Relay suppression takes ages to decay

[741]

I'm driving a large latching relay with a DC/DC convertor. I'm measuring the flyback suppression (simple 1N4148) by tying 'scope GND to the isolated 48V, and 'scope tip to the bottom of the coil - see image.

How can the decay take so long?

Note: The relay is already latched into place for this test (there is another coil on the same yoke to move it to the other latched position). Therefore, it should be acting as a simple inductor (no armature movement).

Repeating the test with four parallel coils made little difference.

The 1k winding resistance should make it decay quite fast, like the simulation?

[Benta]

This has been covered many times in the latest weeks.
It's a bad idea to place a diode like that.
Yes, I know, it's been shown in a lot of circuits and examples. That doesn't make it right.

The diode simply lets the current keep flowing in the relay coil when the transistor turns off, making release time really slow. At some point, the current decays to a point where the relay releases.
That's exactly what you're seeing on your 'scope plot.
A better approach is to use a Zener or TVS across the transistor limiting the voltage, but allowing the relay coil to dissipate the energy stored in it.
Example: transistor is 100 V, supply/relay voltage is 48 V. Choose a 75 V Zener/TVS.

[bdunham7]

This has been covered many times in the latest weeks.
It's a bad idea to place a diode like that.
Yes, I know, it's been shown in a lot of circuits and examples. That doesn't make it right.

The question he asked was why the measured result is so different from the simulation.  2.8H and 1k gives a time constant of 2.8ms.  His simulation shows a decay occuring over a few milliseconds but the actual reading shows much, much longer.  Why?

IDK the answer myself and I don't really understand the scope picture.

[bdunham7]

I'm measuring the flyback suppression (simple 1N4148) by tying 'scope GND to the isolated 48V, and 'scope tip to the bottom of the coil - see image.

I don't really understand your scope shot.  200mV/div?  What scope is that?  Can you put the entire signal on the screen?

[T3sl4co1l]

Obviously, the coil resistance is lower and/or inductance higher (also at longer time scales) than the simulation.

Time scales matter because real relays, solenoids, etc. typically have bulk metal parts, which the magnetic field penetrates only slowly (10s of Hz, if that).  It can take quite some time for fields to equalize, and thus coil current (at turn-on) or voltage (at turn-off) to settle.

You say nothing about where your values came from, so I'm assuming they're completely arbitrary; if 2.8H and 1kohm came from a datasheet, check what frequency it was measured at; the value at 1kHz for example can be quite a lot smaller than at 10Hz (i.e. by about 10x).

(The underlying phenomena here is skin effect, which gives an impedance ~ sqrt(f), and thus inductance ~ 1/f, which is why I can be, at least ballpark confident, that a 100x change in frequency gives a 10x change in measurement.)

Tim

[Andy Watson]

The question he asked was why the measured result is so different from the simulation.  2.8H and 1k gives a time constant of 2.8ms.  His simulation shows a decay occuring over a few milliseconds but the actual reading shows much, much longer.  Why?

Is the 2.8H measured with the amature engaged or disengaged?

[duak]

The measured waveform in the upper image doesn't make sense if the scope ground is connected to +48V as per the schematic.  If it were, the waveform would have to be below the 0 reference level.

Assuming the scope ground is connected to 0V, I would expect the waveform to dip below 0V to approximately -600 to -700 mV, the level at which the FET's body diode is turned on.  Since it doesn't go negative, perhaps the FET is being turned off so slowly that the collapsing mag field in the coil doesn't generate the expected compliance voltage.  If the circuit exists as drawn, I would bet the DC-DC converter output voltage's non zero fall time is causing the slow switching.  I'd like to see the FET gate voltage.

BTW, suppose the DC-DC converter is one of those photovoltaic optocouplers that generates an output voltage sufficient to drive a FET?  When the PVOC output drops, the Miller effect from the FET Cgd is probably more than sufficient to extend the FET turn off time without some additional pulldown circuitry.
 

[T3sl4co1l]

The measured waveform in the upper image doesn't make sense if the scope ground is connected to +48V as per the schematic.  If it were, the waveform would have to be below the 0 reference level.

Look closer... that thin yellow pixel is it being below 0.  This is the little bit above, diode forward-biased, relative to +48V rail, thus positive, as shown.

Assuming the scope ground is connected to 0V, I would expect the waveform to dip below 0V to approximately -600 to -700 mV, the level at which the FET's body diode is turned on.  Since it doesn't go negative, perhaps the FET is being turned off so slowly that the collapsing mag field in the coil doesn't generate the expected compliance voltage.  If the circuit exists as drawn, I would bet the DC-DC converter output voltage's non zero fall time is causing the slow switching.  I'd like to see the FET gate voltage.

How would body diode ever become forward biased when current was massively the opposite direction moments before?

Tim

[741]

Added some more images.

To simplify, I removed the diode and placed 2 parallel 1k (i.e. 500R) as the sole suppression element. The most current they see is the 50mA going round the coil just after the coil is 'let go'.
NOTE: Note all tests involve no armature movement, simply re-energising the coil in the same way I did when it most recently latched. Just 'simple' electronics. We expect about 50mA * 500R = 25V peak back-emf.
<see image>

We now have a total of 1500R in the flyback loop (1k from the relay, 500R from the resistor). This would shorten the decay compared with 'just a diode', but the delay is still massive.
Note the peak flyback, 25V is as expected.

Have been wondering - can the permanent magnet affect things? We have an offset into the B-H curve. Also, an LCR meter would likely use a bipolar signal, different from the use-case of a relay coil.

The guy next to me suggested re-testing with the coil polarity exchanged... I gave that a try.

One issue is this pushes the relay into a different position, so the permanent-magnetic envionment changes, but I doubt it changes much (the magnet sits on the armature, moving with it). Here are the results - notice the decay time has reduced in the 2nd picture where the relay coil leads are swapped.

("R1, R2" are the confusing coil pin-labels given on the relay base)
<see images>

[wraper]

This has been covered many times in the latest weeks.
It's a bad idea to place a diode like that.
Yes, I know, it's been shown in a lot of circuits and examples. That doesn't make it right.

The diode simply lets the current keep flowing in the relay coil when the transistor turns off, making release time really slow. At some point, the current decays to a point where the relay releases.
That's exactly what you're seeing on your 'scope plot.
A better approach is to use a Zener or TVS across the transistor limiting the voltage, but allowing the relay coil to dissipate the energy stored in it.
Example: transistor is 100 V, supply/relay voltage is 48 V. Choose a 75 V Zener/TVS.

It's not necessarily a bad idea. It depends on the relay/load it's switching. Yes diode will result in a slower release. However in many circuits it does not matter or negative effect on relay contacts in miniscule in a grand scheme of things.

[duak]

D'oh, how embarrassing.  Tim is correct about my mistaken interpretation a few posts back about current flow in relay drivers and the polarity of the voltage spike.  The FET drain voltage should indeed rise to try to draw more current through the FET as it switches off.  In spite of working with power FETs since the early 80s, i convinced myself it was t'other way 'round.  That's what retirement does for you.

One project I worked on was for a coil spring lifetime tester.  A number of springs were connected to a mass that slid along a central shaft with a linear guide bearing to reduce friction.  An electromagnet drew the mass one way with a 24 V pulse of a few ms in width and the mass would oscillate back and forth. I used a clamp diode with a series power resistor to discharge the energy in the electromagnet's inductance.  I used some 400 V FETs to allow for quite high voltage spikes if needed.  If memory serves, the spikes were about 100 V peak.  This thing would grind and clatter away for years at 24 Hz in the basement of our building testing springs..