Posted on Leave a comment

Goldilocks part 3

The set of the bears. Plate 7, 1664, by Marcus de Bye, after Marcus Gheeraerts I, 1559. Gift of Bishop Monrad, 1869. Te Papa (1869-0001-67)

Though I’ve run out of three bears analogies, I’ve stuck with the story’s structure: first the porridge was too hot, then it was too cold, and finally Goldilocks found one that was just right. This iteration of the HX caused me to take my hat off, yet again, to the Italians. Many months ago, Dr. Pootoogoo brought a boiler from a later-model Brugnetti to my studio. The flange bolts were so badly rusted that it wasn’t ever going to go back into service without replacing them. It happened to be one with a diagonal HX that I hadn’t examined before and it inspired my to try a similar concept with the horseshoe HX prototype. I was quite surprised that the 60ml HX (baby bear) didn’t deliver water that was cooler than the brew reservoir temperature even though the HX volume was close to the 50ml shot volume. I also started thinking about what the ‘correct’ temperature for the brew reservoir should be. It occurred to me it might not be the best thing for it always to be the same. For example: if the group is at 75 C and the water coming in is at 102 C, the resultant water temperature at the puck is 92 C (these are roughly the numbers for both vintage machines) and we know that the group gains heat after a shot, let’s say for the sake of argument it gains 3 degrees and requires about 2 minutes per degree to recover i.e. 6 minutes. It follows then that for the next shot, if it is to be pulled [i]before the end of the recovery time,[/i] it would be preferable to have the brew reservoir water at a lower temperature than 102 degrees so that when it reaches the puck it will be at same magic 92 degrees. Because of the difference in thermal properties of the materials (i.e. the brass group and the water) and their relative volumes (i.e. big thermal mass of brass vs 50ml of water) it isn’t a one to one relationship. But it is linear – i.e. it will be a constant times the temperature rise of the group. So at any given time during group recovery, the required brew reservoir temperature is the reservoir idle temperature minus the group temperature rise times some constant. In math not English:

Tbr = Tbr_idle – K(Tgroup – Tgroup_idle)

In other words if the heat gain curve of the group could be inversely mirrored by the brew reservoir, then water will be at the right temperature when it reaches the puck [i]no matter when it is pulled[/i]. This is really just destructive wave interference:

If the brew reservoir temperature curve is positive (i.e. there is heat gain), then it will compound the problem of heat gain at the group. But if the brew reservoir temperature drops after a shot, then it will compensate.

The diagonal HX design consists of a large diameter pipe which connects directly to the back of the brew reservoir – essentially increasing the volume of the brew reservoir four-fold. In fact, the concept of the brew reservoir is pretty much gone altogether in this design – the group flange actually becomes one end of the heat-exchanger. A small diameter injector tube runs through the middle of the large diagonal pipe. Line water comes in through the injector too fast for the surrounding water to heat it to boiler temperature and mixes with the water behind the group to get the really stable results that we saw in the earlier testing.

I replicated the basic principal minus the diagonal tube and in so doing figured out why the diagonal design ended up that way i.e. diagonal.

The last kink in the 6mm tubing was only way to thread all the larger diameter fittings onto the injector. And this is a one-way operation: once it is brazed together you can’t take it apart again.

Brazed and (sort-of) cleaned.

And here are the results:

Blue – Boiler
Red – Brew reservoir
Purple – Group neck

Now, although the results aren’t perfect, it shows that the concept works. The group heat gain for successive (unnaturally) rapid shots has been significantly diminished and the recovery time for the group is less than half of what it was (less than 3 minutes). The length of the injector plays an important role in how much boiler temperature HX water mixes with the line water and consequently in the temperature of the water that reaches the brew reservoir. But, as I said, this is a one-shot fabrication and is too much trouble to alter. It would be much easier to change and/or maintain if the injector tube screwed into a straight length of larger diameter tube that maybe ran directly to the group right through the boiler, maybe on a diagonal…

Wait – someone already thought of that.

Enough porridge already.

Posted on Leave a comment

Goldilocks part 2

Illustration by Pervprude

It is high time for more porridge!

Several days after I did the first tests, it occurred to me that the horseshoe tube that I had made when I built the boiler was not the same as the one in the Brugnetti boiler. I wasn’t really thinking that hard about how a heat-exchanger actual works while I was making it. So I took apart the boiler (again). This is the original tube, nicely blackened after a few months of service:

I measured the interior volume of the ‘Papa Bear’ version and compared it with the volume of the vintage HX horseshoe (modeled in CAD): 115ml for the vintage and 135ml for Papa Bear. So I put Papa Bear to the bandsaw, and cut him down to make Mama Bear.

Blue – Boiler
Red – Brew reservoir
Purple – Group neck

(The data for the neck was very noisy for some reason that day – the rapid deviations do not reflect what is actually going on.)

Despite having the same volume as the vintage tube, ‘Mama Bear’ doesn’t behave the same way. The peaks in the brew reservoir temperature are similar to Papa Bear, i.e. around 5 degrees, and recovery times are about the same. It should be noted however, that with adequate recovery time, the neck temperature shows a high degree of stability.

I then modified the horseshoe again, lowering the volume to around 100ml (Auntie Bear??) – but the results didn’t change significantly. So I decided that drastic measures were called for: ‘Baby Bear’ – roughly 60ml volume which is about half of the vintage.

Though the results are better, there are still spikes in the brew reservoir temperature.

Time to alter the Goldilocks plot line.

Introducing ‘No Bear’! Partly to make sure that I wasn’t entirely out to lunch but also to measure the other extreme, I connected the supply to the brew reservoir directly to the mains, bypassing the HX and injecting room temperature water into the reservoir. Unsurprisingly the results are dramatic!

Very cold porridge indeed.

Interestingly however, while the neck temperature rises by a few degrees initially, it recovers quickly (less than a minute) and then drops to 2.5 degrees below idle.

Where does this leave us? The concept of the HX is of course to inject (fresh i.e. non-boiler) cold water through the hot water in the boiler in order to raise it to the ‘correct’ temperature to feed the reservoir. If the machine has been idle for any length of time, the water in the HX will be at the same temperature as the boiler. Subsequent shots will draw cold water into the HX, which, depending on how it is designed, will consistently bring room temperature water up to a specific temperature (either boiler temperature or slightly lower), as long as the heating element can keep up with the demand. So some not very earth-shattering conclusions:

  • Changing the design of the HX will determine the temperature ‘profile’ of the water delivered to the brew reservoir.
  • Somewhere between 60ml and 0ml of HX volume, the water delivered to the reservoir will offset the heat gain and result in equilibrium.

There is one additional question that results from the three bears test: why do identical horseshoes in the vintage and new boilers not exhibit the same thermodynamic behavior? My hypotheses is that materials used for the brew reservoir and boiler are playing a much bigger role than I first thought. Both of these parts on the prototype are made from stainless which is roughly 20 times less thermally conductive than copper and bronze. To test my theory, I put the prototype group onto the vintage machine: compared to the all-stainless boiler assembly, the new group runs around 12 degrees hotter on the copper/bronze boiler. The stainless brew reservoir is slow to acquire heat from the boiler and the water in the reservoir and reluctant to relinquish it to the air or pass it on to the group.

SO…. two rather more significant conclusions:
1 – There should be (or rather, spoiler alert, there is, as will be seen in an upcoming episode) an HX design that meets the requirements of stainless boiler and group combination.
2 – It is time to make a new boiler – using the bronze brew reservoirs that I received a few months ago.

Posted on Leave a comment

Goldilocks part 1

Illustration: Arthur Rackham

[Ed: Re-post from May 12 2018]

I imagine everyone remembers the story of Goldilocks and the Three Bears?

I love the painterly references on the wall – a Degas ballerina on the left, an old master on the right, an enameled reproduction of a Greek vase in blue? But I digress. Apart from carrying a wee bit of Freudian baggage, it is also perhaps not the most pertinent of fairy tales for our times (how would you react if you came home and found someone sleeping in your kid’s bed?). Anyways, what I wanted from it for today’s post is the porridge. You know, the three kinds of porridge that are too hot, too cold and just right? You see where I’m going with this… Except that in my story there are four bears, or rather three bears and no bear, but neither of those versions scan very well, so we are going to stick with the original title. First up is Papa bear’s porridge.

This is one of the first temperature profiles that I took of the prototype machine. The methodology is very similar to the tests on the Brugnetti [Auroras] that I posted previously. As always, the Scace-values are not terribly useful except for comparison.
A reminder of the colors:
Blue – boiler wall
Red – brew reservoir
Purple – group neck
Green – PF receiver on the group
Yellow(ish) – Scace puck


  • The boiler temperature curve is sinusoidal as opposed to saw-tooth and has a delta of 0.7 C between minima and maxima. This performance was obtained using only the P term of the [PID] algorithm. It could probably be improved with some additional tuning, but I haven’t bothered because it is already pretty good.
  • Brew reservoir temperature is stable at idle and shows no trace of the boiler temperature variation.
  • Recovery times for a shot are about 2 minutes per degree for the group and 1 minute per degree for the brew reservoir.
  • Group temperature at idle (~70 C) is quite a bit lower at idle than the Auroras (~80 C and ~83 C for the horseshoe and diagonal versions respectively).
  • Don’t touch the probes.


  • Job (pretty much) done on the PID boiler control.
  • Plenty of room for improvement on the brew reservoir temperature stability which is gaining too much heat after a pull.
  • The whole shebang has to be hotter.

Next up: Mama bear and No bear.

Posted on Leave a comment

temperature profile diagonal HX

1987 Porsche 911 Turbo Cabriolet

[ED: Re-post from May 4 2018]

This is a longer test (over an hour) on a 1987 diagonal HX machine.


The same methods as were used for the horseshoe HX machine were used with a few minor differences.

  • This machine is on 24/7 so there is no warm-up period.
  • It had been at idle for at least a couple of hours.
  • The same pseudo-Scace device and needle-valve was used to simulate shots.
  • The period between simulated shots is generally longer and I did not repeat the faster than commercial usage of the previous test, rather, I waited for the group to recover between shots.
  • At minute 3 and 32 there are cleaning flushes – i.e. pulls that were significantly larger in volume than an actual shot.
  • At minute 23 a PF filled with real coffee was locked into the group and a shot pulled.
  • At minute 25, the steam wand was used to foam milk for a cappuccino.
  • The additional orange line that starts at minute 46 is another K-type thermocouple that was placed on top of a puck of coffee before the PF was locked into the group.


  • This machine has the same pressurestat as the horseshoe HX machine from the previous test and consequently we see the same saw-tooth wave for the boiler temperature.
  • With the exception of the cleaning flushes, the brew reservoir is [b]remarkably[/b] stable – exhibiting almost no discernable trace of the fluctuation in boiler temperature. The temperature drop for the outside of the reservoir after a shot is pulled is less than one degree and the recovery time is between two and three minutes.
  • The group as a whole exhibits the same tendency as the horseshoe HX machine to gain heat with each shot.
  • The recovery time for the group is a little over a minute per degree C of heat gain i.e. essentially the same as the horseshoe machine. As the groups are identical this is unsurprising.
  • The shot test with real coffee plus the additional thermocouple starting at minute 46 shows a 90 C peak shot temperature and is quite likely to be accurate.
  • At minute 40 I removed the PF to prepare for the next shot and the preparation time is longer than usual because I was fussing with the thermocouple. Removing the PF has a fairly significant effect on overall group temperature as the neck falls to the same 83 C that it was when the machine was at idle.


  • The diagonal HX is about as good as it gets as far as temperature stability of the brew water reservoir goes. The reservoir recovers from what little variation there is in less than half the time that the group takes to recover between shots.
  • The slower, more realistic, pace of shot pulls in this test is illuminating. The actual temperature gain seen at the group is between 4 and 5 C per shot. This translates to a group recovery time of five to six minutes – possibly a little slow for a commercial setting, but definitely fast enough for home use and entirely manageable with a cooling strategy such as a cooling flush of a known volume.
  • 1987 was better than 1982.
Posted on Leave a comment

temperature profile horseshoe HX

[Ed: Re-post from April 29 2018]

So, if a pressure profile is a cat among pigeons around here, here’s a fox in the hen house:

What are we looking at? This is a multi-channel temperature log of my early 80s [Brugnetti Aurora] horseshoe HX machine over a period of about half an hour.


Five K-type thermocouples were installed on the boiler, the side of the brew reservoir, the group neck (i.e. extension between the bolt flange and the body of the group), the outside of the portafilter holder on the group and the puck inside the pseudo-Scace. The machine was turned on at roughly 10am and was at idle for at least one hour before the test. The pressurestat is set so that the average boiler pressure is in middle of the recommended band (the green area on the Brugnetti gauge) at 0.9 bar.
At 5:30pm the portafilter was inserted cold into the holder and 6 simulated shots were pulled with the needle valve set to give 20-30 seconds of flow per shot. The spacing of the shots is very quick – about 30 seconds apart – much faster than a ‘worst case’ scenario of constant use in a commercial setting.
At 5:34pm the machine was left to idle for 3 minutes then a further 5 shots were pulled.
At 5:40pm the machine idled for 5 minutes and 3 shots were pulled.
At 5:59pm the portafilter was removed.

Limitations of the methodology:

A) The pseudo-Scace is by no means a perfect analog to actual coffee for a number of reasons:
First, the Acetal puck, although vaguely similar, does not conduct heat in the same way as coffee grounds.
Second, the puck gains heat from shot to shot unlike coffee, which will always be at (close to) room temperature.
Thirdly, the flow rate through the needle valve is not very repeatable.
Fourthly, fluid flow through real coffee is unlikely to be linear, and this non-linearity is not modeled with the needle valve.
So temperature readings “at the puck” should be taken with a grain of salt.
That being said, for comparative study of different machines, Scace-like instruments are a valuable tool.

B) Also, with the except of the pseudo-Scace, these measurements are surface temperatures not actual water temperature.


  • The roughly triangular wave of the boiler (blue line) is from the pressurestat controller with roughly 0.2 bar of hysteresis (dead band). Total variation is approximately 4 degrees C.
  • The (red) brew reservoir temperature (and by inference also the water inside it) correlates closely with the boiler temperature but the variation is considerable damped. Total brew reservoir variation is less than 1.5 C.
  • Average (or baseline) temperature of the brew reservoir is affected only very slightly by continual use rising from 100.5 C at idle to 102.5 C after intense activity.
  • Neck temperature (purple) seems to be a better analog to puck temperature than the outside of the portafilter holder.
  • The group gains heat with each shot if not left enough time to recover and (based on a small sample of only two points on the group) the heat gain is uniform throughout the group.
  • With the exception of the walk-up shot with a cold PF, the temperature gain of the group is between 1.5 C and 2.5 C per shot.
  • Group recovery time – that is, the time it takes for the group to cool from any given temperature gain – is about 67 seconds per degree C of temperature gain.
  • Puck temperature (bearing in mind the caveats above) is lower than expected, especially for the walk-up shot – but the results in the cup are known to be good with this machine at these settings.


  • Brew reservoir temperature demonstrates a relatively high degree of stability even with a dead band of more than 20% of the operating pressure setpoint.
  • Group temperature should be stable if correctly managed. If the machine is left to idle between shots for a reasonable amount of time (between 2 and 3 minutes – i.e. not inconsistent with the time it takes to prepare a shot), it will return to its initial temperature and provide repeatable shot temperatures.
  • The methodology should be changed for future tests to reflect a more realistic real-world usage with a sufficiently long period of time between shots.
  • Not bad for 1982.
Posted on Leave a comment

pressure profiling 1

[Ed: Re-post from April 28 2018]

I did pressure profiling partly as an idiot check, but mostly out of curiosity. As I am using the same springs in the old machines and the new group the performance should be the same. The only thing that really needs to be verified is initial/peak shot pressure which depends on the length of the compressed spring with the lever in the down position (and any variation between the actual springs themselves). The method used to do the profiling was laborious. I made my own Scace-type device with an Acetal disk to fill in for the puck, an analog pressure gauge, a K-type thermocouple and, something not featured in the Scace, a needle valve to control the flow.

Then I filmed some simulated shots and logged the pressure readings on the video every five or ten frames. Boooooooooooring. There are many better ways to do this, but, as I said before, I don’t really need to check the variation over time, just the steady states. I tested three machines: the 1987 diagonal HX, the slightly earlier horseshoe HX and the new prototype group. However, the results between the two antique machines are essentially the same, so I only bothered logging one of them.


It is hard to adjust the needle valve to get repeatable flow rates and thus repeatable simulated shot times, but you can see that the spring performance is essentially a straight line from peak to about 6 bar. At 6 bar, the lever reaches the end of its travel (i.e. the spring is completely extended) and the rest of the curve is just residual pressure.

When the lever is lowered, the line plateaus at the pre-infusion pressure: just over 2 bar for the antique machine which has a pressure regulated water supply and a little over 3 bar for the prototype which was directly connected to the city supply.

Peak pressure for the antique groups was just above 12 bar.
Peak pressure for the prototype is about 11 bar – this was adjusted to 12 bar by shimming the spring by about 2mm.

Posted on 1 Comment

The rite of spring


So Stravinsky’s ballet caused a riot… perhaps it was the shocking juxtaposition of pagan and modern Weltanschauungen, maybe the audience just didn’t like the music. What can I say? Springs are contentious. With that warning, I shall begin trying to unravel the (minor) mysteries of one particular spring.

A little more than a year ago, it was brought to my attention that replacement springs for the Brugnetti Aurora lever group were no longer available. I checked a number of suppliers and looked for substitutes without any success and I immediately went to my local parts dealer and bought the rest of their stock – a shockingly large number: four.

This is a problem if you are me and want to make new groups, or, if you are not me, repair old ones. So I took the dive into spring design and started to think a bit more carefully about a part that I had presumed was going to be “off the shelf”.

I have since come to the conclusion, for a number of reasons that will further elucidated, that the spring that was sold to me as a replacement for part number A.29 for the Aurora group, may not be, well, a replacement for part number A.29.

When I changed the spring on my first machine (well before I embarked on this voyage deep into the jungle down the Congo River) I noticed that the new one was a little taller than the original, making it quite a bit more difficult to install. Installed it was however, and I thought no more about it. That particular machine dates from 1987 and was rebuilt by a local dealer in the mid 90s – but may well have kept its original spring given how much of a hassle they are to change.

Under normal operation, springs will deliver an amount of force that is directly proportional to the distance they are compressed: F = kx (where F is Force, x displacement and k is the spring constant). Put another way the spring constant is simply how much force the spring imparts per unit of compression. So determining the spring constant can be done by measuring force versus displacement. This is the setup used – the force gauge (i.e. bathroom scale) isn’t ideal because of the built-in “intelligence” which automatically tares (zeros out) small readings and shuts off the display, but it does measure up to 175Kg. Displacement is measured using a digital height gauge (not shown) that (as so long as the same datums are used and the gauge isn’t re-zeroed) provides more accuracy than required with enough precision (I can never keep those straight).

I believe that I have identified three different springs that were installed in the Aurora group at various stages:

The replacement “after-market” spring from the local dealer which has an uncompressed length of 133mm.
Antique spring #1 – which has an uncompressed length of 128mm.
Antique spring #2 – which has an uncompressed length of 116mm.

Here are the compression versus force profiles of those three plus a fourth new prototype spring.

The two trials of the after-market 133mm spring suffer from some measurement error – i.e. if you extend the lines back towards the X axis, they don’t intersect the origin coordinates; which they should, as zero spring displacement results in zero force. If they were to be normalized (i.e. shifted down until they would intersect the origin if extended) you can see that they correlate closely with the all of the other trials except the Antique 128mm. In fact the spring constants (calculated using the average slope of all the data points in each trial) for all but the 128mm spring are around 60 N/mm whereas the 128mm is significantly lower at 44 N/mm.

However, the fact that the springs are different lengths is not a minor detail. As the geometry of the piston assembly remains the same, the springs are all compressed to the same size when they are in use i.e. they have an installed length of 96.25mm (corresponding to the lever in the up position) and a fully compressed length (when the lever is in the down position) of 75.5mm.

Installing a springs of different lengths will mean they are operating over different force ranges. The corresponding pressures that the piston will deliver can be easily obtained from the equation P=F/A (where P is pressure, F is force and A is the surface area [19.63cm sq for the 50mm diameter piston in the Aurora group]).

Spring length(mm)Spring constant (N/mm)Force @ fully compressed (N)Pressure (bar)Force @ installed length (N)Pressure (bar)
antique #112844.39233011.914097.2
antique #211659.21239812.211696.0

Presuming that you find tables at all interesting, some interesting points can be drawn from the one above (although now we are getting into subjective and therefore contentious territory). The first line shows that the replacement after-market springs from the local dealer are very likely incorrect as their theoretical operating range is 18-12 bar. I think most people would agree that this is too high. The case for these springs being incorrect is strengthened by the fact that a design analysis of that spring configuration, (i.e. the spring constant, wire size, number of turns, end conditions etc) results in a non-compliant design when used in this application (i.e. the installed length and travel) meaning that it is likely to fail to perform as expected or simply to fail over time.

The second point of interest is that the two antique springs, despite their different properties, yield very similar pressures in the installed configuration – approximately 12bar maximum and 6-7bar minimum. Without knowing more about the provenance of these particular parts, it is hard to know whether they have changed over time or whether they are still operating as designed. However, based on the subjective results of the quality of the coffee that the machine produces when it is operated over this pressure range, I believe that this was the design intent. Further testing of other old springs of known provenance would be helpful to confirm this hypothesis.

For the prototype I chose to use the shorter ~116mm format because it is significantly easier to install and opted for a similar 12-6bar range at the installed configuration. The new spring should be a drop-in replacement for the old Aurora groups.

Prototype Lapera spring on the left, 18bar monster on the right. We fervently hope this little one doesn’t dance herself to death.

Pina Bausch – The Rite of Spring

Posted on Leave a comment

That way madness lies…

Sir Ian McKellen as King Lear. Image source: The Guardian – Photograph: Johan Persson

No, I will weep no more. In such a night
To shut me out? Pour on; I will endure.
In such a night as this? O Regan, Goneril!
Your old kind father, whose frank heart gave all—
O, that way madness lies; let me shun that;
No more of that.

King Lear, Act 3, scene 4, 17–22

The Brugnetti Aurora is my favorite single group vintage espresso machine: minimalist design, bold colors and a fantastic logo in a cousin of the 1962 Eurostyle font. The one in the picture, an HX version, was restored by Orphan Espresso. It was likely assembled and sold in the US by Termozona, an importer from Connecticut who added, what is in my opinion, the criminally design-inappropriate Europa badge on the front.  

Everyone who I have spoken to agrees that they are great machines: simple, well built, small enough for home use, and above all else, capable of producing an incredible shot of espresso. They are, however, rare. The original Sr. Brugnetti closed down and liquidated his factory before the brand was relaunched under new management after a five year hiatus. Brugnetti no longer makes lever machines and I presume that the original group head molds, manufactured by them or a supplier (Tortorelli Meccanica in Sienna?), either no longer exist or would be impossible to track down with my hundred words of Italian.

The heart of the matter is the lever group and the heart of the group is the group body which is a complicated part. So the million dollar question is what would it take to reproduce it?


A discussion of building a lever machine from scratch with reference to the Aurora. thread discussing the origins, shrouded in myth, of the machine. Another couple of restorations of the European Aurora from kaffee-netz in Germany are here and here.

A restoration of a two group model – surely not an original case?