I have been using metallized polyester film capacitors in the output filter of a PWM servo amplifier. I recently discovered that when the PWM duty cycle is near 50%, these caps can get quite hot. Perusing the data sheet, it appears that I am running above the manufacturer's current rating in that condition. Can anyone recommend a capacitor around 10 nF and 250 V DC rating that can handle several hundred mA at 50 KHz? (There is a series inductor before the cap, so the cap will see roughly a sine wave current up to 250 mA or so.) I'm hoping to keep the cost and size down on this, too.
Sounds to me you have a design problem? Are you opening in source/sink mode only and not both ? It so, appears to me that you could have your caps absorbing the wheeling voltages? This only assumes these are the caps that connect to a motor circuit, if that is what you have?
Mica caps can handle that punishment just fine, however, they aren't cheap.
No, I don't think this is a design problem, ie. topology, but just a ratings problem.
Well, to simplify, assume a full-bridge switch, and the direction is set to one setting. One half bridge will be grounded, and there will be very little voltage there. the other half bridge will be switching between ground and the motor supply voltage at the PWM frequency. That's where the problem appears. That filter cap will see a roughly triangle wave current at the PWM frequency, and a magnitude of several hundred mA. Right now I'm using a 0.1 uF cap with a current rating of about 400 mA. The inductor is 47 uH. I'm planning on scaling the cap value down, but then the current rating scales down even faster, so that doesn't help.
I doubt I can find a .1 uF 250 V mica cap that is smaller than the entire servo amplifier! Or, costs less than the sale price of the amp. Yes, they sure have low dissipation, that's why they are used in radio transmitters.
That might help, I may have to redesign the board, anyway.
One big problem is most of the capacitor data sheets give no current rating, many don't even list dissipation factor.
So it's got ~100V RMS on it? Kind of a lot of ripple; suppose I might guess you can't afford much, in order to get a very fast servo or something.
I can't think of a polypropylene with that value which won't work, even the chincy X2 "across the line" caps will take it. Other than EMI/across-the-line caps, anything with "pulse" or "snubber" in the description will certainly do.
If SMT is required, an X7R chip cap (probably 1206 to 1812 sized) would be marginal; a C0G MLCC, though expensive, will still be smaller than any film, and will *never* heat up.
Your only real problem was the choice of dielectric: polyester has a higher loss tangent than PP. The loss also has a nonlinear tempco -- once an internal spot gets to a certaisn temperature, it keeps heating until it burns through and bubbles melted plastic!
Tim
-- Deep Friar: a very philosophical monk. Website:
That's odd. The capacitor only conducts when the waveform is transitioning. As long as the rise time is fairly fast and fairly constant, you shouldn't have much change in dissipation with different duty cycles.
Rise time is also important. For example, if the rise time is 1% of the full waveform, and the dissipation is 1 watt, then average dissipation is only 0.1 watts. What usually ruins the dissipation is ringing and oscillations. These will greatly extend the time that the waveform spends in transitions. I'll wager the scope probe that I just destroyed that you'll see considerable ringing and/or oscillation.
Have you looked at the waveform time on a scope? What do you see?
Umm... 0.1uf is 100nf. What happend to the 10 nf requirement?
47uH in series with 0.01uF resonates at about 200KHz with a series reactance of about 65 ohms. What's your real switching frequency (50KHz?). What is this series LC network suppose to do? If it's on the 3rd harmonic of your square wave, you'll could have about 1/5 of the applied power smoked from the 3rd harmonic through the LC network. If it resonantes on the switching frequency, it can be as high as 2/3 of your applied power. Be careful with your choice of values.
Well, could you supply a maker and model number for the existing polyester cap in question? I want to lookup the ESR. If you have an ESR meter, a measurement would also be nice. 50KHz is close enough to the standard 100KHz used for ESR measurement.
Yeah, but you won't like it. Mica has already been suggested. I'll add porcelain. I use them for RF power amps when nothing else will do. Available up to 1uF. You won't like the price.
For 0.010uf, you're looking at about 0.025 ohms ESR.
Ok, I'll use that. Pwr = I^2 * R = 0.25^2 * 0.025 = 1.6 milliwatts. Yeah, I think it will survive.
First you make it work. Then you make it pretty, fit, and cheap.
--
Jeff Liebermann jeffl@cruzio.com
150 Felker St #D http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann AE6KS 831-336-2558
--
Jeff Liebermann jeffl@cruzio.com
150 Felker St #D http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann AE6KS 831-336-2558
Oops. That's ceramic, not porcelain. Ceramic will work, but porcelain is better:
However, porcelain is only available to 2.2nF. You'll need 4 of them to make 10nF. 0.014 ohms ESR.
--
Jeff Liebermann jeffl@cruzio.com
150 Felker St #D http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann AE6KS 831-336-2558
True, However, I was assuming that it was a series resonant circuit across some high power output used to reduce high frequency hash for EMI compliance. If it really is running at 50KHz, and the series resonant frequency is something like 200KHz, it wouldn't make a very useful low pass filter. Therefore, I'm guessing(tm) that it's a hash filter or part of some attempt to stabilize the unspecified circuit.
--
Jeff Liebermann jeffl@cruzio.com
150 Felker St #D http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann AE6KS 831-336-2558
The standard Philips/Epcos polypropylene parts can have that AC current rating, but only the >=630VDC rated parts.
formatting link
For pricing, try Digikey for an indication. $0.22 doesn't sound out of line, though you can probably source the same part numbers or similar for a quarter of that, off-shore.
No, this is just an RFI filter. The 100 V is at the input to the inductor, so the voltage on the cap is dependent on the PWM duty cycle and the capacitor value. Larger cap gives lower voltage, but higher current. Too large a cap and the inductor gets hot, too. Too small a cap and the RFI reduction is diminished. Still, just knocking the sharp edges off the switching may be enough benefit.
The problem is most cap manufacturers just specify size and the mysterious "dissipation factor" without much info on how they measure it. Panasonic gives curves of rated current and rated AC voltage at various frequencies, and I'm over the limit. There are some HUGE caps that might work, but I'm trying to avoid going much bigger.
No, ceramics have WAY higher dissipation, I measured it on a bridge, maybe 5 - 10 times higher. data sheets bear this out, too. Due to temperature limitations, I solder these through hole parts in later. These things are 8 x 8 x 5 mm, roughly, may bigger than modest SMT parts.
OK, that's helpful, I'll see what I can find in PP!
There is a big inductor in series, so the current never builds up on short PWM duty cycles. The PWM is 50 KHz, so the input to the inductor sees the full voltage square pulse, but the cap at the output only sees essentially a sine wave of much lower amplitude.
No, very little of that. There is a 47 uH inductor and the current design has the 0.1 uF cap to ground. No ringing to speak of, and certainly no "oscillation".
A sine wave, superimposed on some DC (motor voltage).
I'm planning on going to a smaller capacitor to reduce the current. But, the smaller caps have even SMALLER current ratings, so just selecting a smaller cap value doesn't fix it.
Yes.
Yes, I was careful to have the resonance well above the PWM frequency. It is an RFI filter.
Digi-Key sells it as the P10967, or Panasonic ECQ-E2104KB
One of the few parts that have detailed loss graphs.
Yes, it is to keep the motor wires from radiating/conducting interference. There are encoders on these motors, and it is good to prevent noise from getting into the encoder signals. It should also keep the FCC at least happier.
What's mysterious about dissipation factor? It's just the ESR as a fraction of the reactance at some (defined) frequency.
IOW, if the device is of impedance R+jX, then DF = R/X
Another way, it is the reciprocal of Q, or the tangent of the loss angle.
All good LCR meters measure dissipation factor. Manufacturers will probably use something like an HP4274A, or newer. That will display as dissipation factor, Q, ESR, equivalent parallel admittance, or magnitude of Z, with angle. They're all the same thing, it just manipulates the math.
I've got one of those. Use it regularly.
--
"For a successful technology, reality must take precedence
over public relations, for nature cannot be fooled."
(Richard Feynman)
Yes, but can I really trust the current rating on no-name import parts? I will check out these EPCOS parts, $0.22 is certainly an OK price, but I'll have to revise the board for the 7.5 mm lead spacing.
Umm, a C0G 10 nF 250 V ceramic cap would be RATHER expensive. Hey, I looked it up, they are NOT very big, though, so maybe getting these in SMT and not having to hand solder them offsets some of the expense. If going down to the 10 nF value works out OK, then they are only $0.72 in modest quantity, that is only a bit more than I'm paying for the
100 nF film caps, now.
It wasn't so long ago such a big cap would have been outrageously expensive, so this is good news! I'm going to get some and test in the application!
Right, the high capacitance dielectrics are not good for this kind of application. Higher current film caps are available, but more expensive and a lot larger.
Right, and some data sheets do not GIVE the frequency it is measured at, making the data meaningless. You can GUESS what freq they might have used, of course.
You said: 10nF and "several hundred mA", which I'm taking as 300mA worst case. You later said 50kHz, which puts Xc = 318 ohms and V = 318 ohm * 0.3A = 95V, close enough to 100.
I see now that you used a 100nF, which puts it closer to 10V. You'd still have to check thermal limits, but that's on par with the ripple an X7R ceramic chip cap can tolerate. Think of them as the electrolytics of ceramic: they have fairly high DF, but if you keep the ripple low, it's not a big deal.
Alternate rationalization: if it were a resistor, it would only dissipate
10V * 0.3A = 3W, obviously too much for a poor 1812 (or at 100nF, probably multiple devices or a larger footprint), but if the DF is less than 0.1, you'll have less than 0.3W -- probably more than you wanted, but not impossible for a device that size.
Now, you didn't mention how much voltage the cap actually has to carry, so your actual losses due to signal voltage may be substantially higher. Especially if you're driving it hard at a frequency below, but near, resonance (like 50k, which is notably below 78k).
For comparison, I did this:
formatting link
(please excuse the sloppy layout), which has an output filter choke L2, a
220nF X7R (1206 size) bypass cap, a common mode EMI choke and another 100nF in bypass (0805 size).
Operating frequency is about 200kHz, while the cutoff is 38kHz, so the inductor takes the brunt of it here. However, I haven't noticed any appreciable heating of the capacitors running full signal output at 10kHz, which is comparable to your application if the RMS voltage is comparable and the damping factor is constant.
L2 is 47uH each winding, wound on a mix #52 powdered iron toroid, which has moderate DF, maybe 0.05 (it gets hot, but not enough to be a problem). The EMI choke L4 is 100uH, wound on #43 ferrite, which has low DF at this frequency. It's split wound so the leakage inductance is substantial (single digit uH?), which provides a little diff mode filtering, in addition to L2-C3-C4.
Note that R17 gets warm when operated at high frequencies for a long time. The caps are fine though.
Not sure if I've seen sheets that don't at least mention the test frequency (usually 1kHz, anyway), but it's true they often leave it at DF.
There's a few vague classes of cap ratings I've seen:
- General purpose and EMI types:
These are physically small for their value, very cheap, and have weak ratings. Typically, the datasheet only specifies DF.
Examples include: anything physically small (like the 0.1uF 63V MKTs and MKPs made by various manufacturers -- good for timing and supply bypass), CDE's DME series, etc.
The only "teeny EMI" style cap I've seen with [inferred] current rating is a Vishay/BC product, which showed voltage derating along a curve of a constant
100mA or so. Sounds about right for the size.
Abuse story: I once found Illinois Capacitor's 104MKP275K (275VAC 0.1uF, only 13mm long) on sale at Allied. They got pretty damn toasty with only an ampere ripple (though none failed that I know of); realistic ratings are probably closer to the BC part above.
Abuse story 2: on a whim, I tried a few DMEs on an induction coil. These didn't last long:
formatting link
Burned a hole right through the poor thing; subsequent testing revealed the others were weakened as well. Surprisingly, that was ~140Vrms at 50kHz or so at failure.
On topic note: your polyester caps fit in this category, of course.
- "DC Link" caps:
These seem to be pretty popular. Usually oversized MKPs, they have lower DF than electrolytic or mylar, but still have relatively low ripple capacity (you'd size them for a ripple voltage of maybe 5% out of total supply voltage), and usually awful frequency response. For example, Epcos rates some of theirs at 10 or 20kHz, but provides no derating for other frequencies. Due to construction, the derating is probably quite steep in the 100kHz range!
Example: Epcos B32674, though many manufacturers make essentially direct substitutes. Can style ("FilmLytic" and others) caps probably fit this category as well; being physically larger, they have higher ratings of course.
Abuse story: The neat thing about the failure modes of these caps (the MKPs) is, the casing swells as the roll of metallized polypropylene melts, but it's constrained by the epoxy potting. The potting cracks open on the bottom, and the roll extrudes itself through the crack, a striped blob of gray goo! Molten metallized film looks really cool. :)
- Pulse and snubber rated caps:
These always have dV/dt or peak current ratings, and sometimes RMS as well. Good stocky capacitors, but usually optimized for peak over continuous duty. Not always great in frequency response.
Examples: many Epcos parts; CDE 935 and 940C series; many others.
Abuse story: I've seen 935s heat up noticably when used over 100kHz, continuous duty. They are rated for 100kHz duty, but fairly conservatively (the peak-to-RMS ratio is quite large). No failures, but their peak-rated construction gives them similar drawbacks to DC Link types. I've seen these used for induction heating, and apparently they do a fine job in the ~50kHz range.
Note on dV/dt: generally, a manufacturer provides this as a peak rating only. Typically, the RMS value is 10-50 times lower! If a peak current rating is provided, it's usually equivalent by the capacitor equation: I = C
dV/dt.
On topic note: the Panasonic caps you looked at (ECWF(A) or so, I'm guessing?), I believe, fit into this category. I haven't used all that many of these caps, but I'm familiar with them.
- Heavy duty caps (about as heavy as you'll see in a THT package):
Phrases like "high pulse" or "heavy duty snubber duty" are often used around these types. Typically, dV/dt is in the thousands (V/us), even for moderate sizes (0.1uF+). RMS capacity runs in the amperes (same size range). These types naturally have the highest voltage capacity; a 0.1uF 630VDC cap might be rated for 250VAC up to a full 20kHz (~3A) before derating. Ratings often include k_0 values (a measure of RMS-weighted peak capacity).
Examples: Epcos B32671 and etc., Illinois Capacitor PPB series.
Abuse story: I don't think I've ever burned up one of these. I once ran 10A through a 334PPB630K (0.33uF 630VDC) at 300kHz, which is only rated for
1500V/us peak, or 500A peak. Didn't get any hotter than the nearby cabling.
I have actually seen some 223PPB630K destroyed under heavy induction use at
400kHz, but this isn't a fair distinction because the coil wasn't water cooled at the time -- the whole thing overheated, so what do you expect? :)
- More heavy-duty caps:
I'll add section more as a note. IGBT snubber capacitors are available in boxed packages, with bolt lug terminals for direct use on IGBT modules. Needless to say, these are rated heavily. Though expensive, you wouldn't need many for, say, an induction coil of useful power (kW+). You occasionally see oval can (film and oil) caps with similar ratings. And, of course, induction heating caps, which are in a class of their own (usually including water cooling or big heatsinks!).
- Final note:
Concerning manufacturers, I probably have the most experience with Epcos, and therefore mention them often, because they make so many damned products, and they provide reasonable datasheets for their products. Panasonic, Kemet, Vishay, AVX, CDE and many others make products similar to those mentioned, some of which I've used. If you can guess or figure out a particular product fits into the above categories
Now, that's not fair: C0G is essentially lossless. Along with mica, porcelain and related "inert" sorts of dielectrics, the dV/dt capability is extraordinary, dissipation second only to air, and current capacity essentially limited by package conductance (electrical and thermal). IIRC, CDE gives dV/dt ratings on their mica radials in the ~100kV/us range.
I've personally driven poor little 100pF, 50V C0G ceramic disks at 500V RMS,
500kHz (that's 160mA), running only luke warm. Even the same value in Y5P (but higher voltage) overheated and distorted quickly. Disks arc over before the ceramic itself punctures.
I've also ran 470pF 1kV C0G, 1206 size chips, with over 14kV/us peak dV/dt (>6A peak) with no problem, and 1nF 630V 1206s at 1A RMS at 1MHz (~160VAC) with no problems. The real hazard is arcing on these little things.
As for size -- man, I have some old boards with NP0s on them. Tube era stuff. On the rare occasion they had no choice but to plop one down, it's obvious they were expensive: HUGE disks for very little capacitance. Might be... 500V 82pF in a disk an inch across! Needless to say, MLCCs are massively more dense, only a few times the volume of an equal value X7R (which, needless to say, has also shrunken considerably in the intervening years!).
Tim, who takes his capacitors seriously :)
--
Deep Friar: a very philosophical monk.
Website: http://webpages.charter.net/dawill/tmoranwms
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.