Anatomy of a sBitx V2 final thermal failure
|
I know that a number of you have had PA failure attributed to the metal compression clamp coming loos on the V2. I finally saw this happen to one of my posts when screwing it down after replacing failed finals, but each time I opened up the box to replace the MOSFETs the clamp had always been tight. I simply assumed that because of this my blown finals must have had some other cause for failure besides thermal runaway. I had added a choke between my linear power supply and my DE and V2 thinking power spikes might be the culprit, and I also put back to back 12V 5 watt Zener diodes across the drains as suggested by Steve, N3SB.Then today I ran temperature tests on the V2 with continuous key down CW at 5 watts, 10 watts, 20 watts and 30 watts. See the charts. Mica insulators from the factory were used. The temperature sensors were TMP36 analog units, one pressed against one of the IRF24N metal tabs (the drain) and the other pressed against the heat sink less than a cm away. The temperatures were logged every second using an Adafruit datalogger shield connected to an Arduino Uno and stored to an SD card where the data were plotted on Microsoft Excel.
Note: the three lines from each sensor were not well shielded although they were twisted, and as a result the analog DC voltage coming from the sensors during transmit on 80M was slightly artifactual and altered (lowered) during transmit, although this effect was mitigated somewhat at lower power levels by using 0.1 uF bypass capacitors on the sensor lines. On the first three charts (5, 10 and 20 watts) you will see that the temperature reading jumps when the transmission is stopped on key up. This could be used in order to measure the differential between the temperatures measured during transmit and the actual temperatures of the MOSFET and of the heat sink. In the 30 watt trial I went key up several times because the power output over 2-3 seconds dropped precipitously to around 5 watts without any drop in input power or total rig current (which was about 8 amps). I assume that I was looking at impending failure of at least one of the IFRZ24Ns. Finally one of the IRFZ24N PA MOSFETs shorted. Then I cut the power to the rig. The V2 seems more prone to thermal runaway than my DE with its new ceramic insulators. Unlike the test of my sBitx DE when I used ceramic (beryllium oxide) insulators where the power level remained steady at about 30-32 watts and the PA temperature stayed below 60C even after a full 5 minutes, the V2 power output rose quickly even at 5 watts to about 8 watts and the temperature reached 60C after only 13 seconds. Thanks Terry, VK5TM for your message #103822 and the article describing thermal runaway and temperature inflection points. I have more beryllium oxide washers on order but when I learned that there were less dangerous alternative to beryllium oxide (don't grind them--the dust is extremely poisonous. Just stay away from them) I then ordered from Aliexpress some aluminum nitride wafers which appear to be cut for the TO-220 form factor. They're also much cheaper. Jack, N6LN And I should also emphasize that the V2 MOSFETs were very well pressed indeed against the mica discs and the heat sinks, and with thermal paste. My hope is that with more effective heat transfer to the heat sink a thermal runaway protection thermistor circuit will be unnecessary as it seem unnecessary in my DE which has BeO insulator, about 200 times more heat conductive than mica. For what it's worth my guess is that the so-called thermal inflection point may be around the operating temperature of my DE running continuously at 30 watts. |
|
Thanks David for the article. I have measured gate voltage in the past for other reasons. Of course we adjust it initially at the PA_BIAS pot so that current during PTT increases about 200-250 mA. That is, at threshold voltage. Of course if you look at the current draw for more than a few seconds you’ll notice that it drifts up. One could assume that’s at least partly due to thermal instability of the potentiometer, assuming that it’s carbon based. Unlike most metals carbon has a negative temperature coefficient. Furthermore the setting of that potentiometer is on a knife’s edge—a small change of a radial degree or two results in a huge effect on bias voltage and idle current. But as has been described by many the nature of MOSFETs in some operating ranges is that increased temperature leads to decreased threshold voltage, increased current and thermal runaway. That’s why I was about to try adding a thermistor based bias compensation circuit to decrease bias voltage as temperature increases. Happily my experience with my DE is that by increasing the thermal conductivity with a different insulator material I could keep temperatures from increasing so quickly and, it seems, perhaps that’s the easier and better way to prevent thermal runaway.?My point here is not to teach anyone the concept of thermal runaway but instead that (at least in my rig) repeated thermal PA failures may in fact be mundanely predictable with mica insulators. I had previously assumed that was unlikely as long as the clamps were held tight (and I believe that in my case they always were). I put this out there and ask whether my hypothesis is reasonable: whether many of the PA failures that have been reported could be prevented with different insulators that have better thermal conductivity. You may have seen the results I got by using beryllium oxide insulators on my DE. See message #103820. I have here included the before and after temperature graphs I showed in that post from my tests with the DE. I do not recommend using beryllium oxide because it’s very toxic if you breathe dust during grinding, which I avoided by a lucky snap to get a washer into two properly sized pieces. I have properly sized aluminum nitride insulators on order which have almost as good thermal conductivity as BeO and hopefully will keep the die temperatures low enough to prevent thermal runaway as I believe I have managed to achieve in my DE.? |
开云体育John‘s data are so convincing, that I think anyone who is able to should consider switching to the aluminum nitride insulators. ??I think this would make this a much more robust transmitter. ?? I have mine on order.? Last night, I recognized for the first time how nice the CW from keyboard transmitting is! ? It really works! ? I’m a little bit slow to learn all the nice features, but I’m seeing really nice ways to use this little radio, that I bought primarily just to teach me things about digital signal processing. ?? Gordon? On Jul 17, 2023, at 06:58, John Terrell, N6LN <N6LN@...> wrote:
|
|
Jack, Would you mind sharing the link for the aluminium nitride washers. I’d like to order some. Thanks!! Br Peter EI3JCB On Mon 17 Jul 2023 at 04:07, John Terrell, N6LN <N6LN@...> wrote: I know that a number of you have had PA failure attributed to the metal compression clamp coming loos on the V2. I finally saw this happen to one of my posts when screwing it down after replacing failed finals, but each time I opened up the box to replace the MOSFETs the clamp had always been tight. I simply assumed that because of this my blown finals must have had some other cause for failure besides thermal runaway. I had added a choke between my linear power supply and my DE and V2 thinking power spikes might be the culprit, and I also put back to back 12V 5 watt Zener diodes across the drains as suggested by Steve, N3SB.Then today I ran temperature tests on the V2 with continuous key down CW at 5 watts, 10 watts, 20 watts and 30 watts. See the charts. Mica insulators from the factory were used. The temperature sensors were TMP36 analog units, one pressed against one of the IRF24N metal tabs (the drain) and the other pressed against the heat sink less than a cm away. The temperatures were logged every second using an Adafruit datalogger shield connected to an Arduino Uno and stored to an SD card where the data were plotted on Microsoft Excel. |
|
For some reason the AlN insulators from AliExpress will arrive 1-2 weeks before the more expensive ones from Amazon would. Obviously Amazon doesn’t keep these in their own warehouse but subcontracts out to another seller.
For anyone who might want to do a temperature test as I did, go to Adafruit and search for data logger shield? and get a couple of TMP36 temperature sensors from there or from Amazon. Adafruit has a short Arduino sketch that can be edited for two sensors. Just delete the unnecessary part for a photocell. As I mentioned some RF gets into the sensor lines (these are simple analog sensors) but 0.1 uF bypass capacitors minimize the artifactual drop in sensor voltage on transmit. This effect is less on 80 meters but is much worse on higher bands. Again if you look at the temperature reading jump that seems to occur when you key up (measuring the real temperature) you can use the jump to interpolate what the voltage really is from the recordings during transmit. But the result I got is close enough to conclude that a better thermal conductor keeps the IRF24Ns significantly cooler, and that may be enough to keep things working. I set up this testing regime in preparation for doing some experiments to adjust the resistor network I was going to use with my new thermistors that I just obtained in anticipation of installing a thermal runaway protection circuit alluded to by Allison. Happily it seems that such a circuit may not be necessary in my DE since I installed insulators with more robust thermal conductivity, and I hope that when I get new aluminum nitride insulators in less than 3 weeks the same will be true of my V2. I’d rather not use beryllium oxide again for my V2. Jack, N6LN |
开云体育Great info! ThanksGordon Kx4z On Jul 18, 2023, at 06:47, John Terrell, N6LN <N6LN@...> wrote:
|
开云体育I have been following this? conversation and have a question about? that clamp on top of those FETs? Is that a? thick bar or just some light sheet metal? I would think that a stiffer bar might be better as thin metal would have a tendency to bow when the end screws are tightened down. If there is enough? room, a third screw could be put in the center to eliminate the bowing, if there is any. I have a couple of heat sinks? that use a U channel to clamp the transistors to the sink. Just curious. Barry ? Sent from for Windows ? From: John Terrell, N6LN
Sent: Tuesday, July 18, 2023 6:47 AM To: [email protected] Subject: Re: [BITX20] Anatomy of a sBitx V2 final thermal failure ? For some reason the AlN insulators from AliExpress will arrive 1-2 weeks before the more expensive ones from Amazon would. Obviously Amazon doesn’t keep these in their own warehouse but subcontracts out to another seller. ? |
|
toggle quoted message
Show quoted text
On Jul 18, 2023, at 07:03, barry halterman <kthreebo@...> wrote:
|
|
Yes Evan, the differential in temperature readings varies with the temperature reading. Does it change fairly linearly? As I mentions on higher bands the differential is worse. Another curiosity is that in the DAE setup, at least on 80 meters, the pseudo-jump in temperature reading on key up was negligible. And of course adding bypass capacitors at the Arduino substantially reduced this effect. I suspect the difference depends on precisely how the sensor lines are physically placed in the different rigs. The TMP36 data specs say the sensor line gives 750mV at 25°C and has a temperature coefficient of 10mV/°C. That’s supposed to be linear. With RF interfering it may not be, and as you can see the twisted lines for Vcc, ground and sensor no doubt have different levels of confounding RF depending on their haphazard positions. I have a couple of weeks before my additional insulators arrive from China or Ukraine (seems most of the BeO insulators come from Ukraine, like affordable doorknob capacitors; I imagine the cannibalize old Soviet equipment where safety concerns are lax) so I plan to use additional bypass capacitors as you also recommended at the sensor end, and I’ll try to find some shielded cable. As you can see I threw on a couple of snap on chokes on one line in a hurry but that didn’t seem to help. Maybe with a shielded cable a snap on choke will be more effective to prevent common mode RF current from getting to the Arduino.? Jack, N6LN |
|
Good question Barry. The bar is thankfully very thick and I don’t think bowing is an issue. I would expect cracking of the transistor case or stripping the screw threads before the bar bowed much. As you probably know the early V2s had a problem because the screws threaded not into the aluminum of the heat sink itself but rather into threaded inserts that could be pulled out if the screws were turned too tight. Look closely and you’ll see that the upper Phillips head looks slightly different from the lower screw. That’s because the upper threaded insert did actually pull loose and I had to repair it by using a longer screw and hold it with a steel nut on the back side. This required a half-baked and difficult attempt at chewing away som of the aluminum heat fins on the back side with a hand drill which is not very effective and quite frankly very unsafe. Allison suggested using a proper mill but I have no access to one. Anyway my long repair screw works even though it is not positions normal to the heat sink surface. Now I’m hoping the other insert doesn’t come out but I suppose it will. Another potential problem is that the screws probably have different expansion rates from the clamp and the heat sink, steel vs aluminum, with extreme temperature cycling and so even with lock washers the clamp force may weaken. I don’t know whether that is a legitimate concern. I would be happy to pay for a replacement new heat sink without the old inserts as I understand are now supplied with the new rigs being shipped out rather than taking the time to find a mill (don’t know anyone with one and never used one before) or risking injury fixing it myself if and when the other screw insert pulls loose.? Jack, N6LN |
John Terrell, N6LN
