View Thread

a1970gto wrote:

Are you using the y type tri-clamp as your bean extraction? How is that working and could you post a video?

cdrake39 wrote:



Correct - there is a mesh sleeve inserted into the wye so the beans don't drop onto the element. It work's okay, but there is room for improvement. You have to wiggle the handle for the valve quite a bit to get the beans moving out the chute. I'll post a video next time I roast (waiting on new thermocouple so early next week)

jbrux4 wrote:



Ummmm, any way you could walk me through your bean dump? Parts and set-up?

SWEET BUILD!!!! Nice use of what looks like a PC case.

cdrake39 wrote:



Thanks!! It is indeed a PC tower - all the electronics tuck away nicely underneath.

The bean dump is comprised of a tri-clamp butterfly valve which has been bored out and a mesh screen secured over the top. Depending on blower power, you may be able to just drill a bunch of small holes - but that didn't work for my application. When the valve is closed, air can pass through the screen. When the valve is opened, the beans are able to drop out the chute (which is lined with another mesh screen salvaged from a kitchen strainer). It works ok, but the beans sometimes get hung up and you have to wiggle the valve handle a bit. I think a gate valve would work much better and is something I plan to add at some point down the road. Hope that helps!

jbrux4 wrote:



It helps, but I am in need of the visual to put it all together. Could you provide detailed pics or something?

floptronica wrote:

I think your roaster is a thing of beauty, but I can't help thinking that with a bigger blower and that 4" roast chamber, you could do way more than 200g at a time. Is a capacity increase in the plans for this baby?

renatoa wrote:

Even with a zero I term, the behavior should be different when back to normal profile.
P only does not mean fixed power output.

renatoa wrote:

Correct, I mean this part: "25 degrees too low"

The output of a PID controller, when the I and D terms are zero, is given by the formula: output = error * kP

Let's suppose for 150 C degrees setpoint you need 60% power, and kP is 5
In this case, the temperature will stabilise at 138 C, 12 degrees lower that 150 setpoint, because you need that 12 degrees error as an input for the PID, to deliver the 60% output. 60 = 12 degrees error * 5 (kP)
If temperature would increase, then error will decrease, output will decrease too, and temperature will step back to reestablish equilibrium.

If increasing kP, error will decrease, and you can go closer to setpoint, but never reach it. Also, increasing kP make your system sensible to oscillations if you have a sudden outer influence.

jbrux4 wrote:



It helps, but I am in need of the visual to put it all together. Could you provide detailed pics or something?

cdrake39 wrote:

So would you think that decreasing Kp slightly would help smoothen out this curve below? Right now my PID values are set to 2, 0.23, 0.05

renatoa wrote:



I think the P coefficient is low enough as is now, and oscillations are rather from too high I term.
It's easy to find: set I and D to zero, and apply a sharp power step impulse, about 10% of range, let's say from 40 to 50%.
You have to be stable at 40% power when doing this.
Observe the evolution, if having oscillations decrease P, if not increase.
Don't use small steps, be generous with value changes, double or halve, else will waste too much time to find the right values.
So, try P values by succesive doubling, like 1, 2, 4, 8...
If 4 has no oscillations, but 8 start looking as above, then lower to 6 and try again, then finally either 5 or 7 should be the winner. The goal is to reach the highest P with no oscillations.
For the last value you have oscillations, measure the time between two peaks, this will give you KI, as Tu/2, where Tu is expressed in minutes. For example if oscillation period is 5 seconds, then KI = (5/60) / 2 = 0.04

cdrake39 wrote:
The bean dump is comprised of a tri-clamp butterfly valve which has been bored out and a mesh screen secured over the top. Depending on blower power, you may be able to just drill a bunch of small holes - but that didn't work for my application. When the valve is closed, air can pass through the screen. When the valve is opened, the beans are able to drop out the chute (which is lined with another mesh screen salvaged from a kitchen strainer). It works ok, but the beans sometimes get hung up and you have to wiggle the valve handle a bit. I think a gate valve would work much better and is something I plan to add at some point down the road. Hope that helps!

cdrake39 wrote:
Sorry for the delay in getting back to you on this. To give a good understanding of how it is assembled I figured this schematic would explain better than photos. Let me know if you have any questions, or suggestions on ways to improve!

renatoa wrote:

PID oscillations... can be seen in power evolution too.
Lots of hesitations, instead a firm increase, as is the profile shape.

PID tuning would not change this behaviour too much, tuning is done for a specific slope and setpoint, not for a permanently variable setpoint. Outside the range where you performed the tuning will start oscilate again.
Also, the actual PID philosophy impose having over/under-shots, in order to build integrative term.
A code change is required, to compute integrative term taking into account also the second derivative, i.e. RoR evolution. More exactly, when you have a rising error slope, it's ok to add the error to I term. When this slope ends, and the error become flat, no more add to I term, even if error is still there, it's a sign an error decrease will follow.
Much less oscillations with this trick.

renatoa wrote:

Let's continue this as PM if you want, then start a new thread with conclusions.
But for now I will post in TC4 main thread a finding that is really simple, but with visible effects, and I think it should be immediately addressed in the main TC4 PID code.