1
\$\begingroup\$

I have a motor drive which uses 63V electrolytic capacitors, which are the part with lowest-rated voltage on the drive's DC link. I measured the switching spikes on the DC link voltage (carefully, with spring-ground lead rather than ground clip to minimize lead length) and found on the order of 5V spikes at the 20kHz switching frequency with 48Vdc nominal; they last on the order of 400-500ns.

Here's my question: how do these switching spikes affect the 63V capacitor rating? Are they fast enough that I can ignore them and can allow the DC average voltage to get up to 63V? Or do I need to factor in these spikes, along with appropriate engineering margin, meaning I should stop somewhere in the 53-58V range depending on my conservativeness?

(Note: this assumes I am comfortable operating up to the low-frequency 63V rating, which I may or may not choose to do. That's a different question; I'm asking whether I need to take the high-frequency spikes into account.)

enter image description here


Cornell Dubilier's application guide states the following (subject to interpretation of course)

Aluminum electrolytic capacitors can generally withstand extreme overvoltage transients of limited energy. Application of overvoltage more than about 50 V beyond the capacitor’s surge voltage rating causes high leakage current and a constant- Percent Rated Voltage voltage operating mode quite like the reverse conduction of a zener diode. The capacitor may fail short if the electrolyte cannot take the voltage stress, but even if it can, this operating mode cannot be maintained for long because hydrogen gas is produced by the capacitor, and the pressure build up will cause failure. However, special designs are available that use the overvoltage, zener-clamping efect to successfully protect equipment from overvoltage transients such as lightning strikes.

Capacitors used as bus capacitors in large, high-voltage capacitor banks are less capable of withstanding overvoltage transients because the high energy and low source impedance of the capacitor bank can prevent a momentary partial discharge from self healing and cause it to become a runaway shortcircuit failure. For high-voltage capacitor-bank applications use capacitors proven for that use.

edit: The above voltage was measured across a terminal block (Phoenix-style) that connects directly to power planes; so does the capacitor. I have also measured directly across the capacitor itself at 24V --- had to tip up the board holding everything together by hand, press oscope lead and ground ring against capacitor terminals, so I'm being a bit cautious before trying this at 48V:

enter image description here

(Screenshots of Agilent MSOX3034A 350MHz using 500MHz 10:1 passive probe.)

\$\endgroup\$
9
  • \$\begingroup\$ How much current are you switching? What size are the caps? \$\endgroup\$ – Trevor_G Jan 11 '18 at 20:20
  • 1
    \$\begingroup\$ How much do you think a 470uF cap can charge up at 1.5A for 400 nano-seconds? I figure 1.25mV, You are not seeing what you think you are seeing. \$\endgroup\$ – Trevor_G Jan 11 '18 at 21:14
  • 1
    \$\begingroup\$ There are high-frequency plane capacitances on the order of 500-1000pF, plus capacitance from MOSFETs. I never said this electrolytic capacitor was the cause of these spikes. (At such high frequencies the electrolytic cap ESR + ESL makes it look noncapacitive; the electrolytic cap is helping me over longer periods of time.) \$\endgroup\$ – Jason S Jan 11 '18 at 21:39
  • 1
    \$\begingroup\$ That is not what I am saying. I am saying the voltage across the capacitor dielectric can not be changing that much. What you are seeing on the scope is outside the cap and does not violate the cap voltage, assuming you are not already at the cap max V. \$\endgroup\$ – Trevor_G Jan 11 '18 at 21:44
  • 1
    \$\begingroup\$ Because across the wires coming out of the cap IS NOT across the capacitor plates. That is internal beyond the cap ESR and inductance of the leads. The spikes are likely due to capacitive coupling through whatever you are switching with. That is all happening a lot faster than the cap as a whole can react. \$\endgroup\$ – Trevor_G Jan 11 '18 at 21:51
2
\$\begingroup\$

Lifetime of aluminum electrolytic capacitors is generally specified as the time under certain conditions of applied DC voltage, ripple current, and ambient and surface temperature. under these worst case conditions the lifetime is expressed in hours (eg 1000h @90'C). Thus derating these absolute maximum values leads to multiplier values of lifetime for each of these parameters.

The criteria is defined by the capacitor’s electrical parameters has drifted out of some specified limit. The ESR is usually first to go, so as self heating temp. rises with ESR, soon the capacitor will either run so hot that it suddenly shorts out or that it ruptures its safety vent and begins to dry out and drift open circuit. Another failure mode is reverse and over-voltage stress.

CDE uses the criteria for extending Life by Voltage ratio, Mv of applied dc/rated dc;
Mv = 4.3 - 3.3 Vdc/Vr ... ref p2

Thus using Vdc/Vr = 2/3 ratio leads to a lifetime multiplier of 2.1x while using 0Vdc leads to a lifetime multiplier of 4.3x and full rated voltage = 1x

The actual peak voltage at the cap terminals is what counts for breakdown voltage stress. It must be measured direct across the cap. terminals.

I don't think the ripple voltage in your photo indicates much RMS ripple current so the stress factor for ripple current is low.

\$\endgroup\$
13
  • 1
    \$\begingroup\$ Yes they could be from the inductance between the cap and point of measurement. \$\endgroup\$ – Tony Stewart EE75 Jan 11 '18 at 20:24
  • 1
    \$\begingroup\$ Yes they must be from the inductance between the cap and point of measurement. SRF of the pulse is somewhere between 10M and >20MHz from the resonance which is far too high for an E-cap \$\endgroup\$ – Tony Stewart EE75 Jan 11 '18 at 20:31
  • 2
    \$\begingroup\$ @JasonS I think the Ecap is not under load here. It's ripple current would be related to motor commutation frequency. This is parasitic resonance and due to high ESR on supply cap. ESR*C on E-cap is most likely 2~200us range from best to worst and never 0.1 to 0.5us. This can only be done by plastics, FR4 and ceramics \$\endgroup\$ – Tony Stewart EE75 Jan 11 '18 at 21:14
  • 1
    \$\begingroup\$ Half bridges usually have ultra low ESR E-caps near driver V+/- \$\endgroup\$ – Tony Stewart EE75 Jan 11 '18 at 21:17
  • 1
    \$\begingroup\$ really not appropriate to rely on electrolytics for that purpose -- embeddedrelated.com/showarticle/588.php \$\endgroup\$ – Jason S Jan 11 '18 at 21:27
1
\$\begingroup\$

I've never heard anyone recommend running electrolytic capacitors exactly at their rated voltage. If you want to run at 63V, I would recommend replacing the caps. A 20% margin over the highest voltage you expect is probably safe.

\$\endgroup\$
2
  • \$\begingroup\$ Thanks, but I'm not asking about the DC (low frequency) voltage rating; the capacitor manufacturer gives certain assurances, and whether I choose to go to 100% or 90% or 50% is another matter. I'm asking whether the capacitor "sees" such short spikes, or whether they are too short to have any effect. I know that power semiconductors will definitely see the short spikes and they have to be taken into account when determining voltage rating. \$\endgroup\$ – Jason S Jan 11 '18 at 19:55
  • \$\begingroup\$ @JasonS sure you can use it to 100%, assuming you can guarantee you do not go over by a fraction of a percent. Good design practice is ALWAYS to de-rate parts. \$\endgroup\$ – Trevor_G Jan 11 '18 at 20:31
0
\$\begingroup\$

Your spikes have 20 nanosecond Trise. Across 10nanoHenry ESL. With deltaV of 2 volts.

V = L * dI/dT

dI = dT * V /L

dI = 20nS * 2v / 10nS = 4 amps.....................sounds about right?

\$\endgroup\$
1
  • \$\begingroup\$ Thanks but this doesn't answer my question. \$\endgroup\$ – Jason S Jan 13 '18 at 22:36
0
\$\begingroup\$

How do you measure? That looks like an artifact caused by switching and poor CMRR of long ground leaded probe. If you use a very short ground lead (like 1cm),probably you will see nothing.

To support the point: it can be clearly seen that there are three- four dominant frequencies, meaning transfer function of switching to probe is a complex one. It involves ground lead inductance, area of the mesh by probe and it's lead, and so on. Cancel all those, picture will be much clearer and you will be much calmer.

\$\endgroup\$
1
  • \$\begingroup\$ This is better as a comment, not an answer. Also I took the probe hood off with no alligator ground clip. The first scope plot was with a short (approx 1cm) spring clip placed on the ground ring of the scope probe (sorry I don't have a picture) and contacting the power and ground plane through the top screws of a terminal block (approx 1cm to planes). The 2nd scope plot was with the spring clip removed and the probe tip and ground ring pressed directly to the capacitor terminals. \$\endgroup\$ – Jason S Jan 13 '18 at 22:39

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.