Time bomb: The case of the invisible failure mode
Tales From The Cube: An unexpected voltage spike compromises a monitoring device's reliability.
By Walter Lindenbach, Calgary Controls Ltd -- EDN, September 27, 2007
There is a failure mode that is worse than intermittent; no tests, measurements, or parts replacement will directly reveal the cause. You can test the circuit on the bench and in the field for months, and it will work perfectly. Then, within a year, a transistor fails. When you replace it, the device works well for months; then, the same transistor fails again.
I once had to fix an oil-pipeline-monitoring device with a relay driver that behaved this way. One after another, the transistor that switched the relay coil would fail within a year of installation. The relay-coil resistance was about 240Ω (12V dc at 50 mA)—by no means an excessive load for a small TO-92 transistor.
But a little 12V relay coil can produce a 200V spike. I was amazed when I saw it for the first time, and, with the scope that I had 30 years ago, it was hard to see. A transistor was switching a relay coil at about 10 times per second, and the scope triggered on the leading edge of the collector voltage. I turned the beam intensity to maximum, readjusted the focus, and, finally, after turning off the room lights, saw something going off the top of the screen! By turning the vertical sensitivity down and down again, the peak of the spike was finally visible—at 200V! That voltage was probably the transistor base-emitter-breakdown limit, not the peak of the inductive-voltage spike. A transistor with a collector-to-emitter breakdown voltage of 80V will not last long under those conditions, and, if you don’t protect it from the inductive spike, the transistor will fail after some number of switching operations—in minutes or months.
The usual way to protect a transistor from inductive spikes is to place a diode across the inductive load (cathode to VCC). Such an arrangement slows the relay release, but this circuit had no speed requirement so would have been satisfactory.
|
But I think the designer had placed a 10-nF capacitor from the transistor collector to ground to suppress the inductive spike, thus perpetrating another failure mode that looks exactly like the inductive-spike-failure mode—invisible! Only the transistor collector’s on-resistance and the resistance in the capacitor when the transistor switches on limit the current. Now, instead of an inductive-voltage spike, there is a capacitive-current spike, the amplitude of which is independent of the capacitor size. The current spike produces approximately the same result to the transistor as the voltage spike: It continues to operate for some time and then shorts.
I decided to protect the transistor with a different modification: I placed a 12Ω resistor in series with the emitter to limit the collector-current worst-case peak to 1A. A small plastic transistor, such as the MPS8099, can easily tolerate such a peak if it is short. Then, with a 50-mA normal relay-coil current, the drop across the emitter resistor was only 0.6V, which did not alter the performance of the circuit and seemed less likely to cause repercussions than removing the capacitor and adding a diode.
Now the monitoring device worked reliably. There were no more relay-driver failures to cause users to return these units for repair.
Walter Lindenbach started and operated Calgary Controls Ltd from 1970 to 1990, at which point he discovered an allergy to work and retired. Like Walter, you can share your Tales from the Cube and receive $200. Contact Maury Wright at mgwright@edn.com.
-
well that resistor does something, since the transistor last [longer as his maintenance contract] and the fellow is guessing what is happening, but it is a shoemaker solution, you could see many such fix in the industry, than it will be "incorporated" in the next revision in the equipment since it "helped". After while the old part will be replaced with a new one, which does not likes that "such a good solution", and the box starts to fail again ..does it sound familiar?
Alexander Pummer - 2011-8-12 20:36:59 PST -
This comment is REALLY late, but bears mentioning.
The author stated that the original designer had attempted to suppress the inductive spike from the relay by adding a capacitor across the transistor.
When the transistor turns ON, the capacitor discharges through the transistor. The resulting (very) high current spike degrades the transistor until it fails.
The author had several options to fix the problem:
1) remove the capacitor and add appropriate inductive spike clamping to the relay coil.
2) slow down the transistor (often easily done by adding a capacitor from B-E).
3) limit the peak current that can occur when the transistor turns on and discharges the capacitor.
From the description, the author chose option #3. I don't see a problem with that, especially if the capacitor across the transistor really DOES limit the transient voltage spike. Although its not mentioned, I assume that the author did indeed confirm that the transient was well controlled.
As to why he chose that option - its probable that cutting a single trace and adding a resistor across that cut may have been by far the easiest and quickest fix. This would be important if there were hundreds or thousands of boards to repair.
Dwayne Reid - 2011-5-12 20:32:44 PST -
The author is probably doesn't have any engineering education and doen't understand volt-seconds balance on
inductors. The editor should prevent this papaers from being published
Real Engineer - 2007-9-10 12:17:00 PDT -
I had some difficulty following this article and I think the writer is under a few misconceptions.
Assuming the standard relay coil drive circuit, I think the transistor failure was simply caused by a large positive spike on the collector, from the inductive kickback (back EMF) from the relay coil, and the base and emitter circuits don't really come into it.
Adding a capacitor across the transistor makes an L-C resonant circuit and this will reduce the frequency and peak amplitude of the ringing but I can't see any advantages to this approach compared to adding a diode across the coil, which is the traditional and best way to solve the problem, assuming relay turn-off time isn't an issue.
Adding an emitter resistor shouldn't help much, if at all. It limits the transistor current, yes, but by removing the transistor bias (the collector-emitter current pulls the emitter voltage up, and as the base-emitter voltage drops, the transistor turns off, limiting the current). With the transistor turned off, it is more vulnerable to damage from excessive collector voltage, not less. In any case the current will never reach 1A so I think the emitter resistor is pointless.
I think this problem is well-known and it's normal to add a diode across any inductive load; in fact some relays come with the diode built-in. The only reason to use any other suppression method would be for speed, but in that case I would add components in the collector circuit and leave the base and emitter circuits alone.
Kris Heidenstrom - 2007-4-10 13:41:00 PDT -
Assuming there were no turn-off speed requirements for the relay, it seems like a free-wheeling diode around the coil would've been an easier mod than adding a series emitter resistor. A resistor in series with the diode could be added to speed up the turn-off, but then you need to make sure the voltage spike isn't high enough to (again) kill the transistor.
Ron Bauerle - 2007-4-10 09:49:00 PDT
