In Pursuit of a Gravity Escapement.

[Phil Burman]

I started with a half scale copy of Brian Laws' #20 wooden clock to prove it would work at half scale in metal.

proof of concept video.

The next step was a design concept of a prototype escapement based on Brian Laws' #20 wooden clock but with significant changes. With a 250mm pendulum and a 60 pin escape wheel it makes one revolution in 60 seconds and provides a 1 second tick. In the video the escapement is driven by a 1000mm pendulum.

YouTube video here:

60 pin gravity escapement

The design concept of this prototype escapement is based on Brian Laws' #20 wooden clock. With a 250mm pendulum the 60 pin escape wheel makes one revolution ...

youtube.com

 

[tok-tokkie]

I was unaware of that variation of a gravity escape. Your proof of concept has 16 teeth & pins (by my count). Only a single pallet so 1 escape per full pendulum cycle (back & forth). So, for 1 sec pendulum it would need 60 teeth - as you state.
I made a 30 tooth gravity escape which I thought was novel but it is only a minor variation on the original "four legged escapement" developed by Grimthorpe for household clocks. This is a very significant departure from the layout of that design yet preserving the essential virtue of it always being just the weight of the pallet that impulses the pendulum.
I will watch this thread with great interest.
EDIT. I only looked at the linked YouTube after writing the above. That it uses a simple pin wheel for both the lifting & locking of the escape is a great simplification. Beautifully quiet & no need for the "fly" of a traditional gravity escape.
I don't see where the pallet gets lifted. When there are 2 pallets you can see the one being lifted clear of the pendulum. But with this single pallet it remains in contact with the pendulum so I don't see how the energy is transferred from drive weight to pallet to pendulum. Can you enlighten me?

 

[Phil Burman]

Hi tok-tokkie, I thought this might interest you.

The action is very quick so it is difficult to distinguish the individual actions. If you look closely you can see that the pin wheel lifts the gravity arm off the pallet stop pin (attached to the gravity arm) before the pendulum arrives to unlock it.

It is far more noisy than may appear on the video, however there is lots of tuning to be done before the design will get incorporated into a clock.

A simplification may be to eliminate the 12 teeth indexing wheel and have an indexing lever directly engage the 60 pin wheel. This is also borrowed from a Brian Law design.

The proof of concept actually has 15 teeth and pins.

Phil

 

[tok-tokkie]

Thanks. Yes I can now clearly see the lifting of the pallet on the 15 pin proof of concept.
The 60 pin version seems to be very quiet. But, thinking about it during the night, it occurred to me that the escape advances once per complete pendulum cycle so the 60 pin would suit a 'half-second' pendulum. That gives rise to a nice compact Vienna style clock.
I really like the little pinion that reverses the direction so it both unlocks the escape and then raises the pallet.
I have just Googled Brian Law's gravity escape. You took his idea originally but your development to just a pin escape I particularly like. What metal is that - aluminum?

 

[Phil Burman]

What metal is that - aluminum?

Yes I use aluminium for prototypes as it is much cheaper than brass and easier to work than steel. I'm beginning to think aluminium might be an all cases replacement for end plates and other none working parts. I've also been doing some machining with carbon fibre. An anodised aluminium and carbon fibre clock might make an interesting combination.

Phil

 

[Richard Cedar]

I have alway thought that carbon fiber, being light weight, strong and very stable, would be an excellent material for clock parts, especially for large wheels. I think that the combination of the black color of carbon fiber and either brass or steel could be very attractive. I started researching the machining qualities of carbon fiber and kept reading health warnings about the danger / toxicity of carbon fiber particles created during machining. This was sufficient to put me off the idea.

Richard Cedar.

 

[Phil Burman]

Video of the 60 pin prototype gravity escapement with the 12 tooth indexing wheel eliminated and a redesigned indexing lever.

Phil

 

[John MacArthur]

That's pretty classy Phil. Quite simple and robust, and doesn't look like it could possibly have the tripping problem of some gravity escapements. It looks like it could be quite durable. Can't wait to see a whole movement. As Tok says it will run nicely with a half-second pendulum.
Johnny

 

[demoman3955]

I have no clue about any of this, but need to ask how many days would that run with the length of the weights string? The reason i ask is while watching it you could seen the weight dropping pretty quickly, which reminded me of an old tall case o have that has a strange chain and weight set up that i dont understand. I feel its that way to have more chain looping around to make it last longer in the length of the case available, but thats only a guess, and thats the only clock ive seen or owned thats like that.

 

[tok-tokkie]

What strikes me is the opposite approach between my design & yours.
The original design has both a toothed escape wheel and a pin wheel for raising the pallet.
I ended up ditching the pins & using the tips of the teeth of the escape wheel to do the lifting of the pallet.
Here you have ditched the toothed escape wheel & instead use the pin wheel as the escape mechanism.
I looked for a sharp unlocking as the locking face was slid past the tip of the escape wheel tooth.
On yours there is a gentle unlocking as the locking face tip slides around the side of the pin. I notice that the locking face is not radial to the wheel.
Different approaches to the same challenge.
I liked the pinion wheel visually but this is simpler & also offers the opportunity to optimize the action & forces.

 

[Phil Burman]

I have no clue about any of this, but need to ask how many days would that run with the length of the weights string?

The small weight is driving the escape wheel directly, to prove the concept. The finished article will include the movement with a drive ratio of something between 720 to 2800. This will greatly increase the duration but also require a much larger driving weight.

 

[demoman3955]

The small weight is driving the escape wheel directly, to prove the concept. The finished article will include the movement with a drive ratio of something between 720 to 2800. This will greatly increase the duration but also require a much larger driving weight.

ok got ya. I have a few one day and never keep them running because i cant be winding them every day. lol

 

[Phil Burman]

I mounted the latest version of the 60 pin gravity escapement in a movement I have used in the past for testing various escapement types. It doesn't run, even with the massive drive weight. It seems that the escape wheel doesn't provide enough force to raise the gravity arm, I think there are two main issues. The length and weight of the gravity arm and the relatively large diameter of the escape wheel, all of which where directly scaled up from the original 15 tooth escape wheel design together with a drastic increase in the length of the gravity in order to reduce the amplitude to around 1.2 degrees. So time for a major rearrangement.

 

[tok-tokkie]

I learned to minimize the inertia of the pallets and have an adjustable counter poise which let me make it easier for the escape wheel to raise it while simultaneously reducing the impulse. Hence my pallets are 1mm titanium. In my case those worked nicely together so I could get just the swing I wanted. I found it advantageous to increase the drive weight so the lift was quick and decisive rather than relaxed and smooth. As the poise of the pallet is reduced so the raising of the pallet slows unless you increase the drive weight.

My drive train is poor so I have to have spare drive power to reliably unlock the escape. The swing of the pendulum can be seen to vary as the good and poor parts of the drive train come into play. I have not been able to actually identify the bad spots. By reducing the drive weight I get the clock to stop (when the escape fails to unlock). I then mark the mesh of the wheels & pinions but even so I am uncertain who is the ultimate culprit.

I found it a pleasant challenge to sort it all out. May you also enjoy the challenge.

 

[Phil Burman]

My drive train is poor so I have to have spare drive power to reliably unlock the escape. The swing of the pendulum can be seen to vary as the good and poor parts of the drive train come into play.

Thanks for your continued feedback. I am however a bit confused by the above statement. I thought the whole point of a gravity escapement is that it avoids the variation in the force transmitted by the movement. The escape wheel raises the gravity arm a fixed distance and this represents a fixed amount of energy that is transferred to the pendulum when the gravity arm is unlocked by the pendulum. What am I missing?

Phil

 

[tok-tokkie]

The force applied to the pendulum is constant. It is the energy extracted from the pendulum that varies. What varies is the friction force to unlock the escape. The friction force varies directly in proportion to the load applied by the drive at the locking face. It is a very small force but, never the less, if it changes the pendulum has to provide the energy to unlock the escape so the energy extracted from the pendulum varies resulting in a change of the angle of swing so also a change of rate. It is circular error so it is small but the change in applied load persists until the bad spot in the train goes out of mesh. The further up the train from the escape the longer the duration. This error is cumulative.

Prof Hunt of Cambridge University stated that this is the great failing of the gravity escape. That was in a post on my write up of my escape (I think) but it is no longer there.

 

[Phil Burman]

It is a very small force but, never the less, if it changes the pendulum has to provide the energy to unlock the escape so the energy extracted from the pendulum varies resulting in a change of the angle of swing so also a change of rate.

What value in change of amplitude do you see.?

I think the Brian Law design may largely overcome this issue. The unlocking is performed by the gravity arm, after it has separated from the pendulum, and the impulse lever tip can be designed such that it produces very little or no friction force when the pendulum lifts it of the escape pin.

Do you still have your writeup?

Phil

 

[tok-tokkie]

The swing scale on my clock is adjustable so I set 0 at left end of swing then right end of swing shows full swing in mm. At about 48mm it fails to unlock. Initially I set the swing to 54 but since then I have increased that to 56 or 58. I see the amplitude drop to 52 or so.
But there is something else wrong with the clock at present. It stops no matter what. I have been concentrating on something else so have not looked into it. I strongly suspect it needs cleaning & oiling. The pallets run in jewels with end stops. I oiled them with fully synthetic motor oil. I suspect that is where my problem lies. The ball bearings all run completely dry but I doubt that is the cause of the problem. There are tiny bearings at the front where the seconds arbor turns inside the minute arbor which turns inside the hour arbor

I have watched your video at 0.25 speed. I do not see the unlocking happening after separation. On my clock there is a stop for the pallet which determines when the pallet separates from the pendulum. I do not see such a stop in your escape so the pallet stays in contact with the pendulum. The pallet pushes the lock assembly so that it slides down the pin on the pinwheel until the pinwheel is released. Then the pallet is raised so it separates from the pendulum. It is a bit complicated by the fork on the lifting arm as that may stop the pallet earlier.

I agree that you can adjust the friction force by the shape of the locking arm. It depends largely on the radial angle of the locking face wrt the pivot of the escape pinwheel. It seems to point slightly to the right of the pivot which eases the friction force. Lots for you to play with there. I reduced the friction as much as possible by having the titanium locking faces diamond coated. Since there is a minimum charge I also had all the pinions diamond coated at no extra cost.

Here is the link to the discussion on the Trinity clock where Prof Hugh Hunt discusses it

https://mb.nawcc.org/threads/trinity-college-tower-clock-discussion.68648/

That thread was made by extracting the early posts from an earlier thread. In post #15 I refer to the failing of gravity escapes in Hunt's opinion & you will see that that post is not there.

I only wrote up a bit about the escape of my clock.

https://mb.nawcc.org/threads/30-legged-gravity-escape-clock.148968/

I had drafted much more but never posted it.

 

[John MacArthur]

Tok said "I oiled them with fully synthetic motor oil." Tok, as you may remember, a chemical engineer/horologist some time back (can't remember who) said that synthetic motor oil has similar additives to what is in regular motor oil, and that they will eventually corrode and deteriorate metals. This is largely counter-intuitive, but I take it at face value, and have at his recommendation switched to synthetic air compressor oil. I do not have any results yet, and don't really expect to know, perhaps in my lifetime. I do think that synthetic clock oils are a significant advance over old organic ones, but are too light for the more heavily loaded pivots such as barrel and pulley pivots.
Johnny

 

[Phil Burman]

I have watched your video at 0.25 speed. I do not see the unlocking happening after separation. On my clock there is a stop for the pallet which determines when the pallet separates from the pendulum. I do not see such a stop in your escape so the pallet stays in contact with the pendulum. The pallet pushes the lock assembly so that it slides down the pin on the pinwheel until the pinwheel is released. Then the pallet is raised so it separates from the pendulum. It is a bit complicated by the fork on the lifting arm as that may stop the pallet earlier.

You are correct, my bad. Thank you for pointing that out. I haven't yet studied your escapement in detail but in the Brian Law concept the pallet cannot have a stop as it needs to continue to the point of release of the escape wheel and it is the impulse arm contact with the escape wheel pin that stops the gravity arm, which by requirement is after the the unlocking friction has occurred. The act of unlocking will however cause a deceleration in the gravity arm which will progressively reduce the impact of any unlocking friction on the pendulum. It seems to me that the unlocking friction will allow the vagaries of the drive train to influence the impulse to the pendulum, but will only effect a very small part of the total impulse whereas in the dead beat escapement the whole of the impulse is influenced by friction which, combined with the unlocking friction, greatly influences the pendulum amplitude. The minimal effect of the unlocking in the B Law gravity escapement can be further reduced by optimising the geometry and by precise manufacture of components. This seems to be moving into watch making territory!

When considered in isolation it would seem to me that the unlocking friction is indeed the gravity escapements' greatest drawback, but when considered against other competing escapements it would appear to have minimal consequence?

Unfortunately the links in the Trinity College Clock thread you listed appear to be broken.

Thanks for the feedback

Phil

 

[Phil Burman]

The proof of the pudding:

A nine day plot from back in April of B Laws #20 escapement attached to the movement that I have been using to evaluate various escapements. The pendulum has temperature and pressure compensation. It needed a little tweaking for temperature, because of the stability of the amplitude this was the first time I was able to distinguish this level of required temperature compensation from within the general "noise".

Light blue line is amplitude, looks pretty stable to me, amplitude scale in degrees on the right.

Dark blue line is seconds per beat on the left scale. The orange line is atmospheric pressure - 5 mbar per division. The green line is temperature - 0.5 degree C per division. To avoid the pendulum being influenced by the drive weight the movement was wound at regular intervals, hence the downward spikes in amplitude.

 

[tok-tokkie]

The resolution and precision of those Microset traces astonishes me. The variation of the beat is 0.000 004 sec = 0.35 secs/day & you could move that to span the 2 sec line so your clock can be beautifully accurate.

Instead of a tray on the pendulum for regulating weights I have a weight on a M4 screw so it moves 0.07mm per 30° which adjusts the period by 1.26 sec/month. Slight disturbance when you lift it off or replace it but it is lifted with a loop of cord so not very much.

Being able to actually see the amplitude is terrific. It is determined from the period is it not; not by actual measurement of the position of the bob.

You have atmospheric compensation. Below the bob as per Woodward?

I agree that the Brian Law variation of the gravity escape has addressed the unlocking deficiency. It is going to be a challenge to determine how to optimize it. But the Microset will be a great help. Woodward looked to unlock his escapes by a sharp tap rather than sliding motion so it was perfectly consistent. I look forward to your progress reports.

Here is the current link to the Trinity clock The Trinity Clock Lots of analysis from the menu on the left

 

[Phil Burman]

The software for Microset displays down to 0.000001 secs but you need the GPS module for reliable readings down at this level. I believe the internal clock is only good for 1 sec a day.

The amplitude is based on the duration the sensor is block by a suitable flag. The mathematics of converting this into amplitude was developed by David Smith, professor of mathematics at the university of Auckland, New Zealand. What is important for monitoring of course is not the absolute amplitude but the variation.

How did you know I am currently using atmospheric compensation. This particular pendulum also has a super invar pendulum rod, I had to resort to invar rods to compensate for the small amount of brass and steel at the suspension.

I am curious as to why in your design you located the pivot point for the gravity arms a long way from the pendulum pivot point. Doesn't this result in unnecessary friction at the contact point between arm and rod.

One thing that is noticeable with a single gravity arm is that the magnitude of the swing on the arm side is significantly less than on the other side, the Microset calculation for amplitude assumes sinusoidal motion and is not accurate for gravity escapements.

My conclusion at the moment is that I need to build a new movement where all parts of the escapement are easily accessed and adjusted, possibly including the more difficult issue of ocating the pendulum pivot point to match the gravity arm pivot point , hence my question above regarding your design.

Thank you for the current link to the Trinity clock.

Phil

 

[tok-tokkie]

Post #21 you state it has pressure compensation. How have you arranged it?

The maths to convert flag transit to swing amplitude must be complex.

I positioned the gravity arms there because, initially, they pivoted in 2mm bore ball bearings. That was as close as I could get them to the pendulum pivot. I was aware that ball bearings should not be used for oscillatory motion but I took a chance. The clock would stop ticking after a few months &, if I dismantled & washed the bearings it would go again. Eventually I substituted jewels which happened to be able to be fitted into the existing bearing holes but in new brass fittings. All seemed to be well but still the clock runs for a few months only. Much longer now but not indefinitely. So, possibly I should be having a good look at the bearings at the front that hold the three arbors.

Rex Swenson of Australia also used a similar position with no adverse effects. He gives the analysis here Rex Swensen's Web Site Vienna Regulator Page It is a great simplification to the mechanics.
Edit: I have just noticed that he uses a ball bearing on his pallet arbor without problems.

Enjoy the theory sections at the Trinity site. Their target was to detect the tidal influence on the clock & Cambridge is about 50 miles from the sea.

I challenged Hugh Hunt about the positioning of his barometric compensation. It seemed to me the weights should move radially away from the pendulum pivot if they were to have any effect. Moving up minusculy each side of the pivot could not make any difference to my eyes. He very graciously corrected my calculus! The impertinence of challenging the prof of Mech Eng at Trinity College, Cambridge! I knew, of course, that I was wrong but could not see where my error lay.

My clock sits on a Tufnol bracket screwed to the wall. I can reverse it on the bracket but it requires a crux for the pendulum to be added to the clock frame. In the early days that is how I ran it. As you say you can see what is going on & make changes more easily.

It is gratifying that you can see the effect that the impulse has on the pendulum. They are magnificent things which look trivial so disguise their wonderful nature. Having a really small amount of energy extracted makes a significant change to the swing which is then made twice as visible when it is inserted back on the return swing. Dynamics in motion on display. I think your version using pins instead of pointed teeth is a good concept. I also expect it to be an interesting challenge to get it working efficiently. May you enjoy the journey.

 

[tok-tokkie]

I forget my own child! I later realised I had forgotten about the position of the pendulum pivot and took your question to be why were the pallet pivots not coincident. So that is the question I answered.

In the write up of the escape I linked in post #5 I explain that it was because I wanted the pallets to be operated by the lower wider swinging portion of the pendulum that I raised the pendulum. Originally it was at the pallet pivots which are at 2R where R= radius of escape wheel. I raised it to 2.5R initially then to 3R where it now is.

Here is a video I posted in 2012 of the clock when I first got it to run.

A jerry can of water as the drive so I could adjust it. Placed on a pair of planks on top of an old fridge in my outside office. I wanted to publicise it as I was surprised that no one had made a 30-legged escape before.

Initially I started by building a small scale version of the Big Ben double three-legged escape. I could not get that to run. My drive weight is fitted directly onto the hour arbor. So the gear ration hour to seconds is 12X60=720:1. On the double three-legged escape there is another 10:1 gearset to the escape so it indexes 60° giving an overall gear ratio of 7 200:1. I realised that if I could dispense with that then I was in with a chance of getting it to work. 1g of resistance amplifies to 720g instead of 7200g at the drive weight. It was when that was achieved that I posted the initial video.

 

[Phil Burman]

I put a pair of aneroid capsules on the bottom of the pendulum, ala Philip Woodward.

Unfortunately I don't have the math behind the amplitude measurement. However I did come up with my own equation for amplitude base on the assumption that the kinetic energy at the bottom of the swing is equal to the potential energy at the top of the swing. The velocity at the bottom of the swing comes from the time the flag is blocked divided into the effective width (to take account of the sensor beam width) of the flag. It is necessary to take one physical measurement of amplitude in order to calibrate the effective width of the flag.

Amplitude (radians) = v / (l . g) ^ 0.5

It was an interesting exercise but I have no way of inserting it into the Microset software and, like the Auckland professor's solution, it will not work with a typical gravity escapement.

I am now convinced that to get a gravity escapement to work efficiently it is necessary to minimise the mass of the gravity arm(s) similar to that of a deadbeat pallet and to reduce the size and mass of the escape wheel to a similar degree. The down side will be the loss of the overall visual interest, but then my particular goal is precision not aesthetics.

 

[tok-tokkie]

Woodward's placing of the aneroid capsules beneath his pendulum bob was very elegant because the diameters matched. Sattler have theirs attached about 1/3 down the rod which I find rather scruffy.

Yes preservation of energy. But look at the Trinity equations where they introduce a myriad of other factors that come into play.
I recall that they fitted 2 sensors to measure the Trinity clock pendulum transit. I have not checked but I assume the whole section about the software & instrumentation is still there. Of interest but no use if you can't insert it into Microset.

I also was aiming for precision & let the aesthetics be determined by that - but I failed. Your mechanism has two additional moving parts which makes it look much more interesting.

Here is a link to a truly elegant gravity escape clock (also available with proprietary escape) gramat_en – Philippe Wurtz
You can download the drawing of the escape. There used to be lots of diagrams of how it worked. I downloaded them. Here are 3 which may spark some ideas.

 

[Phil Burman]

Post #21 you state it has pressure compensation. How have you arranged it?

Rex Swenson of Australia also used a similar position with no adverse effects. He gives the analysis here Rex Swensen's Web Site Vienna Regulator Page It is a great simplification to the mechanics.

Yes he obtained the gains related to a reduce swing of the pendulum by moving the pendulum pivot point, but I think the correct way to reduce the swing is in the overall geometry of the escapement.

 

[Bryan Mumford]

Phil Burman said:

"The software for Microset displays down to 0.000001 secs but you need the GPS module for reliable readings down at this level. I believe the internal clock is only good for 1 sec a day. "

MicroSet displays down to 1 uS, and is accurate to about 3 uS on single beat readings because of "gate error" in the firmware. That's a second or two per week. If you measure ten beats ("Time: 10") the gate error is overcome in the averaging of ten readings and the accuracy is back to the precision of the timebase, which is one part per million, better than a second per week.

There's a TCXO timebase available which is better than 1 ppm and additionally is temperature stable, which the standard MicroSet is not.

To do better than this you can use the GPS Mode, which will be accurate to fractions of a second per month, or year, or decade, or whatever duration you have the patience to measure. This is so because every measurement is referenced to the "flawless" GPS tick. Any part per million error in MicroSet is a microsecond error over the entire duration of your measurement session, which will be days or weeks. Your error is then tiny fractions of parts per million.

For anyone interested in high precision measurements I will give you advance notice of a new feature I'm working on that will increase the resolution of MicroSet measurements several times, though they are still reliant on the timebase being used (tuned crystal or TCXO). MicroSet was designed to have microsecond resolution and accuracy. That's why the LCD screen and computer software show six decimal places. But internally MicroSet starts with a resolution of one quarter microsecond. This is discarded because MicroSet can only display six decimal places. Additionally, when you measure "Time: 10", MicroSet measures the duration of time for ten readings, divides the whole by ten, and displays an average to six decimal places. Many people use Excel or other software to calculate on the MicroSet data. But by the time MicroSet saves it, you only get microsecond resolution.

I'm creating a "hi-res" mode for MicroSet for people who care about maximum resolution. It preserves the quarter microsecond resolution of readings and it doesn't average down from the whole measurement interval. This means that a one second pendulum measured with "Time: 10" will return a value of 40.000000. You can open these data sets in Excel and divide by 40 to get the one second reading but with a resolution much finer than 6 decimal places. A one second pendulum measured with "Time: 20" returns a value of 80.000000.

This is a feature that will only confuse a large portion of the MicroSet users, who don't work on high precision clocks and for whom an accurate second per week is just fine. Anyone who is working on higher precision timepieces who might be interested in this more technical feature can email me to discuss it. I typically don't follow the forums and may not see any questions posted here.

One other point I might make about the MicroSet display of amplitude is that the Smith Method is available, but is not what is used when you simply "show amplitude" along with atmospheric sensors. The Smith Method requires the sensor be placed at one side of the swing, and you have to enter additional parameters to calculate a result. The simple amplitude display actually measures the duration of time the optical sensor is blocked, which is an analog of amplitude. But, it can be calibrated by observing the actual amplitude and entering this value with "Apply Observed Scale". It's not super high precision, but I don't really know how large the error will be. It won't be much in a clock that maintains a modest stability of amplitude … any error will be small fractions of a degree. And as someone else has already said, what you mostly care about is scale of change rather than absolute value.

 

[Phil Burman]

MicroSet displays down to 1 uS, and is accurate to about 3 uS on single beat readings because of "gate error" in the firmware. That's a second or two per week. If you measure ten beats ("Time: 10") the gate error is overcome in the averaging of ten readings and the accuracy is back to the precision of the timebase, which is one part per million, better than a second per week.

Bryan, how does temperature variation impact on this level of precision for the standard sensor. Say + or - 3 degrees C.

Phil