Coffee Profile Attempts: A work in Progress.....
This page documents my attempts to improve my coffee roasting through exploring various roasting profiles, and some thoughts and experience about process control in a coffee roaster. (Updated 8/03
The profile shown below was done on a computer controlled West Bend Poppery I . It was my attempt to test out a profile suggested by Jim Schulman as a variant of that used in the HotTop drum roaster. It took me a few tries before the profile looked like this -- the large heat capacilty of the WBI makes it challanging to implement a specific profile on the bean temperature without a good bit of trial and error. The top curve shows the temperature of the heated air before entering the roast chamber, and the bottom curve shows the measured bean temperature -- determined by thermocouple at a height 1.5 inches from the bottom of the chamber, with the chamber tilted by a few degrees to induce a good circulation. The batch size was 180 grams.
The roast resulted in a very good balance between
body and character. Earlier attempts would generally sacrifice
one for the other. The main difference between my profile and
that suggested by Jim is the addition of a drying/stabilization stage
that brings the beans to 230 degrees. The profile suggested
by Jim was:
- as fast as possible to 300F
- 4 minutes from 300F to 350F (12.5F per minute)
- 2 minutes from 350F to 410F (30F per minute)
- 3 minutes from 410 to 440f (10F per minute)
- 4 minutes from 300F to 350F (12.5F per minute)
- 2 minutes from 350F to 410F (30F per minute)
- 3 minutes from 410 to 440f (10F per minute)
My profile comes reasonably close to this, with
the segment-changing-temperatures of 230, 300, 350, 410, and 435F.
The Original Control Algorithm.
Process control in the roaster required the simultaneous monitoring of a number of variables. A conventional PID control approach is used for any segment during the roast, that is, a target heater air temperature is determined every 1/2 second, and the heater power calculated to reach that level is implemented by switching the duty cycle of a SSR. A ramp of the heater air temperature can be easily achieved by changing the PID target temperature a given amount each 1/2 second. This aspect of the control process is no different than could be more easily achieved using a ramp and soak commercial PID controller.
In order to implement the overall profile shown above, however, the roasting control has to do things that I think lie outside what can be achieved with a stand alone ramp/soak controller. Consider, for example, the first segment of the roast shown above (from 0 to about 160 seconds). The heater air temperature increases at something like 35 degrees per minute in "Segment 1". What is "non-standard" about controlling the roaster is that the end of the segment is NOT fixed by reaching an endpoint heater air temperature, but instead by reaching a temperature of 230 degrees in the bean thermometer.
If we ignore the second segment in the above roast (a glitch in the fan flow prevented the computer from keeping the heater air temperature constant for a few seconds, as was the plan....), the complexity of the control algorithm is further illustrated by the third segment. This starts at about 175 seconds and 375 deg. heater air temperature (HA-Temp), and ends at about 185 seconds and 475 deg. HA-Temp. The purpose of this segment is to boost the temperature of the heater air enough to increase the heating rate of the beans. This is done as a fast ramp, so that the PID can maintain control. What I think is unusual here is that both the starting point and the endpoint of this HA-Temp ramp is unknown before the roast. The ramp HA-Temp start point is determined by the heater air temperature that is present when we reach a certain BEAN temperature, and the endpoint is set to be a certain number of degrees hotter than the startpoint. So while this ramp segment is more conventional, in that it has a beginning and end that is determined by the temperature actually being controlled, the values used for those endpoints cannot be set before srating the roast. (Changes in ambient air temperature and humidity can make considerable changes in these values).
Overall, segment endpoints have to be configurable to be either: a target heater air temperature, a target bean temperature, or a fixed amount of time. While I do believe that it might be possible to achieve the profile shown above with a conventional ramp/soak PID controller, the profile would end up being different if the ambient air changed, if different beans were used, or if the bean load were different. I am currently convinced that the more flexible approach used here, where segment endpoints can be set by a number of parameters, is necessary for making a reasonably repeatable profile in a process controlled roaster. If anyone is aware of a way to do this sort of control with a conventional ramp/soak controler --- Please let me know!! -- I would really like to learn about it.
The Current Control Alorithm:
While the control algorithm used for the roast above worked very well, it is somewhat difficult to reconfigure. For example, if I wanted the beans to increase by 20 degrees per minute between 300 and 350 degrees bean temperature, rather than the 12.5 degrees per minute used above, I would end up having to tweak the heater air ramp rate and endpoints for many of the segments. Not only would the segment actually altered have to change, but its changes would leave the roaster in a different starting point for the next segment, so that would have to be adjusted too.
This has proven unsatisfactory. Discussions of various profiles and their influence on the taste of the coffee, especially by Jim Schulmann and Mike McGuiness, have driven me to find a more flexible approach for the control process, one where I could tell the computer bean temperature ramp rates directly, and have the computer figure out how to achieve that. This would then allow me to better explore various profiles, in a repeatable manner, and help me better teach myself how to roast coffee.
I have recently been able to achieve this. The "data entry" for the profile has also gotten MUCH simpler. The steps the computer is told in order to get a profile similar to the one shown (I'll get the new data graphed up and posted hopefully soon) above is:
The hard part of getting the computer to do this was making it so that it could directly control the ramp rate of the coffee bean temperature. This was not straightforward. I ran into lots of stability problems. The most interesting part, at least to me, had to do with the difference between regulating to a temperature and regulating directly to a rate. This had me thinking hard about loop control, especially in terms of the closed loop frequency response. It turns out that to get a rate controlled in the same way that you would get proportional control for a temperature, you need to have an open loop gain that varies as 1/frequency! This sure suprised me for a while. So in order to get RATE control, I had to abandon Proportional-integral-differential control, and use instead: integral plus the integral of the integral!!!. Sounds nuts, but it does work, and even makes sense analytically.
This worked OK, and allowed me a stable control, but it did not provide accurate control. Adding in two additional elements into the mix got me control accuracy. The first was predictive determination of heater power required for different bean temperature ramp rates. This is based on measurements in each roast, and so automatically adjusts for bean type, bean load, ambient changes, etc.
The second "fix" was a term proportional to the bean temperature. It makes sense that you would need a certain amount of heat just to maintain a given temperature, and that more is needed to maintain a higher temperature. This is also measured each roast. So the total heater power if the sum of three terms:
1 -- A term simply proportional to the bean temperature.
2 -- A predicted heater amount calculated for the requested ramp rate
3 -- A fed back heater amount determined by the integral plus the "integral of the integral" of the deviation of the ramp rate from the requested ramp rate.
It would be possible to just use 1+3 to get control, and it would get the right rate eventually. But that approach is hampered by the relatively big time constant necessary for loop stability. But adding in term 2 makes it possible to change the ramp rate relatively quickly, and then get them precise (to a couple of percent) after the third term kicks in.
Overall this approach to roaster control works great. I will put up a plot showing 5 different roasts on different days with different bean types, that have nearly indistinguishable measured roast profiles. And changing to a different profile is now VERY easy. I believe that this control approach would work for almost any type of roaster.
My next project is to work on less pricey hardware......
Thanks for visiting, Tom G.
Please make comments or suggestions to: Toms-roaster@columbus.rr.com
I get my green coffee from:
The Original Control Algorithm.
Process control in the roaster required the simultaneous monitoring of a number of variables. A conventional PID control approach is used for any segment during the roast, that is, a target heater air temperature is determined every 1/2 second, and the heater power calculated to reach that level is implemented by switching the duty cycle of a SSR. A ramp of the heater air temperature can be easily achieved by changing the PID target temperature a given amount each 1/2 second. This aspect of the control process is no different than could be more easily achieved using a ramp and soak commercial PID controller.
In order to implement the overall profile shown above, however, the roasting control has to do things that I think lie outside what can be achieved with a stand alone ramp/soak controller. Consider, for example, the first segment of the roast shown above (from 0 to about 160 seconds). The heater air temperature increases at something like 35 degrees per minute in "Segment 1". What is "non-standard" about controlling the roaster is that the end of the segment is NOT fixed by reaching an endpoint heater air temperature, but instead by reaching a temperature of 230 degrees in the bean thermometer.
If we ignore the second segment in the above roast (a glitch in the fan flow prevented the computer from keeping the heater air temperature constant for a few seconds, as was the plan....), the complexity of the control algorithm is further illustrated by the third segment. This starts at about 175 seconds and 375 deg. heater air temperature (HA-Temp), and ends at about 185 seconds and 475 deg. HA-Temp. The purpose of this segment is to boost the temperature of the heater air enough to increase the heating rate of the beans. This is done as a fast ramp, so that the PID can maintain control. What I think is unusual here is that both the starting point and the endpoint of this HA-Temp ramp is unknown before the roast. The ramp HA-Temp start point is determined by the heater air temperature that is present when we reach a certain BEAN temperature, and the endpoint is set to be a certain number of degrees hotter than the startpoint. So while this ramp segment is more conventional, in that it has a beginning and end that is determined by the temperature actually being controlled, the values used for those endpoints cannot be set before srating the roast. (Changes in ambient air temperature and humidity can make considerable changes in these values).
Overall, segment endpoints have to be configurable to be either: a target heater air temperature, a target bean temperature, or a fixed amount of time. While I do believe that it might be possible to achieve the profile shown above with a conventional ramp/soak PID controller, the profile would end up being different if the ambient air changed, if different beans were used, or if the bean load were different. I am currently convinced that the more flexible approach used here, where segment endpoints can be set by a number of parameters, is necessary for making a reasonably repeatable profile in a process controlled roaster. If anyone is aware of a way to do this sort of control with a conventional ramp/soak controler --- Please let me know!! -- I would really like to learn about it.
The Current Control Alorithm:
While the control algorithm used for the roast above worked very well, it is somewhat difficult to reconfigure. For example, if I wanted the beans to increase by 20 degrees per minute between 300 and 350 degrees bean temperature, rather than the 12.5 degrees per minute used above, I would end up having to tweak the heater air ramp rate and endpoints for many of the segments. Not only would the segment actually altered have to change, but its changes would leave the roaster in a different starting point for the next segment, so that would have to be adjusted too.
This has proven unsatisfactory. Discussions of various profiles and their influence on the taste of the coffee, especially by Jim Schulmann and Mike McGuiness, have driven me to find a more flexible approach for the control process, one where I could tell the computer bean temperature ramp rates directly, and have the computer figure out how to achieve that. This would then allow me to better explore various profiles, in a repeatable manner, and help me better teach myself how to roast coffee.
I have recently been able to achieve this. The "data entry" for the profile has also gotten MUCH simpler. The steps the computer is told in order to get a profile similar to the one shown (I'll get the new data graphed up and posted hopefully soon) above is:
1) Preheat the roaster to 200 degrees, heater air temperature
2) Wait for any key being pressed (means that coffee is in!)
3) Ramp heater air at 20 degrees per minute, till coffee is 230 F
4) Hold bean temperature close to fixed for 30 seconds
5) Ramp beans at 60 degrees per minute to 300 F bean temperature
6) Ramp 12.5 degrees per minute to 350 F
7) Ramp 25 degrees per minute to 400 F
8) Ramp 10 degrees perminute to 430 F, or end on key-strike
Previously, it was impossible to "tell" the computer
to ramp bean temperature, I could only make it control the heater air
temperature that I had determined would work, but took many many more
"steps". Now a change in roast profile is EASY to do, any of the
steps after number 4 above can be changed and implemented in seconds.
And the computer gets the right bean ramp rate to roughly within
a couple of percent. (for example, tell it 20 degrees per minute,
and it will be be within 0.4 degrees per minute of that rate).
The hard part of getting the computer to do this was making it so that it could directly control the ramp rate of the coffee bean temperature. This was not straightforward. I ran into lots of stability problems. The most interesting part, at least to me, had to do with the difference between regulating to a temperature and regulating directly to a rate. This had me thinking hard about loop control, especially in terms of the closed loop frequency response. It turns out that to get a rate controlled in the same way that you would get proportional control for a temperature, you need to have an open loop gain that varies as 1/frequency! This sure suprised me for a while. So in order to get RATE control, I had to abandon Proportional-integral-differential control, and use instead: integral plus the integral of the integral!!!. Sounds nuts, but it does work, and even makes sense analytically.
This worked OK, and allowed me a stable control, but it did not provide accurate control. Adding in two additional elements into the mix got me control accuracy. The first was predictive determination of heater power required for different bean temperature ramp rates. This is based on measurements in each roast, and so automatically adjusts for bean type, bean load, ambient changes, etc.
The second "fix" was a term proportional to the bean temperature. It makes sense that you would need a certain amount of heat just to maintain a given temperature, and that more is needed to maintain a higher temperature. This is also measured each roast. So the total heater power if the sum of three terms:
1 -- A term simply proportional to the bean temperature.
2 -- A predicted heater amount calculated for the requested ramp rate
3 -- A fed back heater amount determined by the integral plus the "integral of the integral" of the deviation of the ramp rate from the requested ramp rate.
It would be possible to just use 1+3 to get control, and it would get the right rate eventually. But that approach is hampered by the relatively big time constant necessary for loop stability. But adding in term 2 makes it possible to change the ramp rate relatively quickly, and then get them precise (to a couple of percent) after the third term kicks in.
Overall this approach to roaster control works great. I will put up a plot showing 5 different roasts on different days with different bean types, that have nearly indistinguishable measured roast profiles. And changing to a different profile is now VERY easy. I believe that this control approach would work for almost any type of roaster.
My next project is to work on less pricey hardware......
Thanks for visiting, Tom G.
Please make comments or suggestions to: Toms-roaster@columbus.rr.com
I get my green coffee from: