Testing Capacitors

Intuition

Proving the room is empty before locking it

A building's closing routine is not a walk around the corridors. It is a check, room by room, with a record: not because anybody expects to find somebody in there, but because the cost of being wrong once is far larger than the cost of checking every time. The routine exists to convert a belief into a fact.

Testing a capacitor starts with that same conversion, and the belief being tested is that it is discharged.

A 220 µF capacitor charged to 400 V holds 17.6 J. That is not an abstract quantity: it is enough energy to weld a screwdriver tip, blow a crater in a probe, and pass a very unpleasant current through a hand. The capacitor may have been sitting in a drawer for a week. It does not care.

So the sequence is fixed. Discharge it deliberately, through a resistor. Measure across the terminals to confirm. Then, and only then, start finding out whether the part is any good.

The second half of the job turns out to be more interesting than most people expect. There are three separate measurements a capacitor can be given, they answer three different questions, and a part can pass any one of them while failing another. The meter that most people own answers only one of the three.

Practitioner

Discharging it, and confirming

Two seconds, or the rest of the afternoon.

The second trace is the reason the confirming measurement matters. Left to its own leakage, the same part takes hours rather than seconds, and there is nothing about its appearance that distinguishes the two cases.

There is a further wrinkle worth knowing about, because it surprises people who did everything right. A capacitor that has been shorted and then released does not always stay at zero: dielectric absorption returns some of the charge over the following seconds and minutes. Film capacitors works through the mechanism with millivolt-scale examples; on a large high-voltage part the recovered voltage is far larger than that. The habit that deals with it is to leave the discharge resistor connected while working, rather than discharging once and moving on.

Engineer

Three measurements, three different questions

Four readings, one component, and no contradiction.

Capacitance answers whether the part still holds what it says. A reading of 198 µF against a marked 220 µF is 10.0 % low, which is inside an ordinary electrolytic's tolerance and also consistent with a part halfway through its life. The measurement alone cannot tell those apart, which is the recurring difficulty with this test: it needs a reference.

Series resistance answers whether the charge can get out, and it is the more sensitive test. ESR roughly doubles over a part's life while the capacitance moves only a fifth, so a rising ESR shows up first. It is also the measurement that finds bad joints, since a poor connection is in series with the part and reads as if it were part of it.

Leakage answers whether the insulation is intact.

A leakage of 12 µA at 400 V is an insulation resistance of 33.33 MΩ, and that resistance sets a self-discharge time constant of 7.333 ks, which is 2.037 hours.

The better the capacitor, the longer it stays dangerous.

A cleaner part at 500 MΩ has a time constant of 110 ks and takes 63.54 hours to fall to a safe level. Low leakage is a virtue in the circuit and a hazard on the bench, and the two facts are the same fact.

No single test clears a capacitor, and none of them measures the voltage rating at all.

Two things no bench test reaches, and they are worth stating plainly because a clean set of readings invites the opposite conclusion. The voltage rating cannot be measured without applying the voltage, which is a destructive test on a part that may be the reason the equipment failed. And behaviour under working conditions is not what a meter's small test signal produces: a part that measures perfectly at a few millivolts can still misbehave under bias, ripple or temperature.

Professional

Why two meters disagree about the same part

Three correct measurements, three different answers. One is an instrument error and one is a real property of the part.

Two different mechanisms are at work in that picture, and separating them is most of the skill in this measurement.

The frequency is a real property. A part's series resistance falls with frequency, because part of it is dielectric loss rather than metal. Measured at 120 Hz this part reads 400 mΩ, and at 100 kHz it reads 150 mΩ, a factor of 2.67. Both are correct. Which one matters depends on the job: a reservoir behind a mains rectifier works at the low figure, and a capacitor in a switching converter works at the high one. Comparing a reading against a datasheet figure taken at a different frequency is comparing two different quantities.

The leads are an instrument error.

The red boxes are the lead resistances, not components anybody fitted.

With two terminals, the current that drives the measurement and the voltage that reads it share the same wires, so 40 mΩ at each end is inside the answer. A part whose real resistance is 150 mΩ reads 230 mΩ, which is 53.3 % high. Four-terminal connection separates the two paths: the sense wires carry no current, so they develop no voltage, and the reading is the part alone. At these values that is not a refinement, it is the difference between a useful number and a meaningless one. The multimeter and LCR meters cover the instruments themselves.

Testing in circuit, and when not to bother

An in-circuit ESR check is fast and it is the standard first move on a suspect power supply, because the board's other components mostly do not shunt a low resistance at the test frequency. It has two honest limitations. Everything in parallel with the part is inside the reading, so the answer is an upper bound rather than a measurement. And a good reading does not clear the part, because capacitance and leakage were never measured.

A capacitance measurement in circuit is usually worthless, because anything in parallel adds to it and the reading is a fiction.

The judgement that follows is a practical one. Use in-circuit ESR to find an obviously bad part quickly. Remove the part when the reading is ambiguous, when the value matters, or when leakage is the suspicion. And where several identical parts sit on one board, measure all of them: the comparison between siblings is often more informative than any absolute reading, because they have shared a history and a temperature.

What the readings are worth without a reference

Everything in this lesson depends on knowing what the part looked like new. A datasheet gives it, a new part of the same type gives it, and a sibling that has run cooler gives an approximation. Without any of those, a single set of readings establishes only that the part is not obviously dead. That is worth knowing and it is not the same as good, and failure modes and ageing is where the drift those readings are looking for comes from.