Disaster-preventers: speculations on things to watch

[R.G.]

I got off into thinking about what to look for in a tube amp to head off disasters - that is, to prevent the normal wear-outs from becoming expensive failures.

Let me preface this with: I know full well that it's not always possible to prevent or predict failures, and not always possible to stop them when they happen. But there are some things that are indicative of problems, and some things that need taken care of immediately.

I'm just thinking these over.

The model for this is that there is an incarnation of Maxwell's Demon sitting inside the amp, checking over and over for any problem you've told him/her/it to watch, and reliably squawking when one of these conditions is noted.

Some things I've thought of already:
- power tube overcurrents of sustained duration
- loss of the bias supply in fixed bias amps
- B+ levels too high or low
- B+ with too much ripple ( this indicates either overcurrent, poor function on primary filter caps, or loss of a rectifier in a full wave setup)
- temperatures too high; likely targets are surfaces of the power transformer, output transformer (+choke??), air temp in the top of the cabinet; others?
- AC input power too high - too many watts going into the box

What else?

[defaced]

-No load on OT secondary
-Excessive power tube screen current
-Excessive power tube grid current (for the 0.0000001% of amps out there that can run in AB2)
-Excessive current draw on heater or other secondary (my guess is a heater short would blow the mains fuse, but other windings might not)

Kinda in parallel with this, but not really error checking, more error preventing:
-Inrush current limiting

[R.G.]

-No load on OT secondary
-Excessive power tube screen current
-Excessive power tube grid current (for the 0.0000001% of amps out there that can run in AB2)
-Excessive current draw on heater or other secondary (my guess is a heater short would blow the mains fuse, but other windings might not)

Kinda in parallel with this, but not really error checking, more error preventing:
-Inrush current limiting

Mmmm. Good ones. Thank you.

But you're backwards on the heater short - a heater short is least likely to blow a mains fuse, because the voltage has been stepped down so low that it changes the primary power (and hence current) less than the B+ winding, so it's least likely to make the primary current big enough to pop a fuse. Maybe, could happen, but you can't count on it.

[nickb]

Impending spilled beer detector?
Approach of moron with hot soldering iron alert?

More seriously, Failure is often predicated by abnormal heat. Now, CMOS IR cameras and micros are pretty cheap...

[trobbins]

Perhaps:
- high mains ac voltage
- dc grid voltage for cathode biased stage
- wrt 'too much ripple', the key failure/imbalance issue is relative magnitude of mains frequency ripple voltage, versus 2x mains frequency voltage.

[NateS]

Good catch on excess ripple. Watching heater current can also tell you if a tube is already dead, or missing, or possibly the wrong type. Does transconductance go crazy with a tube about to die? I wouldn't expect both power tubes of a pair to go out in unison, so the bias balance should catch this one. For no load on the secondary, it seems like some of the protection circuits discussed in the past would be better than monitoring.

On the temp monitors, I had initially considered NTC thermistors, especially since I found one with a lug, but MCP9700 is $0.25 in qty 10, and TO92 will fit inside the barrel of a 4ga crimp ring. It's essentially a temperature programmable voltage regulator. When I get around to temp monitoring, I was mostly planning to monitor my FET's, but that's largely 2nd guessing design, rather than bullet proofing. For bullet proofing watching the transformers makes a lot of sense.

[Steve Conner]

I second the call for some sort of wheel or leg system that would enable the amp to run away when it sees a guitarist with a soldering iron.

[NateS]

A little while ago we were mocking the RTC, but in 98% of RTC applications, they are simply used in data loggers, not control.

If you're doing warranty work, a datalogger might be nice. Ohhh last Friday, we saw an extended brownout condition. Are you using a variac? And those tubes you bought on ebay only lasted 200 hrs.

[R.G.]

Impending spilled beer detector?
Approach of moron with hot soldering iron alert?

For some things, Maxwell's Demon is not enough - you need to find a deity.

More seriously, Failure is often predicated by abnormal heat. Now, CMOS IR cameras and micros are pretty cheap...

Yes - heat!

Perhaps:
- high mains ac voltage
- dc grid voltage for cathode biased stage
- wrt 'too much ripple', the key failure/imbalance issue is relative magnitude of mains frequency ripple voltage, versus 2x mains frequency voltage.

Ah! The relative sizes of mains frequency voltage versus 2x mains. Good point! How about also the ratio of mains voltage to ripple?

Watching heater current can also tell you if a tube is already dead, or missing, or possibly the wrong type.

Yep.

Does transconductance go crazy with a tube about to die?

I think it mostly just drops a lot. But I don't really know. The actual method of death probably matters. An open heater means gm goes to zero, and poisoned/worn out cathode means it just drops. Shorted screens can do a lot of different things.

I wouldn't expect both power tubes of a pair to go out in unison, so the bias balance should catch this one.

That's a good one! If the bias needed for a tube is over X different from the other one(s), that means it's not matched. If it *changes* much over life, then it may well be dying - loss of cathode emission is what this catches.

For no load on the secondary, it seems like some of the protection circuits discussed in the past would be better than monitoring.

I think you're right. I can't see any good way of doing anything about this one, unless you have managed to conjure up a particularly burly, manly version of Maxwell's Demon.

TO92 will fit inside the barrel of a 4ga crimp ring.

Not to lessen the goodness of the MC9700 comment, but this is a jewel. A sturdy way to reliably mount an isolated temperature sensor device is really, really a great idea. The immediate thing that comes to my mind is the use of the DS18B20 temperature sensor. It comes in a TO-92 package and allows you to sense a semi-infinite number of points on a two- or three-wire string with the one-wire protocol. It does temperatures to nine or twelve bits directly.

[NateS]

A sturdy way to reliably mount an isolated temperature sensor device is really, really a great idea. The immediate thing that comes to my mind is the use of the DS18B20 temperature sensor. It comes in a TO-92 package and allows you to sense a semi-infinite number of points on a two- or three-wire string with the one-wire protocol. It does temperatures to nine or twelve bits directly.

I haven't actually crimped one, but I expect the plastic would hold up. You can always crimp it to mandrel first if it's too fragile. Slather it with thermal grease, crimp/heat shrink/hot melt. as needed. DS18B20 offends me on pricing. It really comes down to how many sensors. The per unit price works against the savings from 1 wire. What needs monitoring? OT,PT. Any regulator, or just ones that typically block high voltage, or low current, but couldn't handle both? (If you're dropping 200V off the screens full throttle voltage, and it starts sucking down more current than you can safely throttle that low?)

I hadn't considered the frequency content of B+. What's the mechanism here? I guess I should pull out my OCD filtering on the B+ Voltage sensor now.

If you want to get really dumblesque about soldering irons, a discretely packaged light sensor (hint, LED's can be used as inputs) + whatever level of sadism you subscribe to (you probably don't want thermite charges in a consumer product,.)

[R.G.]

I haven't actually crimped one, but I expect the plastic would hold up. You can always crimp it to mandrel first if it's too fragile. Slather it with thermal grease, crimp/heat shrink/hot melt as needed.

I would get a steel rod or bolt and file it to shape as a mandrel, then crimp. A few tries with a file/sandpaper and then trying the fit on TO92 would result in a nice, snug fit with some thermal grease or catalyzed thermal epoxy.

DS18B20 offends me on pricing.

Me, too, but it does offer some technical advantages. I use geothermal water loops for heating/airconditioning. The conditions of the heating and cooling can largely be read off from the incoming/outgoing water temps, and the DS18B20 works GREAT for that, and I don't need to worry about wiring so much over a couple of hundred feet of wire.

It really comes down to how many sensors. The per unit price works against the savings from 1 wire. What needs monitoring? OT,PT. Any regulator, or just ones that typically block high voltage, or low current, but couldn't handle both? (If you're dropping 200V off the screens full throttle voltage, and it starts sucking down more current than you can safely throttle that low?)

Yep. Even a bimetal clicker would work if you could live with the premade temp settings. And an analog temp reader works as well, just takes an A-D channel for each one, or an external multiplexer.

I hadn't considered the frequency content of B+. What's the mechanism here?

The big one is that if you lose half your rectifier, the ripple frequency halves.

If you want to get really dumblesque about soldering irons, a discretely packaged light sensor (hint, LED's can be used as inputs) + whatever level of sadism you subscribe to (you probably don't want thermite charges in a consumer product,.)

Thermite's not fast enough. I'd prefer something like sirens and blindingly-bright strobe flashes. Something like the warnings in the Nostromo when Ripley set off the self-destruct sequence.

[trobbins]

[QUOTE=NateS;325447]I hadn't considered the frequency content of B+. What's the mechanism here?QUOTE]
Further to RG's comment, as a fullwave rectifier moves from fully balanced diode conduction to one diode not conducting at all (ie. half wave rectifier), the relative magnitude of 100Hz ripple voltage compared to 50Hz ripple voltage changes. Comparison technique makes the detection method insensitive to operating power level (except for zero power level, but most amps have some idle power level).

A DSP form of detection would no doubt be easiest, but I've used two simple analog filter circuits tuned to 1x and 6x mains frequencies, and then rectified and filtered and sent to a comparator to identify when an industrial 3-ph rectifier or battery charger using SCR regulation has a bad rectifier type fault.

[Merlinb]

What else?

Excessive prorogued current imbalance in the OT primary halves, which would lead to permanent saturation.

HF oscillation?

[R.G.]

Excessive prorogued current imbalance in the OT primary halves, which would lead to permanent saturation.

Do you mean permanent remanence in the iron? Or just that the operating conditions would be off, perhaps leading to transformer heating until this was corrected.

Most OT iron is deliberately made as magnetically soft as it can practically be, and tends to have low remanence - which makes sense; it's a poor permanent magnet material, easily de-magnetized. So I'm thinking you're referring to too great an imbalance in operation, which would cause heating losses to go up in the OT, as well as being pretty tough on the tubes driving the saturated side.

HF oscillation?

That's a good one. Easy enough to sense by the same method as sensing ripple off the B+, but with a much higher pass frequency.

[NateS]

Plate current = primary current, so if you LPF that, you can analyze balance.

I have doodled and sketched HFO detction until HFO. I presume you mostly care about the output stage/global. In the end though, this feels like more of a design issue than a likely failure mode. Especially if you're aggressively discarding content you don't care about, It could however do double duty for catching spiking on the primary (back to the no load case), but since it can cause transformer death, maybe interesting.