Reflow oven first tests with questions...

[alank2]

Hi Everyone,

So I'm ready to make the jump to smd and try my hand at some stenciling and baking.  I worked on an avr based PID controller last year with a converted toaster oven that has 2 elements on top and 2 elements on bottom.  My first test shows that I tweaked the PID algorithm pretty well for the sensor in air (not attached to a PCB), but then I thought I'd attach it to a small test pcb with a paperclip and see how that would go.  Here are some pictures and I've got some questions at the end for anyone who has experience here!

I am using a profile that goes like this:

warmup ramps to 150 deg C @ 3C/s and the stage does not end until 90s after the the temp reaches 100C
reflow ramps to 200 deg C @ 3C/s and the stage does not end until 90s after the the temp reaches 185C
cooldown ramps down to 25 deg C @ 6C/s

Oven does have a convection fan I run all the time during the cycle.

My oven has a metal pan that sits on top of the metal bars so I left it in, but I wonder if I'd get better results without it (allowing the bottom two elements more direct access to the pcb's) - So this is question #1.

series 1 is PV (present value, temp), series 2 is SV (set value), series 3 is CO (controller output).  The CO is 0 for completely off to 120 for completely on (120 half cycles @ 60 Hz):

Sensor placed about 1" above the metal tray.  I suspect when I was playing with this last year, I must have tweaked the PID parameters by putting the sensor in air:



Sensor paper clipped to a small pcb:



Question #2 is - should I be taking the temperature of the pcb surface?  Is this the best place to sense?  Assuming this is true, I'm going to need to go back and tweak my PID parameters for this condition to improve performance.

Question #3 is - I am a bit surprised as the temp spikes with it was attached to the pcb.  I expected the pcb to dampen the temperature changes much moreso than the air but yet between 400 and 500 seconds it was a spiky up and down deal.  Does this make sense?  Why would this be?  My sample time is 5s between pid operations so each temperature is an average of all the readings the unit took for that 5s (4 per second so 20 measurements from a max6675).

Any other thoughts?

Thanks,

Alan

[Paul Price]

Looks like you might have some problem with the temp sensor or the wiring/interface to it.

[alank2]

Looks like you might have some problem with the temp sensor or the wiring/interface to it.

I'm using a max6675 on a breadboard.  The above is a 5s average of 20 measurements from it.  Maybe I need to get a list of 250ms measurements and see what they look like to see if there is error or bad readings coming through...

Thanks,

Alan

[Paul Price]

Just how does this oven work for you? Does it do the heating/cooling/soldering correct job if you feed it a PCB?

[alank2]

I haven't tried yet, I want to get the profile right before trying...

[Paul Price]

Your first graph shows almost an ideal response, that is why I cannot understand the erratic readings from the second picture of operation.

[alank2]

Your first graph shows almost an ideal response, that is why I cannot understand the erratic readings from the second picture of operation.

I know, I'm going to run another test with the probe pressed against a pcb again and see if it turns out the same or similar.  I know I need to tweak the PID parameters for the sensor being on the pcb vs. measuring open air.

[Paul Price]

You got it right in the first graph shown..revert to that software and probe configuration. You second graph shows rapid oscillations of control output. This means you've either made a big tweak in your software or your temperature probe is gone sour.

[alank2]

The difference between the first and second is that the first one has the probe held in air, touching nothing.  The second has the probe bead pressed against the board with a paperclip.

[Paul Price]

This does not seem to explain the entirely different control output behavior, I say try it again with the probe off the board. If that works the same as  your first diagram you are ready to cook PCB pie.

Could the probe tip be possibly grounding or shorting out to something on the PCB on the hot seat?

[AlfBaz]

This does not seem to explain the entirely different control output behavior

Wouldn't attaching the probe to the pcb add thermal inertia, thereby screwing the PID params that worked in free air?

[alank2]

Could the probe tip be possibly grounding or shorting out to something on the PCB on the hot seat?

I wonder about that, I'm going to repeat the test, maybe it was grounding.

I did a CO change test today with the sensor on the PCB.

My controller output CO can be from 0 (fully off) to 120 (fully on, 120 half cycles, 60 cycles).  I first set the CO to 5 and let it stabilize, and then changed it to 15 and let it stabilize.  I'm going to try to re-tune the PID parameters using this and see how it goes.

[alank2]

Wouldn't attaching the probe to the pcb add thermal inertia, thereby screwing the PID params that worked in free air?

I think you are right, absolutely, working on returning now.

[hamster_nz]

Does anybody have a rough analogy for what is going on?

I always assumed that it worked like this



Because most of the heat is IR and can go through air without heating it, at a first approximation it looks like this:



With the air temperature behind a low pass filter caused by the metalwork (and your PCB). I think this explains the graphs quite nicely (with free air being smoother, and the PCB surface being spikey.

[Paul Price]

Some Ideas to Consider

These thermal diagrams makes some sense but how does this help establish real-world values to the variables expressed so much more qualitatively than quantitatively.

A probe thermally interfaced to a PCB with paper-clip would not induce the wild CO oscillations you show. Show me how this could be possible. Could just turbulent air cause this? The temperature on/off CO hysteresis must be very much too low.

Look at it this way. The internal temperature measured by in free air is actually feeding a integrator (the inertia of the PCB's thermal mass). What this means is that if the target temperature is achieved in air, it is only a matter of seconds or minutes before the PCB warms/cools itself to the same temperature. That's all there is to it.

What this means, if you give the PCB enough time to rise to the hot air temperature, you can then use this information to derive an algorithm to determine how long the new plateau of temperature must be sustained to raise the board to very close to the same temperature of the internal air (at the critical solder paste melt temperature stage of this thermal joyride).

The temperature will rise from it's its present temperature to 63.7% of the air temperature in one thermal time-constant and the thermal time-constant is directly proportional to the thermal mass of the PCB. in 5 time-constants the board will reach 99.9% of the air temperature.

But this may all boil down to that you could be able to attach and shield(from moving air. it is the board temperature you interested in, not the turbulent air) one or more temperature probes that are thermally intimate with tiny spots on the PCB(using thermal heatsink compound) surface and then use these temperature readings  to tell if the board, in fact the whole board, is at the right temperature.

.

[hamster_nz]

Some Ideas to Consider

These thermal diagrams makes some sense but how does this help establish real-world values to the variables expressed so much more qualitatively than quantitatively.

A probe thermally interfaced to a PCB with paper-clip would not induce CO the wild oscillations you show.

Look at it this way. The internal temperature measured by in free air is actually feeding a delay line (the inertia of the PCB's thermal mass). What this means is that if the target temperature is achieved in air, it is only a matter of seconds or minutes before the PCB warms/cools itself to the same temperature. That's all there is to it.

What this means, if you give the PCB enough time to rise to the hot air temperature, you can then use this information to derive an algorithm to determine how long the new plateau of temperature must be sustained to raise the board to very close to the same temperature of the internal air (at the critical solder paste melt temperature stage of this thermal joyride).

The temperature will rise to 67% of the air temperature in one thermal time-constant and the thermal time-constant is directly proportional to the thermal mass of the PCB. in 5 time-constants the board will reach 99.9% of the air temperature.

.

Air isn't what is doing the warming in these small ovens, it is the IR radiation from the heater than does most of the heating, and air is only really involved in taking heat away from the the board. And even blowing air in at room temperature you cant can't cool a PCB off as quick as it heats up.

You can prove this if you don't use a tray under your PCB when doing reflow - the area of the PCB that are shadowed from the elements  by the wires of a metal shelf will be much cooler, leading to uneven soldering - likewises you get a pattern of lines on the bottom of any toast you make (but do this before you use the oven for its true purpose in life...).

[Paul Price]

Fascinating, but look at the edit I made to what you've already replied to. Will multiple probes upon spots on the board exact the temperature of the board? Would it be necessary to engage in dithering the PCB's position to ensure even heating?

I am soon ready to attempt to build one of these thermal joyrides myself.

[alank2]

New PID parameters improved things quite a bit so far before any tweaking:

[AlfBaz]

New PID parameters improved things quite a bit so far before any tweaking:

Nice...
Doesn't look like you'll get any faster heating based on you controller output. I'd be inclined to increase your "soak" time to ensure "heavier" components are up to temp before going for reflow.

If you have any parts that are borderline crispifying from prolonged reflow temperature, perhaps an alarm at the end of reflow to remind you to open the door for a quicker cooling ramp

[Paul Price]

The story looks good. Howja doit?

Where did you place the thermocouple?

[alank2]

Thanks AlfBaz - thanks for the tips!

Paul - thermocouple is clipped to a pcb and setup with a little tension to push the bead of the couple into some soldermask.  I changed the PID parameters to PGain from 5 to 3.25, IGain from 0.002 to 0.005, DGain from 100 to 50.

I then repeated the test with DGain double to 100.  It looks like the CO was a little jumpier, but it didn't overshoot as bad:

[hamster_nz]

I'm a little confused at the lag between the green and blue lines - are they all aligned in time? I have only a few seconds between power on and temp change (basically the time for the heating element to warm from cold) but yours looks like a bit more.

My oven was just the smallest, least expensive domestic bench-top oven I could find. Maybe yours is different...

How I avoided overshoot in my design was to have a set point about 5 degrees below the target temperature, active for 10 seconds, that held up the profile until it's target temperature was hit. This is then followed with the desired temperature / hold time.

So if aiming for 125, it hits 120, the heater goes off and it  'coasts' for 10 seconds up to the target temperature before the 125 setpoint becomes active.

I managed to get an OK profile

[StrangeloveMD]

Nice work thus far!  Luckily I've only just been looking in on this whole "temperatures change during the reflow process" thing, so I can actually find some of my bookmarks...

Attaching Thermocouples:
Tis from '98, but good.

LEDs are fun.

TCs in General:
According to OSRAM, thermocouples do exist.  We are on the right track.

And of course, the real opus of concern; IPC-7530: Guidelines for Temperature Profiling for Mass Soldering Processes (Reflow & Wave)
The document shows its age, but mostly talks about temperatures in terms of air temp as measured 1~4cm above board topside (this will depend on the oven) as I recall.

Hope something in there is useful to you, as I believe a few of the concerns brought up in this thread are addressed.  I also think it may be prudent to preheat the oven as much as possible, and consider a board heater (or preheat mode for the oven) that lets the board slowly ramp up to an easier "starting" temp.


hamster_nz:  Thank you for posting your curve with the paste limits, it's rather my preferred format.  Now, if only we could all move on to plotting them as bounded by board limits.  We'd really be onto something!

[Paul Price]

How big was your board? What was on it..just resistors and caps..any MCU in 100+ leaded packages?

[hamster_nz]

How big was your board? What was on it..just resistors and caps..any MCU in 100+ leaded packages?

I've been building batches of half a dozen daughter boards at a time, each with about a dozen or basic bits on it - nothing harder than a SOT23-3..