for your cadillac is like this, designed for 20 years operation. perhaps it is overkill for your application (and thus not cost competitive except to those you can be educated on its value). but it did make for a good learning experience for you.

perhaps there is a compromise to be had. now the trick is for you to USE that education.. take that knowledge and discover the compromise. re-work a china cheap controller from the ground up with a cost target in mind. find the compromise in the design that optimizes for the application. for example, if using a bootstrap gate driver, then the PwM signals to this chip should be differentially driven. i don't think there is any cost impact for this, maybe $1, so it should be negligible for low production runs. the reason is bootstrap has no isolation so dv/dt of mosfet will - due to parasitic capacitance - inject a current into your digital controller working to corrupt your PWM causing to adjacent mosfets to turn on simultaneously. the probability of this happening is MUCH higher for bootstrap controller as compared to your galvanically isolated gate driver in your Cadillac. think about this and see if you can figure out what i'm talking about and then attempt a design. what other things can you do? you know what needs to be done, but what can you do on the cheap? what can you afford not to do? if you do something, how much $ increase will be acceptable? just the use of IR2100 chip will increase your cost above what china can do becasuse they use level shifting transistors instead which is way cheaper so you already took a cost hit .. was it worth it? what did IR2100 bring you that level shifting transistors would not? think about it...

 

[HighHopes]

here's another idea i was just thinking about while cutting the grass.. (came inside to type this out)

what if you viewed it from a different angle. maybe the question should be "i want to use the 1kw rated china controller as is cause forsure that design is the cheapest, but must avoid the blow up".

then, do a survey to discover what are the typical failures. polling endless-sphere it looks like it is lowside mosfet blow for unknown reason. statistically it should be more random than that because the application is symetrical for 4-quadrant controller but we see it is the low side mosfet that goes.. obviously it is more stressed than the uppers. so solve that problem to make the stress balanced. then, realize that the stress is too much for all mosfets and solve that problem. problems to be solved in "value engineering" manor.

what about this type of approach?

maybe you only need 2-quadrant design since nobody wants to drive their ebi-cycle/e-motorbike backwards anyway?

 

[speedmd]

Exactly! Find biggest factor /condition effecting design robustness. Solve it /make it bullet proof, and so on, identify next factor. "Sometimes, it is just easier doing it the hard way".

 

[Thud]

My point of view:
Its important to identify the motors being driven in these discusions.
The whole 99% thing became aware to me through a mesage frome member Bertie.

It was specific advice regarding XieChang controllers along with the tiny, ultra low inductance motors like the C80-100 outrunner's or small colossus. (We are talking 1/4 of the resistance & inductance of the hottest wound hub motor)

More important is keeping the block time limit at minimums (or zero with with XPD software) if spinning a high rpm motor.

I blew 3 controllers on the bench with unloaded motors messing with 120% settings...following Burties advice solved my controll issues.

After that, the cheapo controllers seem to perform just fine when kept inside their intended operational envelope, Ive learend a ton playing with these.....mostly to keep realistic expectations :lol: & take everyones claimed outputs with a grain of salt....power in is not power out.

 

[HighHopes]

that's probably good advice, but we want to take it a bit further. to understand the WHY and the HOW. what is it about low inductance machine that requires 99%? or is it some other reason that requires 99% and low inductance machines just makes the symptom worse. would 98% be better?

from a user's point of view these questions are irrelevant because after the change was made, it "worked" so why do you care? but from a designer's point of view these questions are everything.

 

[Thud]

Big moose enlightened us on this particular weakness regarding the Xiechang controllers.
Here is snippet he laid on this e-tarded bike racer after a session of popping controllers:

Ok... I am interested in what other experts would say, so feel free to opine and agree/disagree with my opinion.

The fact that it was the bottom FET that failed, tells me a lot. Now, I don't have the controller in hand with my scope and test tools, so this is a bit of conjecture.

The usual practice is to switch the phase leg onto the bus with the top fet to the rail, so it turns on and stays on. Phase current is measured and the bottom FET is PWM'd to keep the currents in check. Now with the large outrunners the inductance is very low, so the current builds "quickly" according to dI=Veffective * dt/L So dI can get very very large in a short time. If the controller is designed with a fixed cycle time in the do loop and L gets too small, the physical currents can outrun the controllers timing to control them.

That is we have a loop that executes over a microsecond to measure current with ADC, calculate stuff, then keep lower FET on or turn off lower FET that works with a range of L values. Now if we cut L by a factor of 10 or 50 but don't speed up the execution time in the loop; we get internal punch thru on the FET or a lead bond failure. If your FETs failed open, perhaps it was a lead bond failure to the die. It would be interesting to decap a blown FET for inspection.

For very, very low inductance motors, the current cutback must be interrupt driven and not a polled function in the control loop of the micro.

Just some BigMoose Musings(tm)

I am the wrong person to talk tech talk on the electronics, but it is my understanding that the xiechang's do NOT directly measure phase current...i have only conjecture on the math they use to estimate the phase current. The popular theory is its a simple multiplication of the battery current read at the shunt.

I have perpetrated the 99% speed & zero block time to any & all who attempt to set up the small out runners with the Chinese e-bike controllers...i have had too many Pm's from guys who insta-destroyed a brand new lyen controller....with a modded shunt & other voodoo that works wonders on a DD hubbie.....but it totally wrong for the lil guy's(with very sharp teeth & short tempers )

 

[zombiess]

Highhopes, you pose an interesting idea. From the majority of blown controller posts I have seen and the 4 to 5 blown controllers I have examined I have seen only 1 with the low side gets blow up. This was my personal 18fet test controller that I ran at 75v 180a phase, 150a battery on my low L hub motor with 0.0 block time and 99% limiting. It worked for one hard launch, a few feet of part throttle and then failed on the next hard acceleration. These controllers both old and new only pwm the high side. If the RC motor users at failing the low side gets More often, that would be interesting to me. The one major variable is that I mat h the gets that go into my controllers and I only know of Steveo who has built one of these up using matched gets I donated to him to see if he could replicate my results on a motor that was blowing up his 36 fet controller regularly. So far he is good with a 24 fet using the gets I sent. I should start a poll and see which side people blow the most often and on which type of motor. My guess is high side due to the extra stress of switching.
Btw when we talk about block time on these controllers we are talking about a setting that allows the controller to exceed its set battery limits for a certain amount of time. This feature is there because the controllers are designed for low power and hub motors. Sometimes this combination can cause an issue of tripping the crude protection circuit because of starting on a hill.

 

[John in CR]

Which is high side and which is low side? I tried searching but no help. I've got a box of blown ones to check. All done in by low L motors.

 

[Arlo1]

Which is high side and which is low side? I tried searching but no help. I've got a box of blown ones to check. All done in by low L motors.

Think low side = low voltage so its - .....

 

[zombiess]

Which is high side and which is low side? I tried searching but no help. I've got a box of blown ones to check. All done in by low L motors.

High side of the phase FETs has the drain (middle pin) go to the B+ and the source go to the phase output. Low side has the drain go from the phase out and the source go to the B-

It's really easy to figure out the groups by looking at what the middle FET leg connects to on the battery side, if it goes to B+ it's high side, if it's connected to the phase out, then it's low side.

 

[Njay]

It's still quite hard for me to think brushless, but I think you may be seeing it wrong, zombies. If the controllers don't do synchronous rectification (you say they only PWM the high side), it is the PWM'ed transistor's bottom partner that recirculates the current when the top is OFF; therefore, it re-circulates it through its body diode. So, it's the bottom FET that suffers (a lot!!!). And it is very "bad" to do that when compared to doing synch rectification.

I have started by not doing sync. rectf. in my learning controller and then at a certain moment I implemented the sync, and the difference in MOSFET temperature at the same conditions between the 2 methods is colossal (on the transistor doing the recirculation). Not only the dissipated power is higher due to higher Vfw of the diode but also that transistor is the one that gets current for longer time when the duty cycle is very small - with the time constant on the ms as Lewboski mentions, once the current reaches very high values (low inductance motor) which are "driven" by the small duty cycle ON phase, the OFF phase, where the current re-circulates through (as I think) the bottom MOSFET, it stays at high values for possibly the entire OFF phase - higher voltage drop (Vfw) and for longer (than the top MOSFET).

 

[zombiess]

Njay, you are probably correct, I don't know motor/controller operation that well. Don't both sides experience freewheeling diode losses?

 

[Arlo1]

It's still quite hard for me to think brushless, but I think you may be seeing it wrong, zombies. If the controllers don't do synchronous rectification (you say they only PWM the high side), it is the PWM'ed transistor's bottom partner that recirculates the current when the top is OFF; therefore, it re-circulates it through its body diode. So, it's the bottom FET that suffers (a lot!!!). And it is very "bad" to do that when compared to doing synch rectification.

Yup these controllers are using about the oldest tech of BLDC control....

But remember the Low side that needs to have the Diode conduct is off its a different low side that is on.

 

[Njay]

Njay, you are probably correct, I don't know motor/controller operation that well. Don't both sides experience freewheeling diode losses?

During the switching transitions yes, but what I'm talking about is that the bottom guys have that experience also during the entire OFF part of the PWM cycle.

But remember the Low side that needs to have the Diode conduct is off its a different low side that is on.

Yes, I have that in mind.

 

[HighHopes]

NJay, it's like you are saying "controller failed due to over temperature of the lower mosfet" and your solution is to "lower the power dissipation of the lower mosfet by control method of synchonous rectification".

synchronous rectification, for those that don't know, is when current is flowing through the msofet body diode and you turn the mosfet ON during that time so the current will chose the path of least resistance flow through the mosfet die instead (backwards so to speak). ** IF ** it happens that the losses are lower for mosfet conduction then it can lower the power dissipation so that's good.

so firstly, you have to know for sure that this is why controllers are failing and secondly this solution on works with 6-step/trapezoidal commutation, which as it happens, is exactly the method used by china controllers. that first part is hard to do actually because your china controllers do not trend data, do not have fault annunciation (a fire does not count), it does not detect nor distinguish different types of faults (under voltage, over frequency, over temperature, short-circuit, over current, etc. etc.). so it makes discovering the nature of the fault difficult. so this worked in your case, and the theory could be correct so i like this idea as a potential universal solution (would take a bit more analysis of 6step commutation to make sure there is sufficient cool of time). there is still more you can do, but this is a good first step and specificially for solving problem of over temperature failure of lower mosfet. starting a thread on failure modes would be very good but i think you willl have difficulty collecting useful data, probably it will all be anecdotal.

for a more modern/advanced sinusoidal this trick would not work because the mosfet both upper & lower are ON & OFF the same way over a cycle, so therer are no long periods of diode conduction .. it is perfectly symmetrical design. in theory you could turn on the mosfet when diode is conducting but in the next pulse the mosfet would be conducting in the normal direction and so it would never get a cool down period and likely explode. also with sinusoidal inverters all 3 phases are active each pwm period so you must be blocking voltage with the diode.. if you turned ON a mosfet for synchronous rectification at each time the diode would have been conducting you would short out the drive soon enough.

ps. the upper mosfet does not necessarily take more stress just because it is switching. keep in mind that it takes simultaneously less stress because it does not pass conduction losses 100% of the time. conduction losses + switching losses + leakage = total mosfet power dissipation averaged over one full electrical cycle. so it depends on the datasheet curves for both mosfet & diode and depends on the algorithm to control. for sinusoidal inverter you want to have switching losses = conduction losses (ideal). for 6step maybe a different ratio is appropriate ... or ... you selectively decide when to apply the synchronous rectification to ensure the diode die losses is the same as the mosfet die (probably this means that you most offten decide to allow the mosfet to pass the current, but perhaps not all the time as when the mosfet gets hot so you decide to let the diode take some heat to better balance). ... just thinking out loud.

 

[Njay]

NJay, it's like you are saying "controller failed due to over temperature of the lower mosfet" and your solution is to "lower the power dissipation of the lower mosfet by control method of synchonous rectification".

It's not . It's like I'm saying it could be it or that it could be a relevant contributing reason. I posted mainly because I think zombieSS was seeing it the other way around (the top heating instead of the bottom). Anyways, I think that, in the context of trying to "save" this kind of controller (with minimal effort), it's pretty useless to know that the rectification method is the main failure reason (assuming it is) unless someone gets his hands on the controller's source code, or has a tool chain for it and the time/knowledge to re-write the code from scratch.

I don't have one such controllers nor motor, but I agree that there's lots of information missing. For example, I haven't see a scope shot of the rail near the FETs, we don't know "how bad" the layout actually is yet. Gate - source shots would be interesting to see too, as across the shunt resistors.

I have 2 things in the back of my head that I would like to know and are related to rectification/body diodes:

- At least some Altraxx DC motor controllers (I don't know if all) apparently do async rectification and have diodes and not MOSFETs on half of the half-bridge; historical, technical or economical reasons?

- When one matches MOSFETs for turn ON, do body diode curves also get matched?

 

[Arlo1]

What if we use the signals from the hi sides and fed an inverter like the prius where is only three wires run to the inverter for gate signals. We could have the low side on untill it gets a gate signal for the hi to be on and build a delay in the inverter.

 

[Njay]

Something can be done but not just a simple inverter. The controller still needs to control that MOSFET when not PWM'ing the top and a dead time must be ensured, times 3 + cut and solder a few traces. Not my concept of minimal effort though.

 

[Arlo1]

Yeah I think I am mostly done with these controllers I will use them for toys. But when I need real power and reliability I will continue to work on my own design. I can't wait to get back at it. Its to bad Im all alone working on this here but its sure awesome I have you guys online to help me through it!

 

[fechter]

I like Njay's diagram. That shows what's going on.

I remember wondering about those 9 FET versions where they paralleled the bottom FETs but not the top ones. Eventually I figured out this was an attempt to even out the dissipation a bit. The forward voltage drop of the body diode x the current puts the thing in danger of getting outside the safe operation zone and would be way more than the dissipation of the same FET turned on.

They do make really nice off-the-shelf gate drivers intended for synchronous operation that have built in dead time to prevent shoot through, etc. From a manufacturing standpoint, I just don't see how this adds so much to the cost that they don't do it. Well, there's lots of things about the design of those Chinese things I don't understand.

Adapting this in to an existing controller would be challenging, but not that bad. It might take a little flying daughter board to hold the gate driver and get it as close as possible to the gates.

 

[HighHopes]

NJay, no problems i'm just talking things out here. interesting question about matching diodes.. i don't know the answer, probably only zombies knows that.

Yeah I think I am mostly done with these controllers I will use them for toys.

exactly, the design it is clear they are meant as throw away items good for a toy but not good for EV. EV application requires more reliability.. you don't want to get stranded on the road!

if you all keep at it, i'm sure a good compromise design will emerge. that is the beauty of the internet and chat forumns like this after all.

 

[zombiess]

NJay, no problems i'm just talking things out here. interesting question about matching diodes.. i don't know the answer, probably only zombies knows that.

Yeah I think I am mostly done with these controllers I will use them for toys.

exactly, the design it is clear they are meant as throw away items good for a toy but not good for EV. EV application requires more reliability.. you don't want to get stranded on the road!

if you all keep at it, i'm sure a good compromise design will emerge. that is the beauty of the internet and chat forumns like this after all.

Good question on the diodes is right. I don't know the answer as I never thought about matching that parameter. Not even sure how I would do something like that or if it would even be required. How much variance do most diodes have?

When I calculate out diode losses on the switching side assuming a 1.3V drop I'm seeing 64W of heat at 1% PWM and 0.65W at 99%. I'm not sure how to calculate the freewheeling losses on the other bank.

 

[HighHopes]

just to belabor the point ...

@Lowboski - Infineon Hybridpack for EV application uses a 2nd order filter, with a whopping 70uS group delay. not sure i agree with this approach but they don't seem to mind. i've attached the bode plot so you can see their frequency response. FYI, they expect their current sensor to produce 5V full scale at phase current sense input, but i am simualing with 1Vac so you can see the transfer function for what it is. nice flat gain up to about 1.5kHz which is perfect and then sharp attentuation for switching frequency and higher frequency noise due to rise/fall of mosfet which is >100khz, so these parts are nice. but the group delay... yikes.