Calculating controller power input capacitance

[zombiess]

Thank you for posting this HighHopes.

Is there any type of Polypropylene capacitor that we should stay away from? I know there are a few types including metalized.

 

[Teh Stork]

I'd like to add that low voltage film (any type) are quite uncommon. So for a 300VDC inverter, film based caps are a good choice; I agree. For ebikes running at around 100V, I see no other combination coming close to the performance from electrolytic together with ceramic caps. The small ceramic caps (with their low ESR/ESL) keep the bus stiff while switching, the electrolytic caps deliver the bulk capacitance.

 

[Njay]

HighHopes, thank you very much for attending my request.

The phase inductance is also fixed because your motor is pre-selected (how did you get 2.64uH?, that is incredibly low value).

It's not a phase inductance. It's a battery <-> controller wiring (parasitic) inductance. The "phase inductance" (I actually have a DC motor, but let's call it "phase" as is of the most interest to the forum) I need is around ~39uH and the vast majority of bicycle hub motors have it higher; but for instance there's colossus (the one our friend Arlin is using) which has 8uH.

Is the expression also valid for DC motors? I see it's from v = Ldi/dt scaled by (D-D^2)^-1 , but I don't know more. I'll look it up later somewhere if it's some well-known thing.

 

[HighHopes]

I see no other combination coming close to the performance from electrolytic together with ceramic caps.

teh stork - could be you are right as my description of method was the proper way (for approximate sizing), not the most common way. the method is geared towards getting away from what you see common out there to more analytical but simple method that should lead to decision to use polypropelne without fear of being too expensive.

Is there any type of Polypropylene capacitor that we should stay away from? I know there are a few types including metalized.

zombiess - what makes you think metalized polypropylene capacitors are bad for DC link application in VSI topology?

Is the expression also valid for DC motors?

Njay - yes, still valid. it is for any PWM voltage source inverter driving a machine with Bemf.

I see it's from v = Ldi/dt scaled by (D-D^2)^-1

its not magic anymore if i have to derive the equations. that i need to check my lab notes to find the papers this equation came from.

Ripple Current:
1. delta_I = d*(1-d)*Vbatteri/(f*L)
how is this ?
start with common inductor equation: V = L*di/dt & integrate both sides & rearrange to get:
2. delta_I = VL*delta_t/L : VL = votlage across motor phase inductor
more discussion,
VL is when the upper mosfet turns ON it puts Vbattery across the phase inductor. but because its a motor with PWM it already had a voltage fighting the applied voltage, the motor's bEMF. so then VL = Vbattery - VbEMF. but what is bEMF voltage? it would be tempting to get crazy with the cheese wiz with more equations but we are only making a rough dimension of the capacitor so i think it is good enough to take an approximation. when the mosfet is OFF the voltage does not suddenly drop to zero, it is roughly maintained due to motor continue to spin and the voltage is roughly what the applied was so we can use bEMF = d*Vbattery. so VL = Vbattery - d*Vbattery. I should put a side note here that we could make a worse case assessment and say that the bEMF (d*Vbattery) is worse case equal to zero, i.e. when the motor is at stand still. but maybe that is a bit excessive if we think that the period of time that this occurs is small in comparison to period of time when motor is running. but perhaps not for EV application.. start/stop..start/stop.. hmm... anyway, for now i will leave it as bEMF = d*Vbattery and just accept that the capacitor is stressed when motor is starting up.

3. delta_I = (Vbattery - d*Vbattery) * delta_t/L : delta_t = mosfet ON time = d*PWM_time = d*1/PWM_frequency = d/f
4. delta_I = (Vbattery - d*Vbattery) * d/f * 1/L
5. delta_I = Vbattery*(1-d)*d/(f*L) : here you have to plot the formula of (1-d)*d to learn that peak is at d = 0.5
6. delta_I = (1-0.5)*0.5*Vbattery/(f*L)
7. delta_I = 0.25*Vbattery/(f*L)

 

[zombiess]

Is there any type of Polypropylene capacitor that we should stay away from? I know there are a few types including metalized.

zombiess - what makes you think metalized polypropylene capacitors are bad for DC link application in VSI topology?

Nothing makes me think metalized is bad, I am just asking if it is or not since I have seen there are a few types. I literally have no idea if there is a difference in the electrical characteristics. I know from past reading that if I am dealing with high ripple currents are trying to absorb a fast transient I would use polypropylene caps.

Now I have the math to calculate how much I need, thanks

 

[HighHopes]

ah. i misunderstood your question. i don't know the answer, only used metalized polypropelene film for >1kw application when purchasing COTS (commercial off the shelf) which frankly is rare in areospace but getting more common.

the snubber caps that go near the mosfet phase leg(s) are selected with different criteria in mind. the technology and construction method of cap are selected based on their pulsed voltage spike handling capability. value is chosen to compensate the phase leg inductance including busbar & DC Link inductance. for 1kW or less and good layout with quality parts you don't even need snubber caps.

 

[HighHopes]

to answer a question, here is my worksheet using realistic numbers and has a bit more info on how the math is done and capacitor RMS current is selected from the datasheet.

You'll notice I don't even concern myself with ESR value of cap, it is sort of taken into consideration indirectly when looking at the I_Arms rating. Only important thing to consider is to never select an electrolytic capacitor that does not have a 10kHz rating number specified. But of course this is not even worth mentioning anymore because I have convinced you to go to Polypropylene right? :wink:

 

[speedmd]

Did some life testing on capacitors some years back. Polypro film caps are in a different league all together. Much much better. Problem most times with them is they are 4 - 10 times larger than electrolytics. No comparison in real life performance if you can fit them.

 

[HighHopes]

speedmd, that's a fair point. if your design priority is for volume of space rather then electrolytic will be better choice BUT the performance will suffer.

if i take a 100uF electrolytic cap 500V rated and compare that to a 100uF polypropylene cap 500V rated yes the poly is much bigger. but the performance is not the same (as you mentioned), which means that the electroltyic cap will explode because ripple current is too high; but yes, it is smaller. if you do not want it to explode (obviously) then you have to change the inverter specification when using electrolytic.. lower the bus voltage, increase the PWM frequency and/or increase phase inductance. so, smaller cap yes but also comes with different inverter. <-- this is based on same reliability or same inverter life expectancy in the comparison.

 

[Njay]

You'll notice I don't even concern myself with ESR value of cap, it is sort of taken into consideration indirectly when looking at the I_Arms rating. Only important thing to consider is to never select an electrolytic capacitor that does not have a 10kHz rating number specified. But of course this is not even worth mentioning anymore because I have convinced you to go to Polypropylene right? :wink:

Let me do my math, then I'll let you know

Thanks for the trouble of writing all of this. Looks much easier than I thought .

 

[zombiess]

I have convinced you to go to Polypropylene right? :wink:

You make a very good sales pitch for it. Getting rid of the tall electrolytic caps can make packaging a bit easier. This is something I've had to be very conscious of when selecting electrolytic caps.

Maybe a combo approach could work well for cost / packaging... mainly packaging. How would one go about calculating the math for that? Would you first calculate the polypropylene cap, then go back and make up the difference with the electrolytic caps?

BTW, your calcs and reasoning are going to make me go back and look at my cheap design I showed you and see how it works out by eliminating the electrolytics and replacing them with good caps. I don't like the idea of using the electrolytic caps on the design the way it is now because I have to lay them over due to packaging issues... that means long leads and we all know that is bad. I've seen several pics and at least 2 physical controllers where the electrolytic cap lead was melted off from them being laid on their side.

 

[HighHopes]

How would one go about calculating the math for that?

i won't answer the question (purposely), but i will give you some insight.

in aerospace, especially military, usually the priority of design is, relability, weight, volume, cost. reliability is highest even above life safety because mission critical level means mission succeeds with .. or without .. the pilot. but don't lose sight that weight & volume are # 2 & 3. so there is a LOT of effort that goes into estimating a footprint and power density. actually the preliminary design cycle (design on paper) is almost as long as the actual design cycle! for budgetary price you need 20% accuracy which is actually hard to do unless you have similar products to estimate from. for preliminary design your analysis has to be within 10% of the final product in terms of weight, volume & cost. so the math is very accurate.. has to be .. for results within 5% and then the tolerance +/- 5% gives you worse case 10% accurate (or bang on!). so the math for estimation is complex, thorough. it is differential calculus with focus on optimization algorithm for various inputs & variables. you remember the dissipation equations for a swithcing power mosfet i showed you? my lab notes are actually 37 pages long, just to calcualte the losses in the switch! mind you, it has graphs & pictures and explinations. i have publish papers on that subject alone. the DC link capacitor analysis is 2 pages + another 5 for inclusion of other elements that affect sizing, dissipation factor, end of life. it goes on and on..

 

[zombiess]

you remember the dissipation equations for a swithcing power mosfet i showed you? my lab notes are actually 37 pages long, just to calcualte the losses in the switch! mind you, it has graphs & pictures and explinations. i have publish papers on that subject alone. the DC link capacitor analysis is 2 pages + another 5 for inclusion of other elements that affect sizing, dissipation factor, end of life. it goes on and on..

Is it sad or awesome that I drool over the thought of being able to read that kind of information? I never thought I would ever be this interested in one aspect of electronics. There is something special about gate driver / inverter design that is attractive to me, but I'm not sure what it is.

I'm such a nerd it's scary. I was in our public health lab today replacing a printer and was talking to some of the microbiologists, one really cute one in particular She was showing me some cultures of Candida Albicans (yeast that causes thrush) while I was working and then a mystery yeast she had yet to identify when another tech asked me if I wanted to see rabies. I jumped at the chance and looked through the stereoscopic microscope at 1000x examining the brain tissue under fluorescence. Pretty awesome stuff, they even explained to me how they use the antigens to create the sample slides. I love nerding out on science and engineering. I'm beginning to think learning is my biggest hobby because I am full of all kinds of random info. So much so that people actually think I am smarter than I am LOL.

If a zip file full of scanned notes and formulas with explanations shows up in my PM inbox...

 

[Njay]

(...) for budgetary price you need 20% accuracy which is actually hard to do unless you have similar products to estimate from. for preliminary design your analysis has to be within 10% of the final product in terms of weight, volume & cost. so the math is very accurate.. has to be .. for results within 5% and then the tolerance +/- 5% gives you worse case 10% accurate (or bang on!). so the math for estimation is complex, thorough. it is differential calculus with focus on optimization algorithm for various inputs & variables. you remember the dissipation equations for a swithcing power mosfet i showed you? (...)

Do you characterize the devices yourself or rely solely on the datasheet info?

 

[speedmd]

Good points on accuracy of the caps. A few of the other things I remember on the testing of the polypro caps was they did often loose capacitance over time / use. Metalized film caps are much more robust than electrolytics going by what I saw in life testing. If a minor leak/ high resistance short develops on a electrolytic, it will most likely always fail dramatically in a short period of time destroying anything in its path. The polypro caps, will at reasonable power levels most times self heel by burning away the metalization/ connection surrounding the short and at times even taking out of service a round section several layers thick that could be deep inside the wrap of film. They isolate insulate and seal it off from the shorted area and keep on working like nothing happened, but at a reduced capacitance. Only when this gets too sudden/ too hot does it go into full melt down mode, which is very rare from my experience unless you are really overdoing things. I was overdoing things regularly 25 -50% of ratings in testing for days on end. Long ago, but I can still smell those occasional totally melted down units.

 

[HighHopes]

Do you characterize the devices yourself or rely solely on the datasheet info?

usually i am happy with using only the data sheet for more common stuff. for example, op-amp circuit as bessel filter. i select the parts, make the design, and then monte-carlo the input to output to see how the variation is due to variation (a 1% resistor is more like 2% when you consider temperature, aging etc.). and then sensitivity analysis to find out what component is driving most of the variation and then go spend more $ to get better quality part (maybe one cap should be C0G dielectric, or one resistor should be 0.1% tolerance or, 1% tolerance if fine but oversized at 2512 in order to not be so influenced by self heating). there are a million tricks.

for some critical aspects, or things that cost $, i would go into detail for design of the part itself with full characterisation on the bench to prove concept. i think i did LEM's very first current sensor rated -40C to +125C for them beacuse they had no such offer in their catalog at the time. the design was not that hard but man was it difficult to find an encapsulation material that could seal the electronics (important in low atmosphere environment) and NOT have minor crack where it contacted the plastic. IGBT 6-pack module, capacitors, inductor, i have characterised completely and designed electrical & mechanical. worked with the various manufacturers for each technology to show them how i wanted it built (and being told "it won't work" just to find out later they started issuing patents like crazy when it was shown to work). it puts a smile on my face to see the products are still shown on the first page of their website as advertisement. i had one of the very first SIC modules (Si IGBT + SiC diode) 600V, 600A and built a hell of an inverter at power density or efficiency unheard of. one of those modules cost more than my car, the inverter almost my house. IGBT modules very particular about how it was built inside the module. how it was bonded, ribbon, thermal path to baseplate, die spacing, wafer selection, hermetic seal.

for the inverter i am doing now (combining Infineon's Hybridpack with Texas Instrument's Instaspin-FOC with custom wound low voltage induction motor) is just by the datasheet with no characterisation and extremely limited test facility i have in the basement (1970's oscilliscope is all i have). but i'm not discouraged, our for-fathers in electrical engineering accomplished far more with WAY less. plus i have all of you to bounce my problems off and brainstorm a solution

 

[Njay]

So it looks like you ride the top of the wave . It's pretty nice that guys like you, Lebowsky or BigMoose come down to the beach once in a while to give some lessons to the rest of us who like to watch but would like to ride the big waves too

I'm still chewing this cap dimensioning thing. Lots of info and doubts floating around, need some time to let it settle down, will end up with a spreadsheet with 2 sheets, one for the "bulk" capacitor and another for the "snubber" capacitor. A "Power Unit" PCB for a single TO247 half-bridge is being designed for some time now, constantly redoing things, and still need to re-do the bulk capacitance part. Never something so small gave me so much work!

 

[HighHopes]

well, don't get too lost in the details. i was just giving you a picture of how it is done in professional way, but it is not always necessary to work that way. for DIY project there are really no specs to meet other than basically functional.. so no need to go to the nth degree. i guess my point is that it can be done but just take your time to think about things, learn something, apply it. take your time.. no rush. there's not been an open source quality inverter posted to the net in its full glory in a long long time (OSMC the first one? has there even been a second in the past decade?)

 

[HighHopes]

for snubber cap, i think the theory is clear that it is there to balance the stray inductance. but for me i always chose the value "by feel" and didn't really use math for that. i had no snubbers for <1kW drive, and 2.2uF for 50kW continuous drive. look at the physical packaging and DC link cap quality to determine how much snubber i needed and then just picked value out of the sky. i used to calculate it, but virtually impossible to determine what the stray inductance really is, so i gave up that method.

only thing that is important for snubber cap is that it must be rated for very high pulse voltage application. values tend to be between not needed and 2.2uF when good layout (physical assembly) and quality parts.

 

[Njay]

I'm trying not to get lost in the details, but I want to have some math behind, preferable simple and conservative within reason. For example, for calculation of MOSFET switching power dissipation for inductive load I use Pd = (tr + tf) * 0.5 * Vbat * Id * fpwm , which is an equation that overestimates losses but it's simple and easy to understand (KISS KISS KISS). I want to build a reliable controller; I plan to ride a motorcycle with a DC motor, and I don't want a MOSFET failing short (as they like to) and have full power suddenly applied to the wheel, or sudden rear wheel brake. I know I must make it fail safe through other means, but my philosophy to always start with a sound design.

The guidance I gathered so far for the snubber cap design is:

1) Minimum RMS ripple current
Assume the capacitor provides all the current during the MOSFET switching time, so
[pre]Iripple > Imax * (tr + tf) * fpwm[/pre] (1st written by rithee5 above in this topic)
Example: Imax = 60A (peak, not continuous), Vbat = 43.2V (almost discharged 16s LiFePO4), switching time tr = tf = 500ns, switching frequency = 16KHz -> Iripple -> 1Arms

2) Minimum capacitance
2.a) The capacitor must take/provide half (triangle area) the max current during the switching time while keeping the rail at +- a percentage w of Vbat, so C = i * dt / dv -> [pre]C > Imax/2 * tr / (Vbat * w)[/pre]
This gives me values much higher than it seems needed. Example: Imax = 60A, Vbat = 43.2V, 5% max Vbat ripple and 500ns switching time -> 7uF.

2.b) Should store enough charge as to provide Irr while keeping the rail at +- a percentage j of Vbat, so Q = CV -> [pre]C >> Qrr_diode / (Vbat * j)[/pre]
Example: IRFP4110: max 210nC Qrr @ 75A 85V 100A/us 125ºC, Vbat 43.2V, j = 5% => C >> 0.1uF; x 10 -> 1uF + 7uF from 2.a) = 8uF

3) Maximum ESL
As small as possible

for snubber cap, i think the theory is clear that it is there to balance the stray inductance.

It seems that Irr through the parasitic inductance (layout + caps) is the main problem.

but for me i always chose the value "by feel" and didn't really use math for that. i had no snubbers for <1kW drive, and 2.2uF for 50kW continuous drive.

With what kind of MOSFET switching time?

 

[HighHopes]

its like we're sit'n at a pub having a couple of BEvERages, discussing our favourite passtimes

we can agree to disagree on your equations

The guidance I gathered so far for the snubber cap design is:

i think you need to have clear in your mind what the difference is between the "DC Link" capacitor and the "Snubber" capacitor and then taylor your equations to suit.

in my opinion, the DC link capacitor is there for balance of energy and the snubber ccapacitor is there to compensate for stray inductance.

if you want to learn more about snubber caps, this document will help http://www.semikron.com/skcompub/en/AN-7006_IGBT_PeakVoltage_Snubber_juli_2011.pdf

ps. i would use IGBT with DC motor. my first project was DC motor and mosfets were used and it was fine. that was for personal project with others online, not professional product. but.. ya, i get worried about their failure mode. IGBTs tend to fail open.

With what kind of MOSFET switching time?

in my professional life i only ever used IGBTs because they are considered more rugged and they have better failure mode and also you can control better the parameters when you design at the die level. so technically.. mosfet switching times i can not say. IGBT switching times (basically the same question), i have done 50ns switching (that was crazy) up to about 800ns. 270Vdc up to 540Vdc. 1kW up to 50kw or there abouts. they were all a little different. i do not design for a target switching speed, i design (or lab bench determine), power dissipation vs. noise generated and when i find optimal setting then i look to see what the switching time is. i know ahead of time only the range i will be in, so i have some idea what to expect.. but actual switching rise/fall time of mosfet is not a design parameter for me per say.

 

[Njay]

I don't drink beer, but sure like the discussing .

The guidance I gathered so far for the snubber cap design is:

i think you need to have clear in your mind what the difference is between the "DC Link" capacitor and the "Snubber" capacitor and then taylor your equations to suit.
(...)

Yes, I do have it clear. I was really speaking of the snubber capacitor in my last post. The little one which is there just to solve the parasitic inductance issues caused by not having an "entirely good and well positioned" bulk capacitor.

ps. i would use IGBT with DC motor. my first project was DC motor and mosfets were used and it was fine. that was for personal project with others online, not professional product. but.. ya, i get worried about their failure mode. IGBTs tend to fail open.

I'm a bit worried now too, 'cause I haven't thought about the top MOS failing being able to cause trouble too. If the bottom fails you can open the system's contactor in a short time. If the top one fails... I always wondered why Altraxx had a diode as the top switch. Humm... something to think about.

With what kind of MOSFET switching time?

in my professional life i only ever used IGBTs because they are considered more rugged and they have better failure mode and also you can control better the parameters when you design at the die level. so technically.. mosfet switching times i can not say. IGBT switching times (basically the same question), i have done 50ns switching (that was crazy) up to about 800ns. 270Vdc up to 540Vdc. 1kW up to 50kw or there abouts. they were all a little different. i do not design for a target switching speed, i design (or lab bench determine), power dissipation vs. noise generated and when i find optimal setting then i look to see what the switching time is. i know ahead of time only the range i will be in, so i have some idea what to expect.. but actual switching rise/fall time of mosfet is not a design parameter for me per say.

I understand. My 500ns switching time is actually derived from switching power dissipation requirements. It's just that I find it more interesting to speak about switching times, because the problems they cause seem more hard to solve (di/dt) than thermal.[/quote]

One question: with all the focus you have to put on reliability and fail-safe, professionally, have you seen field failures and what cause them?

 

[HighHopes]

field failures of course. they are big industry thing "plane on ground" event. thankfully it is rare.

during the original product design cycle there are a lot of checks & balances to avoid field failure. design reviews, preliminary, critical reviews. lots of people invovled with lots of experience. qualification of the product is crazy with how the product is poked/prodded/stressed to see if it will fail.

generally the aerospace design use 20yr old techniques & methodology in order to benefit "lessons learned" thus avoiding the more common pitfalls. for example we use sine-pwm and/or space vector modulation only for generation of 3-phase sinusoid.. not chaos controller (non-linear) because they haven't been in industry for 20 years yet. needless to say there is a lot of tribal knowledge in this field as people tend to hold on to tried & true designs that are proven. but on the other hand, practically nothing is boiler plate app-note design. a great deal is custom designed from scratch because commercial stuff and generic info in app-notes will never meet design requirements which are strict and incredibly thorough (systems engineers have 10+ years of design experience first so they know exactly how to specify a product). for designs then you have to be creative, but always reliable.

reliability is deeply analyzed too. each component considered. you do not get the wide digikey selection of parts, no, a component specialist gives you a list of "approved" parts. you can use this series of resistors, this series of op-amps, this supplier of current sense, opto-couplers, igbt, diode. everything. then in the design each subsytem is analyzed "if this signal fails Open/Closed/Indeterminate, what will the outcome be and how will the system be protected from failure".

so for example, take your motor cycle with DC motor. if your mosfet fails closed you will get DC current into machine which will cause high torque. this is a failure. how will you detect this occurance? what will you do about it? then.. the circuit that does this detection.. ask the same question! what if your failure detection circuit fails Open/Closed/Indeterminate. ha! we do not normally consider as if two subsystems fail simultaneously (double failure) as it is usually considered too expensive to design for that condition (though sometimes you have to because spec says to do so).

 

[Arlo1]

HighHopes can you help walk me through the cap calculations....

Ok so my numbers 8uH motor phase to phase & Maybe add 2uH for wire from the controller to the motor wires... 118v fully charged battery.

So you calculate for amps

>>DC Link ‐ Ripple Current:
i=d*(1-d)*Vbat/f*L

i=0.5*(1-0.5)*  72V/20khz*50uH
i=18App or 13 Arms

So for me......
i=0.5*(1-0.5)* 118/22khz*10uH
i=134 App or 96.7777 Arms

So this alone seems hard but maybe if I up the PWM and find the caps. What about calculating the total farads???

 

[Arlo1]

I could up the PWM to 160 kHz to get close to the same as your required amps...
Its time I learn and understand this thread.... Maybe this is my problem..... ?