Oh wow. In that database of pull metrics - did you ever analyze the statistics of the parameters?

E.g. are they uniformly distributed - or is the CBR gaussian with a sharp peak?

As you can imagine, I've been contemplating that task, as the data approaches useful volume. Its usefulness could be skewed for some parameters, by the tendency to 'freeze', or hold a value whilst varying others to reach a set point. This point (preferred flavour) can be achieved by more than one combination in the parameter set.
However, it shouldn't affect the distribution of coefficient brew ratio. It's a good subject and relatively easy for me to re-list. It shares a cell with a couple of other notes, which format I have been intending to change, so I'll just get on it and we shall see.

Its probably how my roaster arrives at his 'espresso recipe' - just pick a point on the CBR distribution curve that maximises bean consumption...

AssafL wrote:Oh wow. In that database of pull metrics - did you ever analyze the statistics of the parameters?

E.g. are they uniformly distributed - or is the CBR gaussian with a sharp peak?

Here's the normal distribution of Brew Ratio for two medium roast blends of a similar type: Honduras, Colombia, Guatemala, mainly Cattura and Bourbon varieties; all Bianca shots, with varying profiles.



5% intervals. X axis gives the shots per "bin" or interval. Mean:0.447. Standard deviation:0.028.

So, as far as this pull metric goes, yes it is a gaussian distribution and the standard deviation. is quite narrow.
When reviewing this, I looked at the taste notes for each shot by st. dev. All the shots with positive taste notes and "enjoyment ticks" were within one deviation each side of the mean, with one exception. Once outside this range, taste notes showed more negative comments.
So two quite small standard deviations in Brew Ratio covers my acceptable taste range for these coffees.

Do other coffees fall outside this range?

As coffee ages does the peak slide to the right?

Not outside that range, but other coffees have different mean B.R's and some have a wider distribution partly due to my trying more extreme shots and having inconsistent technique.

BR movement on aging is inconclusive for this pair of medium roasts. One roast was 15 days old to 23 days over the usage and the other ran from 38 days old to 43 days. So similar roasts and varietals but quite a difference in lifetimes.
A point worth making is that the shots I used for the distribution were all off one fairly consistent grinder, whereas I usually switch grinders between single dosing for myself on the M68 and a busy day with multiple shots for visitors off the Ceado doser.

I'll probably switch roasts to match grinders in future, so that I can build consistency!

What do you find in these respects? I imagine there's a lot of depth in your brew data?

I roast 200gr samples. So most of my pull data isn't for a specific roast.

Also I tend to force myself to periodically try new strategies such as darker roasting, brick wall preinfusion and lungo - ristretto territory. Just to try new flavors. Otherwise I end up in a comfort zone.

Hey all,

I've updated My short and concise build to preinfusion/pressure profiling with some new pictures...

Graham J wrote:This is probably somewhat off-topic, but does reflect a profiling shot development. Apologies to Jake_G!

No apologies needed! I love that this conversation is happening here.

Graham J wrote: Here's the normal distribution of Brew Ratio for two medium roast blends of a similar type:

You had me at "normal distribution" Do you find that you favor lungo shots, or do you find that your shots simply blond later than a 1:2 ratio?

robertmw wrote: I think the ideal solution for controlling temperature of an E61 grouphead, while maintaining the majority of the existing design of using boiler water going through the thermosiphon, would be to have a controlled valve on the thermosiphon loop.

Yes. To quote another user: "A thousand times yes!". I think the cat's meow would be to have LM's thermostatic mixing valve with either bi-metallic control as Assaf suggests, or just take it over with a PID on the needle valve to control group temp directly. You would need a state machine to make full use of this, with passive temperature management during idle periods (no cool water available since the TS loop is closed circuit) and active group cooling available when flushing or pulling a shot.

another_jim wrote:The ideal is a cascade control where a sensor on the group resets the boiler's setpoint to maintain a desired temperature. These were kind of tricky when using two old school PIDs, since even a hair alteration on the integral controller of the outer loop (in this case the group) could sent it into oscillations. With programmable controllers like an Arduino, it's a piece of cake. Every ten minutes or so, change the set point of the boiler to bring the group about 2/3rds of the way out of its error. (i.e. if the group is at 90 and you want it at 93, raise the boiler temp by 2 degrees. 10 minutes later, if it's still 1 degree too low, change the boiler by 2/3 degree, etc etc)). That will create a massively stable control with the group always settling in close to the desired temperature.

I'm familiar with these control loops and one thing cascade loops require is a state machine to be reliable or at a minimum manual control for startup. One question I have about your particular suggested scheme, is how does the set point get back down? I see how adding up the steady state error can basically raise the offset, so that 93°C set point can trigger a boiler PID target of say 105°C when 103°C isn't cutting it, but what happens when the group hits 94°C? Is there another 10 minute latency while we wait for the boiler to cool to 104°C? I don't mean to drag us through the weeds, but the lop-sided nature of passive heat transfer (heating element MUCH more aggressive than natural convection cooling) does leave one wondering how you balance it out. Hence my interest in tempering the HX loop with cool water...

Tonefish wrote:With PIDs and electronic control this should not be a problem, after all, dealing with the issues you describe above is what PIDs do

As others have pointed out PID fails at controlling this without additional controls to keep everything in line. I like the idea of having group temperature feedback into the control system, but I don't see an easy way to incorporate this without a microcontroller...

So many good thoughts, so little time.

Cheers!

- Jake

Jake_G wrote:As others have pointed out PID fails at controlling this without additional controls to keep everything in line. I like the idea of having group temperature feedback into the control system, but I don't see an easy way to incorporate this without a microcontroller...

As mentioned above, PIDs with electronics shouldn't be a problem. My $19.99 drone has a microcontroller, so what's the big deal?

Well, okay, maybe too much to ask for now, but you don't ask and you don't get.

Jake_G wrote:I see how adding up the steady state error can basically raise the offset, so that 93°C set point can trigger a boiler PID target of say 105°C when 103°C isn't cutting it, but what happens when the group hits 94°C? Is there another 10 minute latency while we wait for the boiler to cool to 104°C? I don't mean to drag us through the weeds, but the lop-sided nature of passive heat transfer (heating element MUCH more aggressive than natural convection cooling) does leave one wondering how you balance it out. Hence my interest in tempering the HX loop with cool water...

I was unclear. On a programmable controller, there is no need for the outer (aka supervisory) loop to be PIDish.
-- The boiler is controlled by a PID. However, the desired controlled temperature is the thermosyphoned group (or a group heated through via other long lag devices by the boiler, e.g. a massive HX (Cimbali, La Spaz), a plate or direct bolt-on (Gaggia, Rancilio, Reneka) or a flooded group (LM, Synesso)). This group sits at a temperature below the boiler, but the difference depends on lots of factors and is variable.
-- The supervisory loop monitors the temperature difference between group and boiler setpoint (delta) and sets the boiler at desired group temp + delta. When the desired group temp changes because the person changes it, it changes the group temp immediately. But it changes its intenral delat variable very slowly, based on a long log. Depending on how smart it is, it can take use into account (if the controller knows the pump status), or the room temperature, or logs from 3 weeks ago, etc, etc. Memory is cheap, analytics are cheap, so there's no earthly reason to use PID loops for this kind of thing except a lack of imagination in the designer.

Jake_G wrote:
You had me at "normal distribution" Do you find that you favor lungo shots, or do you find that your shots simply blond later than a 1:2 ratio?

OK - I used "normal" as defining what physicists like to call a Gaussian distribution and math and stats people call a"normal distribution". Symmetrical about an axis, generated only using mean and Std.dev. etc. Not "normal" as in widely prevailing, or usual. Best to be clear, sorry if this is not your point! This dataset was abstracted responding to Assaf's question on looking at distribution of brew ratios.My response was originally to questions on setting up shot parameters on a profiling machine. This is how threads "ramble" ..
I don't think this preferred B.R. is usual, outside of the particular beans and roast. For example, my preferred BR is centred around 1:2 on my current Finca La Aurora, Mexican, fully washed, light/medium roast. However the subject of the distribution graph was two medium blends which I found best at a B.R. of 1:2.2.

I think the more useful points to share are these:
1.A single standard deviation each side of the mean covers 95% of the good tasting shots.This is about 3.5g of water or 1.6g of coffee, so significant taste differences arise at ½ these amounts. For this taster. On these beans. But other people can easily test this, if they keep a pull or brew data log.

2.Getting back to the long and rambling path to p.i. and profiling - data from brew records is super-useful for setting up profiling shots and avoiding getting lost in all the variables.
Once I have an acceptable and consistent tasting espresso, I can progress to varying the shot profile, while keeping the B.R. constant within ½ standard deviation. This creates a lot of taste interest. After varying p.i. and varying end pressure, I'll have a good idea of the taste range and can then loop back to changing B.R. and temperature, as extra interest.

This may reflect the fact that I'm not all that experienced in this area! However it's proving to be a good base to work from to study variable p.i. and shot profile.

Chapter 19: Something's wrong with the machine. The water won't come out.

Ok.

It's the middle of October, 2018. I've been getting along with my S20 pretty well for nearly a year now using some form of preinfusion or another to be more forgiving of my puck prep, and pick up some flavor notes here and there. For the last 10 months, the has been achieved through line pressure preinfusion. Basically all I did was install a kill switch to the relay that controls my pump and leave it off until I want to extract at full pressure. As an added benefit, this also means the Gicar controller can't turn the pump when it wants to fill the boiler, either. The machine has been very quiet this year.

Just this week, I finally worked up the nerve to cut a hole in my gicleur cap and get the valve installed. Everything fit, but there were some complications. First, I couldn't adjust the flow nearly low enough. It bottomed out at 140g in 30s with the pump running. I reasoned it out that this must be due to water leaking in behind the stainless steel bolt that I fashioned into a threaded bushing to adapt the 8mm threads of the S20 to the 5mm threads of my needle valve. After all, the gicleur I removed had only a 0.5mm orifice and it flowed 250g in 30s, so I knew the leak was smaller than that. I left it there overnight and pulled a relatively non-eventful shot in the morning.

When I asked Heidi how her morning coffee was yesterday afternoon, she said:
"Something's wrong with the machine. The water won't come out."
I said: "I know. It's on purpose. How was it?"
She said: "No. Something's wrong."

Well, she was right. I walked in and found the group dripping incessantly into the drip tray (good thing I'm plumbed in and out!). When I tried purging the group, water barely trickled out... I shut off the water supply and pulled the group cap off to find that my 3D-printed bushing had disintegrated under the heat and pressure of the group. Whoops. I pulled the valve and put the gicleur back in, and popped my "just in case" plug into the cap to plug the hole. It leaked just a little bit. I started to warm up the machine and kicked the pump on to see how it would hold the pressure. The pump pressure seemed to seal the plug as I thought it might and I let the machine warm up a few before checking it again. This time when I kicked the pump on, I saw the plug move. But the plug was all the way into the cap... pump on, plug moves. Towards me. Through the hole in the cap. I pulled the cap and pushed the plug back out...

Crap.

Don't 3D print hot parts...

Alright.

Think!

First I grabbed a 6mm bolt with a flared hex head with an O-ring around the threads and plugged the hole with that. I popped the hex head with O-ring through the hole in the cap from the inside and secured it with a nut on the outside. It worked!

Then, looking at my needle valve arrangement, I installed an O-ring at the base of the threads on the bolt I was using for a bushing and thought again about how to seal the barrel of the needle valve against the bore of the gicleur cap. My design using two O-rings was sound, but the part was too small, and the cross section between the inner O-ring and the outer one was too thin. It just didn't have the strength it needed to work well.

Then it hit me. And shame on all of you for not calling me an idiot sooner!

I had two O-rings, each with a cross section of .070" effectively stacked on top of each other (with nothing more than what amounts to a thin candy shell between them) to seal between the barrel and the bore. These were "0" series o rings. They make 1 series O-rings (not thick enough...) and 2 series O-rings with a cross section of .139"...

BINGO.

I just happened to have a 203 O-ring in my arsenal. I dropped it into the bore of the cap and then inserted the needle valve. Perfect fit. No bushings, no machining. No custom parts. Hindsight is 20-20, right?

I pulled the gicleur back out, threaded the valve and bushing assembly back in place with the new O-ring behind the bolt head to seal up any internal leaks and gently snugged the whole shebang up. Then I carefully lined up the gicleur cap over the barrel of the needle valve and threaded the cap back in.
Water back on - No leaks.
Power back on - No leaks.
Pump on - No leaks.

I closed the valve and tested the group... No flow. I'm happy. A quick test of water debit at various valve positions so closely matches what Graham posted that there's no need to go over it in any detail. I set the flow rate to 45g in 30s and went to bed.

This morning, I pulled my very first Slayer style pre-brew shot. I didn't adjust the valve at all, so brew pressure should have been a steady decline after puck saturation, much like when Assaf left the FLB in place and posted telemetry here.

Here's a video of the results. I started the shot right before I hit record. Doh! Cheers!

- Jake