Hi:

I just finished building and installing a bolt-on, programmable pressure profiling pump system to my uber-stable LM Frankenlinea. I did this because I wanted to test prevailing ideas about preinfusion, the benefits of lever machine pressure profiles, etc. on a very stable platform on which selected brewing parameters may be quickly changed while holding others constant to values which are the current state of the art. At the same time, if the tests warranted, I wanted to make the system transparent enough that I could put together similar systems so that interested folks could just do a pump swap, with no extra effort involved. Depending on how this all goes over the next few weeks I may end up writing something more comprehensive, but for now I'll give progress reports as I learn stuff.

Acknowledgements up front for folks who who ought to be recognized: Andy Schecter implemented brew pressure profiling over a year ago, and should be credited as being a true pioneer on this. I'd like to thank Michael Teahan for tipping me off to a source of the required pump, and to Clyde Smith at Fluid-o-Tech, for his advice and encouragement. I should also mention here that the path I'm treading is not new. There is a body of work that has recently been performed by machine manufacturers, but I am not privy to details.

As I said earlier, the idea here isn't necessarily to pioneer any device, but to test the idea of pressure profiling in a reasonably systematic way, look at any perceived benefit from pressure profiling, publish the results in places where the results might have some impact, and in the end maybe make it possible for others to tailor brew parameter space for themselves if the results warrant.

Now, a little about the device: Unlike Andy's, the programmable pressure-profiling system (PPPS) is built from an array of commercially available parts. The pump is a rotary vane pump, which is magnetically coupled to 3-phase magnetic drive. Power is supplied to the pump drive by an accompanying 3-phase, variable-frequency converter that operates the drive over the range from 1100 to 3500RPM. The RPM of the pump is varied in response to an output signal provided by a fast-acting, industrial process controller. An electronic pressure transducer mounted in the water delivery line to the espresso machine provides a feedback signal to the process controller. The pump / drive combination is really quite interesting. Power consumption of the magnetic drive is 1/3 the power required by a traditional AC motor / pump combination. The system also has reliability advantages. Since the pump is magnetically coupled to the drive, there is no shaft seal to wear out. The pump plumbing is configured so that most of the water delivered by the pump impeller is recirculated to the pump inlet. The amount of recirculation is adjusted so that the pump runs at nearly maximum speed when the output pressure is approximately 140 psi. This sets the pressure range of the pump from almost no pressure at 1100 RPM to 140 psi at slightly less than 3500 RPM.

The test machinery: The test platform includes equipment considered to be state-of-the-art within the coffee industry. The pump is installed on a 2-group, LM FrankenLinea AV. Water supply to the pump is from a 5-gal. plastic carboy, filled with carbon-filtered water conditioned by an acid neutralizer and solar salt ion-exchange softener. Unlike a stock Linea, the Linea used here has no exposed plumbing. Flowmeters and group solenoids are mounted on top of the group caps, with water passageways integrated into the cap design. Temperature stability and reproducibility of this machine is better than 1 degree F under all duty cycles. Temperature control is achieved using a boiler mounted thermocouple, with industrial fuzzy-logic process controller. The machine has 0.6mm gicleurs installed in both groups. A Mazzer Robur is used for grinding the coffee. Coffee is roasted in-house (literally, since that's where Espresso Research Company HQ resides) on a Has Garanti drum roaster with 1 kg. capacity.

Progress: I've got the pump installed on the Linea and I've learned how to control the pump. The coupled system of pump and espresso machine produces an interesting control problem on startup. Cool makeup water entering the boiler expands, raising the boiler pressure to 12 bar. On pump startup, the brew solenoid opens, ejecting a small amount of water. At the same time the pump activates, with speed controlled by the process controller. The controller at first detects falling pressure, and speeds the pump to compensate, resulting in a pressure burp on startup. The problem is reasonably managed by selecting the fuzzy logic control parameter in the controller, and assigning an initial pressure setpoint of around 1.5 bar, rather than my initial setting of zero. The controller now detects a falling pressure, but the response is to allow the pressure to fall, ramping quickly to 1.5 bar for 1 second, then initiating the programmed pre-infusion cycle. There is no issue of pressure burping if the pump is configured to mimic a normal AC-motor-driven pump. In this case, at startup the pump speed rapidly increases to that which is required to support 9 bars pressure as measured in the group.

I have loaded a program into the controller that mimics an e-61 pre-infusion profile on startup, with straight-line (linear) increase in pressure over an 8-second period to 9 bars of pressure. The pressure profile remains constant at 9 bars for 5 seconds, then reduces linearly to approximately 7 bars at pump cutoff. The pre-infusion result is visually interesting to me in that liquid appears at the same time over the entire surface of the brew basket (bottomless portafilter), indicating that saturation of the cake is occurring more or less completely. It's intuitively satisfying, at the least in that the subsequent liquid flow buildup is uniform and very similar to what I have observed from e-61 machines. The effect of the pre-infusion ramp on required grind fineness is that my Robur needed to be set to produce finer grind by several divisions of the knurled outer surface of the adjusting collar. This is similar to what I experienced when reducing gicleur diameters in the Linea during comparison tests against a pre-production LM GS3 last winter ('05 - 06). I attribute this to reduced "slamming" of hot water into the cake, reduction of fines migration, or whatnot. Anyway it seemed reasonable to me based on past experience.

The pressure tail-off is an arbitrary exercise just because I could do it, and because Andy Schecter reports some benefit. It appears possible to control the flow rate of the extraction by controlling pressure tail-off, and by doing so somewhat delay the onset of blonding. I don't know much more than that at this time. Paul Pratt kindly sent me the pressure profile of his Faema lever machine, which I will try when I get a little more orderly in my methods. Essentially his machine starts at near zero pressure, builds very quickly to 8.5 bars (piston release), and then reduces to 3 bars at the end of the extraction. I'm presuming the pressure reduction is linear since the piston is spring-driven.

Taste - Obviously this is the important, bottom-line issue. Qualitative observation of extraction is only meaningful if it can be correctly correlated to optimum taste. It's REALLY early to draw any meaningful conclusions. But I didn't screw up by installing the pump system. The coffee is at least as good as I was producing before, and I was doing a pretty good job prior to the pump install. I've not yet had time to alternate back and forth between straight 9 bar extractions and profiled extractions, because my first efforts have been in sorting out the system. I hope to do testing soon that will be reasonably valid. I may bring another conical grinder home to pair with the Robur so that I can one grinder correctly for profiled shots and one for constant pressure shots. And I've got to find some test victims - maybe local folk that can really hold their coffee.

More soon!

-Greg

This is cross-posted at alt.coffee and Coffeed. Both sites are not necessarily visited by HB folk, and both sites have different target audiences.

Hey Greg,

Very interesting!

Looking forward to hearing more.

Are you doing all these tests at constant temperature (that is at one temperature) or are you changing the temp for various test/coffees?
Perhaps too early to be worrying about that.

And it looks like you are running into the confounding thing about coffee (as you mentioned). It's the taste that matters, and that's so freakin' frustrating from an engineer's point of view. Of course, that's the beauty in it as well...

The problem is that tests like the one you are embarking on approach poisonous when you are required to taste everything (to say nothing of tastebud fatigue).

Anyone know the LD50 for espresso?


Kudo's again on the test rig, sounds really cool.


ciao

lino

Things have come a long way since I attached a variac to a vibe pump and watched the gauge four years ago.

I'm not sure how far Andy ever got with his experiments; I'd still be interested in seeing how a "constant flow" profile looks, i.e. a pressure profile that created a constant flow pour. Andy verbally communicated that this may require very high pressure early (and may not be a good idea for the first 10 seconds or so), and a declining pressure at the end.

The amount the pressure needed to be tapered depended hugely on where it was measured. The puck is a declining resistance, especially as the fines start degrading 20 seconds into the shot, while the gicleurs are fixed. Measuring points up stream of the group show far less of a declining profile to achieve even flow than the pressure measured directly above the puck.

Andy said that he hadn't really investigated it that systematically. My thinking is similar to yours in that I immediately hit on the idea of constant flow rate shots once the cake has been saturated. Also, the controller allows 16 ramp / soak cycles in one program, which means you can parse up the extraction into small enough steps that the presure profile in effect becomes a continuous curve. The best pressure profile may indeed use a maximum pressure well in excess of 9 bars, if only for a couple of seconds. I'm not discounting anything at this point.

If you didn't mind, could we consult on how one should construct test sequences for this work? I have some willing victims, and I can pull shots and deliver them without the subject knowing the pressure profile used, I think. Your statistics knowledge is waaaaay better than mine.



-Greg

gscace wrote:If you didn't mind, could we consult on how one should construct test sequences for this work? I have some willing victims, and I can pull shots and deliver them without the subject knowing the pressure profile used, I think.

Ken and I will be doing an article discussing our experiences, recommendations, and disagreements about "home coffee research." Ken has a medical research background and favors well planned, well designed experiments which can be easily and solidly analyzed, since the data collected is balanced and pure. I have a qualitative and historical social science background, so it's a "just do it" ask questions and read archives type thing, revise the theory as you go, then slam whatever data you have with fancy stats to overcome its deficiencies and torture out some answers.

After our initial discussions we realized that the best approach is to do both. Start out informally. The idea is not to prove anything but to come up with good theories:
-- what exactly does varying the pressure do? My guess is the key variables here are three: 1. rate of water flow through the puck, 2. the rate of solubles absorption (puck weight loss), 3. the water retention and the overall --is it sandy or pasty feeling? -- texture of the puck, again as a .dt You may want to stop shots at various points mid-extraction and do some puck analysis and weighing to get a feel for this.
-- What is the taste and reliability effect of different profiles. I think if you know what is happening to the puck during the extraction, you'll start having very specific ideas about what to taste for. For instance: Is a certain pressure profile far more consistent tasting than others -- this would be absolutely huge in terms of making such profiles mandatory in cafes. Does a certain pressure profile allow for deeper and sweeter extractions of high dosed shots -- my paper on getting 20% plus percent extractions requires lower doses -- a lot of people feel this is more cost than benefit. So profiles that get sweeter, higher extractions at US dosing levels would be huge too.

The key to this phase is to revise your hypotheses at the drop of a hat as you gain experience -- you want to shoot down ideas as fast as possible, so you get to the meat of it quick. Never make a big production of anything, just go for a very hard but quick test of anything you think is happening. If it fails, move on; if it passes, the idea gets into the formal test round.

It's easy to have ultra-valid experiments with perfect and pure data that will never prove anything. All you need is a dumb hypothesis like "Profile A tastes better then profile B" Whatever the outcome, nothing has been learnt, since the hypothesis is meaningless.

This is fine when you are very sure there's nothing at all depending on the difference between A and B, you've tested it out, and found nothing systemtatic during the informal phase. Then the formal test of a hypothesis like this just puts the stake into the heart of a bad idea.

But my guess is you went through all this trouble because you think there's gold in them pressure hills. The task now is not to prove anything, but to find the gold -- work with the profiling system, find out how different profiles affect the extraction, and how that in turn affects the taste. When you've gotten solid and specific ideas, it'll be time to take it to the assayer's office and develop a test that really proves it.

I've become somewhat more sceptical that I used to be on how much gold is in them pressure hills. But I'd be delighted to help on either or both the search or assay phases.

another_jim wrote:
Ken and I will be doing an article discussing our experiences, recommendations, and disagreements about "home coffee research." Ken has a medical research background and favors well planned, well designed experiments which can be easily and solidly analyzed, since the data collected is balanced and pure. I have a qualitative and historical social science background, so it's a "just do it" ask questions and read archives type thing, revise the theory as you go, then slam whatever data you have with fancy stats to overcome its deficiencies and torture out some answers.

After our initial discussions we realized that the best approach is to do both.

It's a lot of hard work to actually pull off an experiment varying factors in espresso production in a way that you can actually "prove" anything. I have recently completed a trial comparing previously frozen coffee vs. never frozen, in a complex design with multiple variables controlled. Even if you know what you want to test, it's damn difficult to set up such a trial and even harder to actually execute it, that is if you want to produce results that will really "mean" anything. Stay tuned to these pages for the freezing article which is currently being written.

It sounds like you are nowhere near ready to design a study that has much chance of reaching any level of statistical significance. How could you; the modifications you are working with are so new that you haven't actually figured out how best to use them.

I'm with Jim on this. Use your "willing victims" to help you to narrow in on what it is that you think works, and after you have a really good idea of what that is, design a simple comparison between a standard process and a modification that you think adds something, while holding other factors constant.

It would be great if you had another identical system that you could use for comparison shots in real time. Since few will have this sort of equipment lying around, other approaches such as more complicated grading of shots (rather than comparisons to another shot sitting in front of the taster in real time) would presumably be needed. In any event, I think the taster needs to be blinded to what is going on, or the judgment is going to be clouded by what the taster expects to be tasting.

ken

Sounds like a great initiative on your part - perhaps a little controls nightmare at first but an engineered package could be very enticing to both cafes and individuals.

I would very much like to be included on your "locals" list.

Eric S.
erics@erols.com
301-587-5033

Here's an update on the last few days of progress with brew pressure profiling. First, I have a greatly enhanced tolerance for caffeine. Besides that...

I've fooled around with profiles a bit. What I've learned is that a person can drive himself half crazy if he has a very tuneable system and no practical knowledge on which to draw. If he's half crazy to begin with, the combination produces someone who is full-crazy. This can be bad. I've also formed the opinion that there is something to soaking the coffee cake at a low pressure and then ramping the pressure up. Right now the pre-infusion cycle is set to 2 bar for 3 seconds, with a subsequent ramp to from 2 bar to 9 bar over a 4 second interval. Liquid coffee appears evenly over the perforations on the bottom of a brew basket just as the pressure begins its rapid rise. I like this preinfusion cycle over a straight line rampup in pressure for 7 seconds from startup to 9 bar because it produced more crema in the limited pairs of shots I brewed alternating between straight line ramp, and low pressure wetting with faster ramp.

Those who have observed many shots worth of extractions squeezed through a bottomless portafilter are familiar with the cone of coffee that hangs off the brew basket during the "meat" of the extraction. The liquid appears quite viscous for a time, hanging off the bottom in a hyperbolic shape, eventually thinning out and becoming lighter in color at the end of the extraction. When this happens the hyberbolic-shaped cone seems to 'withdraw" toward the brew basket as the fluid gets runnier. It seemed reasonable to try to control the volume flow rate of the extraction toward the end of the brew cycle in an effort to make the liquid flow more orderly, and maybe more viscous as well. So I programmed various downslopes in pressure at the end of the brew cycle, arriving at a profile that includes the previously described pre-infusion cycle, a constant pressure period of 7 seconds at 9 bar, then a downslope for 20 seconds to 6.5 bar. The flow rate of liquid in the downslope is reduced when compared to the constant pressure case, and the color is darker at the end of the extraction. Extractions seem to take a little longer, using the visual clues that I'm used to using as a guide for when to stop brewing.

One thing that I've learned for sure is that there is a fundamental difference between what I'm doing with the profiling pump, and how pressure profiling works in a lever machine. In a lever machine, the spring-loaded piston drives water through the cake. The pressure of the water is determined volumetrically. By this I mean that the pressure is dependent on the position of the piston, which means that pressure is directly related to how much water has been forced through the coffee. In the system that I am investigating, the pressure is related to elapsed time. One could grind very coarsely and push a whole lot of water through the cake in 7 seconds, yet at 7 seconds elapsed time the pressure will have just peaked at 9 bars. On the other hand, if the coffee is extremely coarse in the lever machine, then at 7 seconds elapsed time, the pressure will be much reduced compared to a proper extraction. And the amount of water corresponding to that specific pressure value (and forced through the coffee) will be identical to the case in which the coffee grinds are correctly sized. In the system that I'm using, it appears that the time-based profile has to get matched to the grinds size so that the pressures vary in time appropriately within the extraction (I dunno if I know what "vary appropriately" completely means yet, but I maybe know a little more than I did last week). I'm finding out that not only am I matching grinder settings to collected volume, viscosity, color and crema, I'm matching the grinder settings to the pressure profile as well. This adds another parameter to the mix, but it's not unmanageable and I'm getting more comfortable with it.

I've also learned that there are pressure issues at play in all pump-driven espresso machines that affect orderly pressure ramping at the start of brewing and may or may not have a small effect on coffee taste. Relatively cold water is pumped to the brewing water reservoir (boiler or heat exchanger) in espresso machines without feedwater pre-heat. The water is heated within the reservoir, and the pressure within the reservoir increases due to thermal expansion of the liquid water. When the group solenoid is activated, the water pressure is instantly released and a small spurt of water flows onto the coffee. I've reduced the effect of this to a large degree, but I haven't eliminated it entirely. I think that the way to almost completely eliminate it might be to install a normally open solenoid-actuated valve inline with the pressure-relief line from the brew boiler to the pressure-relief valve in the drain box. I would connect the valve to the pump circuit, so that the valve would close when the pump is activated. Rather than adjust the pressure-relief valve to the recommended 12 bar, I would adjust it to crack open at 2 bars. When adjusted in this way, the boiler would maintain only 2 bars pressure when idling. The solenoid valve would close on pump actuation, allowing pressure to build to brewing pressure levels. At the end of the brewing cycle (pump de-activation), the solenoid valve would re-open, again limiting boiler pressure to a low level.


WRT taste - I've been mostly investigating a lot of the physical effects of variable pressure on the extraction, and learning about how to do it. I've been doing this because varying pressure doesn't make the coffee suck. It tastes good enough to warrant spending the time learning about the system, and I am enjoying the taste and the learning process. Yesterday I pulled a couple of shots for a third party both with the profile described above, and without any profile at all. There was a difference in taste. Next weekend I hope to do some blind testing in which I brew shots for folks with both profiled and non-profiled pressure, seeking to learn if the difference can be reliably detected and which shots are preferred.

gscace wrote:One thing that I've learned for sure is that there is a fundamental difference between what I'm doing with the profiling pump, and how pressure profiling works in a lever machine. In a lever machine, the spring-loaded piston drives water through the cake. The pressure of the water is determined volumetrically. By this I mean that the pressure is dependent on the position of the piston, which means that pressure is directly related to how much water has been forced through the coffee. In the system that I am investigating, the pressure is related to elapsed time.

Nothing a lot more time and money couldn't address .... just bring the flowmeter into the control loop so that you can vary the pressure profile based on volume passed through the system instead of time. If some of this work plays out to be meaningful, that might be an interesting thing to try.

I am just now realizing how important your grind setting (and repeatabilty) are to make any of this testing work. I wonder how consistent the LM Swift grinder is and if one would prove to be helpful in your testing?

BTW - Are you having nightmares yet?

I'd thought of precisely what you suggest - that you could make the system volume-based rather than time-based. I'm also thinking that I probably won't do so. Locking the system into flow is prolly less flexible, and flexibility is what I need at this point. The grinder is up to the job and my technique is good enough that the additional requirement of time-matching the extraction through grinds size isn't being a problem. That being said, I'm not sure how good my dosing stacks up with a Swift. There are other issues with the Swift (non-conical burrset being one) that prevent one from taking up counter space at Espresso Research. A Robur is gonna have to be good enuff.

-Greg