Bob temperature compensation

Discussion in 'Clock Construction' started by rogerj, Oct 24, 2016.

  1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.
  1. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    I hope this isn't a stupid question, I have only recently got interested in Horology and last year began building a clock. It has a one second pendulum driven by an electronic interpretation of a Hipp Toggle. I have plenty to learn.
    So the question is about the pendulum and bob. I bought 1 metre of 8 mm dia Invar and had a bob made from steel 8" high x 2" dia. With the bob supported at the bottom I had to extend the invar rod with steel (1.5" 8mm steel stock and the rest steel 4ba threaded rod.)
    I understand that ideally the bob should be supported half way up it's height. But have I inadvertently incorporated approximate temp comp for the bob with the 4" of steel at the bottom of the rod ?
    It SEEMS to me that with the bob expanding upwards, its centre point will go higher, but be lowered by the same amount due to expansion of the steel element in the rod ??
     
  2. Jim DuBois

    Jim DuBois Registered User
    Donor NAWCC Member

    Jun 14, 2008
    1,214
    9
    38
    Retired from computer program/product/project mana
    Magnolia, TX
    Country Flag:
    Region Flag:
    Well, the bob will expand at a different rate that will your steel extension even if both are of identical steel. That is due to the mass of both being dissimilar as well as the rod will heat or cool lagging the mass of the bob most likely. The steel extension being in contact with the material of the bob will stay more or less at the temperature of the center of the bob. But, that said, we may well be splitting hairs with worrying about what will be a small percentage of changes in length...the length of the steel piece as well as the length of the bob. Also, temperature compensation was much more important before most of us moved to decent heating systems. A clock in a jewelers store in the northern US in 1880 might see temperature shifts of 40-50-60 degrees F from no heat overnight during winter to sometime in the day a full blast of the old pot bellied stove. Today, most clocks don't see a shift in 24 hours of more than a degree F or so. Also, some of this will depend on how accurate you want your clock to be. Usually in most clocks timekeeping errors in the pendulum with an Invar shaft will be overshadowed by errors in the train, outside vibrations, changes in humidity/air pressure, and so forth. While a Hipp toggle can assist in making a very accurate clock I would test the rate of the pendulum as you have built it and see what it does over time. If you should put a recording thermometer in the case and measure the temps over several days at the same time checking the rate you may find you don't really have a problem with the pendulum assembly.
     
  3. Phil Burman

    Phil Burman Registered User

    Mar 8, 2014
    92
    0
    6
    Suffolk England
    Country Flag:
    Basically the answer is yes. Including a high expansion material support tube on the rating nut and extending to support at the center of the bob is an accepted method of providing temperature compensation. invar doesn't have a zero coefficient of expansion it is approximately 10% that of steel so a 100 cm. rod of invar would expand as much as 10 cm. of steel. Aluminium is a good candidate for temeprature compensation due to its' large coefficient.

    Phil
     
  4. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    Thanks for the helpful replies..
    My Invar supplier only sold 8mm rod in one or two metre lengths. It was expensive and at the time I ordered it I thought I needed a bit more than one metre (was it 46" ?) but two metres would be far too expensive and lots of waste..I regarded the inclusion of the plain steel extensions as a disadvantage. Recently I started to think about getting the bob drilled from the bottom to the halfway point and changing to centre support.
    Fortunately I then had the thought that prompted the question in the OP and it would seem that the mod would have gained me little if anything. I appreciate the finer points raised but the clock is finished and keeping time to standard I had hoped for so I'll leave well alone.
    Roger
     
  5. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    I thought I would add another question to my older thread as its directly related to the above Q&A's.
    The clock generally kept good time but did show variations particularly with temperature so I had added a small 12 watt heater (a resistor), controlled by a digital stat with a 1 degree differential in the case.
    Recently, with the assistance of a friend, I was able to monitor the clock closely with a Raspberry Pi and the the minute by minute variations, due tho the stat cycling at times, could clearly be seen. (Not in the overall rate but in the minute by minute analysis)
    So I wanted to experiment more. I changed the heater to 36 watts and the stat to one with a 0.1 degree differential set to 24C. Although this worked and the rate was within a second a week, the very short term jitter (seen in computer analysis) in the rate was alarming and yesterday at 8.15 GMT I switched the heat off and opened the case door for half an hour. The ambient was 20C and the rate, as expected, IMMEDIATELY slowed to losing.
    Ambient has stayed close to 20C throughout, just dropping to 18C overnight and the pressure stable at around 1023 Mb.
    I've attached a graph of the rate which shows that the clock has now lost about a second in a day and shows no sign of changing at the moment.
    I can't see how the pendulum is over compensated and anyway the effect would not be immediate surely.
    The question is then, Can this losing rate be entirely explained by the increased air density of 4C cooler air ? If yes, how is compensated for in pendulum design.
    [​IMG]
    The red line/arrow indicate when the case temp was changed abruptly and the rate beforehand was typical of the preceding three days. The other two red lines are at midnight.
     
  6. Tinker Dwight

    Tinker Dwight Registered User

    Oct 11, 2010
    13,680
    28
    0
    Calif. USA
    Air density is a tiny factor. Cooling would cause the air to
    be denser. This would not slow the clock, it would speed it
    up by increasing friction, shortening the natural swing.
    It might be an issue with your heating method. It is difficult
    to get a uniform temperature without using an outside chamber.
    You'd tend to get a very stratified temperature because of
    the losses through the case.
    You can buy that aluminum covered foam insulation cheaply
    and make a chamber. A small muffin fan can keep the temperature
    even inside.
    Tinker Dwight
     
  7. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    Thank you for your thoughts Tinker. Can I add that that the total arc of the pendulum is 2.5 degrees and it is under the control of an opto/electronic Hipp Toggle. Although I can't rule out a change of arc, and hence circular error, the time between impulses remains unchanged at 52 seconds. (Admittedly it can only change in 2 seconds increments.
    I absolutely accept what you say about the temperature control I tried for just thee days. it was disaster for all the reasons you gave and that's why the trial was stopped. I have already made another heater from an old cpu cooler with a 2" fan, plus reduced the wattage back to 12 for another trial sometime..
    However the denser air making it run slower (while the barometric pressure remains the same) has confounded me..but that's not surprising being new to this.
    So what I am asking is, what WOULD cause a pendulum clock, which was keeping good time, to instantly start losing a sec/day when the air temp was abruptly lowered by 4dC. ?
     
  8. Tinker Dwight

    Tinker Dwight Registered User

    Oct 11, 2010
    13,680
    28
    0
    Calif. USA
    Think about what the pendulum is doing.
    In a vacuum, it crosses the center with a set amount
    of energy.
    This carries the bob out some distance where it gains
    in altitude equals that energy. It takes some amount of time
    to slow to a stop and swing the other way. That
    time is what causes the rate.
    Each part of the cycle takes 1/4 of the total time.
    Now, we put it in air. Friction robs some of that energy
    as does moving some mass of air.
    The time it takes to reach the altitude, equal to the energy,
    is reduced. It slows the speed of the pendulum but that
    means it takes less time to swing out.
    There is a slight increase in time it takes on the return
    to center but not as much energy is lost as on the outward swing
    where it had a larger starting velocity.
    The pendulum is not making a large enough swing to make circular
    error an part of slowing the rate.
    It is the increased loss of energy that speeds up the cycle by reducing
    the natural swing distance.
    It isn't until the air gets really viscus like water that the bob's return trip
    from the end of the swing is so slow that the rate is slower.
    In air, the return speed is closer to the pendulums natural vacuum
    speed from a shorter swing. More time is shortened on the outward
    swing because of the greater loss of energy, from the higher velocity.
    It is another one of those counter intuitive things like how the center
    of gravity being lower with the same moment of inertia makes the
    pendulum speed up.
    As the viscosity goes up there would be some point that an increase
    in viscosity would slow the pendulum rate. A pendulum with big wings
    would be past that threshold. It would run with a slower rate.
    If plotted, rate against viscosity, you'd see it was a humped curve
    and not a straight line.
    I've never looked at the math but I suspect it is around where 1/2
    the energy is lost to the viscus fluid.
    A good heavy bob is not over that hump.
    Tinker Dwight
     
  9. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    The bob in this case is 8"tall x 2 " diameter and weighs 6lb so I presume there is no way it is over that hump.
    Can you confirm that in the circumstances I've described, where all environmental things I can think of remain the same, that you would expect the clock to speed up when subjected to the 4C temp drop ?
    Of course I'm also puzzled why, after an hour or two, the clock hasn't speed as the pendulum settled at a new shorter length as you would expect if the temp comp wasn't correct - which I'm pretty sure it's not..
     
  10. Tinker Dwight

    Tinker Dwight Registered User

    Oct 11, 2010
    13,680
    28
    0
    Calif. USA
    The length to the mass is the main contributor to the rate.
    If it got colder and there was no compensation, I'd expect to see the
    rate faster.
    If your seeing slower, I'd expect it was a problem of over compensation.
    Do remember, unless tested, you don't know the exact temperature
    coefficient of any of the parts you've constructed with.
    It is always best to tweak the temperature compensation with the
    rate close first. The rate doesn't have to be supper exact for the first run.
    Both usually effect each other a little but should converge.
    Tinker Dwight
     
  11. John MacArthur

    John MacArthur Registered User
    NAWCC Member

    To throw in another variable: denser air is by definition more bouyant, lowering the weight (though not the mass) of the bob. This is essentially "floating" the bob more. Less weight means less restoring force of g to center on the pendulum. Just another mean little trick things play on us precision pendulum guys.

    Johnny
     
  12. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    Morning all..well it is in the Uk...perhaps the best day of the year in prospect ! Too good to be staying indoors messing with clock so my little Austin 7 is going for a run with the club..
    As for the clock, it's remarkable what sleeping on a problem can achieve..It suddenly dawned on me that I've been looking at the wrong problem ..ie an explanation for why the clock slowed when I thought it would gain..In fact it resorted to being a normal clock and running at the speed it should do.
    It was the condition prior to dropping the temperature was the abnormal condition ! For whatever reason - and they don't really matter - with the crude temperature control and the air currents and wildly differing and moving temp layers I had had created (alluded to by Tinker in #6) that was no good and I had re-rated for that condition.
    Lesson learned.
    My experiment with small heaters was an attempt to avoid getting the temp comp correct this by keeping the temp constant. Mistake.

    I'm a fan of Synchronome master clocks and eventually decided to have a go at making something similar..There is nothing beautiful about way mine is made. Synchronome claimed within 2 seconds /week and the aim was to equal, or perhaps better that. When I first ran the computer prog with the original gentler temp control operating it achieved +/- 0.25 sec for a week with final loss of 0.16 seconds..
    This kind of thing can get obsessive, hence the "improved" (but useless) change to a higher temp.
    I'll start a trial without temp comp and discover what I need to do to improve the compensation.
    BTW @John..I looked at your web page and admired those beautiful clocks..Fantastic workmanship..
     
  13. John MacArthur

    John MacArthur Registered User
    NAWCC Member

    [QUOTE=rogerj;BTW @John..I looked at your web page and admired those beautiful clocks..Fantastic workmanship..[/QUOTE]

    Thank you Roger. I'm nearly done with #4, and will (soon, I promise) post a thread of its construction.

    Johnny
     
  14. tok-tokkie

    tok-tokkie Registered User

    Nov 25, 2010
    262
    0
    16
    Cape Town, South Africa
    Country Flag:
    You ask what the time difference would be for a 4°C temp change.
    Your bob is 8 inches long = 200mm in round terms.
    It is supported at the bottom. Assuming the invar is temperature stable so the lower edge remains exactly where it was relative to the hinge point.
    Using the elementary period equation for a pendulum (in MS Excel notation) for a single swing (half of the complete cycle).
    T = Pi()*SQRT(L/g)
    L=T^2*9.81/Pi()^2 where T=1 & g=9.81 m/s^2
    L=993.961mm the length for a 1 second pendulum.
    Simplifying it as a pendulum with zero weight rod then that is the length to the middle of the bob.

    Now the expansion of your bob for 4°C.
    dl = L*a*t where dl = change in length. L=0.1m (distance from bottom of bob to its center). a = alpha = coefficient of linear expansion of steel = 0.000 013 m/m°C
    dl = 0.1*0.000 013*4 = 0.000 005 m

    So the period of the expanded pendulum will be:
    T = Pi()*SQRT((0.993961+0.000 005)/9.81) = 1.000 003 secs
    There are 86 400 secs per day. So your clock will take 86 400*1.000 003 = 86 400.234 secs to do that.
    You clock will be 0.23 secs/day slow.

    Complete maths of a pendulum are given at the Trinity Clock site at Cambridge University. Right now the site is down but Google should take you there.
     
  15. Phil Burman

    Phil Burman Registered User

    Mar 8, 2014
    92
    0
    6
    Suffolk England
    Country Flag:
    Hi Tok, I don't think you can ignore the thermal expansion of the invar. Invar has a Coefficient approximately one tenth that of steel but the rod is approximately 10 times longer than the bob to its' centre, so I think the OP is right that the setup is approximately self compensating. However the time/temperature lag difference between the rod and the bob will make a mess of any short term (read several hours) variations.

    invar 0.00000113 m/m/deg C
    steel 0.00001200 m/m/deg C

    Phil:)
     
  16. Tinker Dwight

    Tinker Dwight Registered User

    Oct 11, 2010
    13,680
    28
    0
    Calif. USA
    The ideas is as Phil shows. You want the compensater to be
    about 1/10 the length of the invar if steel. The problem is these
    numbers show more digits than the +- of these materials actually
    have. The steel can vary quite a bit. The optimal would be to have it
    with 2 nuts. One on the inside of the bob to adjust the percent of
    the bob used for compensation and a second to set the relative
    length position of the bob for the rate.
    Roger's plot shows hours so I'd suspect that differential thermal inertia
    wasn't the issue. Stratified air in the clock enclosure was more likely the
    problem.
    Tinker Dwight
     
  17. tok-tokkie

    tok-tokkie Registered User

    Nov 25, 2010
    262
    0
    16
    Cape Town, South Africa
    Country Flag:
    My calculation was inverted. The bob would be raised by the expansion so the length should be reduced and the period would also be reduced. But the delta would be the same.
    As for the expansion of invar. It is very dependent on the actual composition of the alloy. Also I have seen reports where the invar rod is heat cycled before use to stabilize its thermal expansion. In the case of Philip Woodward's W5 clock he found that there was perfect zero expansion of his invar pendulum - though it was much shorter than a 1 second pendulum.
    I was wanting to show that simply by supporting the bob at the underside instead of the center the period of the pendulum is affected by a quite small temperature change. The effect of expansion is very marked.
    The optimum material is fused quartz.
     
  18. Phil Burman

    Phil Burman Registered User

    Mar 8, 2014
    92
    0
    6
    Suffolk England
    Country Flag:
    Tok you were right the first time, the OP reported a 4 deg C drop when he switched off the heater, the fan and opened the door. With regard to invar coefficient, I'm begining to think you may be right. I currently have zero compensation on my pendulum and that is starting to look like the correct setting for zero temperature effect. The calculated length of the compensation tube (303 stainless) was around 140mm but that showed to be way over-compensated. I also read where Woodward reported that his W5 required zero compensation, but he didn't have an explanation as to why. The Microset timer software shows just how complicate things get, with what looks like several effects all going on at the same time. I've given up trying to understand the theory and have finally taken on-board Tinkers advice to run it for 24 hours with stable conditions, then adjust the ambient temperature, wait for stabilisation and then run for another 24 hours. Ignore all the short term noise and use only the stabilised rates at the two temperatures.

    Phil:)
     
  19. Tinker Dwight

    Tinker Dwight Registered User

    Oct 11, 2010
    13,680
    28
    0
    Calif. USA
    Are you calculating the increase in the rotational inertia of the bob,
    when you say zero compensation? Even at the center of the bob, the
    bob grows and adds a small amount of slowing, just because it is bigger
    at a higher temperature.
    This usually means it will under compensate, though. This is the main
    issue with the mercury compensator but then mercury expands quite
    a bit. Most only look at the change in CG.
    Tinker Dwight
     
  20. Phil Burman

    Phil Burman Registered User

    Mar 8, 2014
    92
    0
    6
    Suffolk England
    Country Flag:
    Hi tinker, yes I had considered that but I didn't think the effect could overwhelm the expected increase in length of the 1 metre of invar rod, in any case as you say a rise in temperature would cause the bob to get larger which would increase the MoI due to the L squared effect and this would slow the pendulum down. However I do keep meaning to do the calculation in order to check. At one time I planned to make a spreadsheet taking account of all issues but then realised that the uncertainty about expansion coefficient values would make this into an even more complicated hair pulling exercise. All the books I read recommend to do it by experimentation, then what you get is right regardless of the theory. It looks like I may now have to devise an short, adjustable, reverse compensation arrangement, with the bob hanging from the compensation tube rather than sitting on it!

    Phil:)
     
  21. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    Hi..I've only just realised that the conversation had gone to page 2..Still reading and learning..I've started a trial with no heat and will report back in 3 days if a pattern emerges. It seems that you are all saying that temp comp which changes with expansion, air density and flotation is very complicated. I do try reading up on this stuff but so far have only found these issues explained separately and nothing about how they operate at the same time. I hope the trial will indicate what I need to do.
     
  22. John MacArthur

    John MacArthur Registered User
    NAWCC Member

    *Very* much so... Analysis/theory/opinion about this has been raging on and off for several centuries, and still goes on today among some of the best minds in horology and physics. There still seems to be no absolute agreement. In my thinking, it seems that in the sub-parts-per-million scale, chaos enters into events, with no way to predict the absolute outcome. Woodward calls this "random walk", but even with his rigorous analysis, not everyone agrees that it is purely random. In some ways, it's unfortunate for us precision junkies that we can measure our pendulum timekeeping so accurately that we can see every little fluctuation.

    Johnny
     
  23. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    As promised..here's a graph of the clock's performance without any heaters..It is much better than when the heating was running which is a lesson learned. However I haven't learned anything more because the pressure has been high throughout 1020 > 1030 and the ambient temp close to 20C..maybe dropped a little at night. The trial will continue to see where it leads..You can see that until midday Thursday it was very good indeed, then quickly started losing and adopted a slower losing rate until now when it is 0.14 sec slow..I can't see any reason for this or any correlation with any environmental changes..I will see how things are after a week. Roger
    [​IMG]
     
  24. Tinker Dwight

    Tinker Dwight Registered User

    Oct 11, 2010
    13,680
    28
    0
    Calif. USA
    It looks more like there was a peak on Thurs. than a drop started Thurs.
    Overall, it looks like it is doing well.
    I'm not sure how you are driving the pendulum?
    You need to record humidity, pressure and temperature
    along with the rate to determine anything useful.
    Tinker Dwight
     
  25. Phil Burman

    Phil Burman Registered User

    Mar 8, 2014
    92
    0
    6
    Suffolk England
    Country Flag:
    The science of clocks and watches by A L Rawlings page 71 gives the buoyancy error due to change in atmospheric pressure:

    cast iron @ 7 kg/litre = 0.0073 seconds per day per millibar
    brass @ 8.5 = 0.0060
    lead @ 11.3 = 0.0045

    You can prorate for materials of other densities.

    So for a barometric change from say 1020 millibar to 980 millibar with a lead bob would give a rate change of 0.18 seconds per day.

    Phil:)
     
  26. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    @ Tinker: The pendulum is being impulsed every 52 seconds in true Matthias Hipp style..ie an electro-magnet beneath the bob.
    I would really like to include temp and pressure reading in the recordings but I'm using a small program written for the Raspberry pi which was given me and I have no programming skills to alter it. I do have a weather station though that records and I'll experiment with combining the results even if only in Photoshop.
    @ Phil: Thank you for those figures.My bob is mild steel so close to cast iron..With the stable high that have predominated over they past week it proved to me that the variations are more temp related - or due to mechanical imperfections caused by the not very precision which I was able to use in the construction.
    Meantime, as you know, clock watching is a slow business, so I will certainly come back here when I have something definite to report..Roger
     
  27. Tinker Dwight

    Tinker Dwight Registered User

    Oct 11, 2010
    13,680
    28
    0
    Calif. USA
    Are you mechanically firing the coil or using an opto-electronic
    interrupter?
    I don't think it effects things but it may.
    It could be that the number of swings changes. You might be right on the edge
    of changing the number of swings. That would show such erratic behavior.
    It might be a good time to learn some coding.
    I prefer the Arduinos look-a-likes, simply because they are so cheap and you
    can dedicate each to its own function. Humidity, temperature and pressure sensors
    are easy to connect and there is code already done to access these devices.
    The can also monitor the # or swings and what ever else you'd like.
    Tinker Dwight
     
  28. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    I have written up the cHipp Toggle clock, Tinker. Google "Hipp Toggle" it will be obvious which link to follow..But do excuse the ramblings of an amateur if you go there..As far as the time between impulses is concerned I have never seen it change from 52 seconds when I occasionally time it. The exact swing amplitude can vary within a couple of millimetres (width of the "flag") can change within that 2 sec window I think.
    I did play with an Arduino before a friend offered me help with a Pi program..I have tried to get my son, who does know how to code, to give me a hand but so far without success !! I doubt I have the ability to learn coding at my age but don't tell me that 10 year olds can do it,,that hurts :)
    Basically, I find monitoring a clock in this way opens a whole can of worms..Today's downward trend in the rate looks terrible...but in fact when I used to monitor by comparison with a radio controlled wall clock, I would never have noticed....and been pretty happy with it. lol
    I've added a graph of the performance up to today..Just short of a week's worth. The pressure remains generally high - 1024 yesterday and this morning it's up to 1036. The short term blips must be temp related but how much of the drift downward is due to a rating error and how much to pressure, time will tell. I'll leave it alone for another week at least and hope for a passing low !
    Roger
     
  29. Phil Burman

    Phil Burman Registered User

    Mar 8, 2014
    92
    0
    6
    Suffolk England
    Country Flag:
  30. Tinker Dwight

    Tinker Dwight Registered User

    Oct 11, 2010
    13,680
    28
    0
    Calif. USA
    What are you using as a time reference?
    As long as the drift rate is relatively constant it
    is doing well.
    I understand the way a Hipp toggle works but also
    know it can be adjusted to center on a particular
    swing count.
    Tinker Dwight
     
  31. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    Hi Tinker..
    The Pi software was designed for a Synchronome so expects a pulse every 30 secs. I added a /30 in hardware to make it work with mine..Every 30 seconds it records a time-stamp from a NTP time server on the net..
    I tried to word it so it didn't look like you needed to look up how a Hipp Toggle works ! But I don't think I succeeded.
    If you do Google "Hipp Toggle" it will be the first hit..Don't know why Google thinks it's the most relevant source of info out there....
    There is a full description with my simple circuit diagrams...
    Originally I timed it to 30 seconds but recently removed some resistance from in series with the impulse magnet (that was easy to get at without stopping the clock) and it went to 52..Sometime I'll try adjusting it to 60 seconds
    Roger
     
  32. Tinker Dwight

    Tinker Dwight Registered User

    Oct 11, 2010
    13,680
    28
    0
    Calif. USA
    What clock source are you using as a reference?
    It isn't a problem of software, it is a problem of reference used
    that I'm asking.
    Tinker Dwight
     
  33. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    Every 30 seconds the software is triggered to see what the offset error is between the pulse and the time off an NTP server. A value in milliseconds is recorded. The average error over every 5 min period goes to the graph..Does that make sense ? Roger
     
  34. Tinker Dwight

    Tinker Dwight Registered User

    Oct 11, 2010
    13,680
    28
    0
    Calif. USA
    #34 Tinker Dwight, Apr 19, 2017
    Last edited: Apr 19, 2017

    I doubt the server is providing an accurate time standard to the level you need.
    The best you can use is either a temperature compensated quartz oscillator
    or a GPS that has a one second time pulse out.
    Your server shouldn't have any long term drift but I wouldn't rely on it for
    sub millisecond timing.
    Unless you want to see some other funny aliasing you might want to set
    the power pulses to an prime number, not near 30 or 60.
    Are you using a true mechanical switch or an optical interrupter
    for the Hipp Toggle?
    Tinker Dwight
     
  35. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    Hi Tinker.
    In reverse order..yes I am using an optical interrupter as the Hipp detector.
    Next time I alter the impulse period I will try for 59....but I wish I understood the reason !
    On the 17th Mar I asked my friend to send me a graph of the un-averaged raw data over a short period..Here' a short sample of the going showing what we think are the small cyclic circular errors, albeit samples of a 52 sec period wave sampled every 30 secs.
    [​IMG]

    I'm not knowledgeable enough to debate the accuracy of an NTP server but I have discussed it with my friend who had thought of the possibility of short term jitter from the internet route and the WiFi to the Pi. I believe that's why he had decided on taken a 5 minute average of the previous 30 second samples. As for the long term accuracy we had never doubted it...I'd be interested in knowing why you question it..Certainly his Synchronome or my DIY clock are ever going to break any records..Roger
     
  36. Tinker Dwight

    Tinker Dwight Registered User

    Oct 11, 2010
    13,680
    28
    0
    Calif. USA
    Most computers only need to check the time once a minute.
    The build in time reverence is the over driven crystal used for the
    processor speed.
    Then they use an estimated error from a two way path over the
    net.
    They are usually within a second of the correct time but usually
    as much as 5-10 milliseconds fast or slow every minute.
    Some of the variation you see is likely from such differences.
    The daily drift rate is real but some of the ups and downs may just
    be the processor responding to its drift and variability.
    In the graph you're showing the short term are likely your pendulum
    but the general waviness may not be.
    Using an prime number of swings will shorten the time of sampling error
    caused by beating against the reference frequency. It ensures that you
    don't have something like a 1 hour beat in the sampling method and seeing
    it as some error in your pendulum. I might choose something like 53 swings.
    If you used something like 60 and the computer was correcting on 60 seconds
    the beat frequency might be hours that you see as long term drift in your
    pendulum. It might actually just be systematic errors in some part of the
    sampling noise.
    An interesting thing to do is to save all the samples and do an FFT on the data.
    Decimate the data at different amount and again run a FFT. This will allow
    you to see the different types of regular noise. These will usually show up
    as peaks at regular points. Your pulse rate of the pendulum should be clear
    but other errors can be hiding.
    Tinker Dwight
     
  37. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    Thanks for that insight..I will pass your info to my friend to see what he makes of it..
    As my clock outputs 1 pps and is divided by 30 in an external board, it would be possible to test the sytem by substituting 1 pps from a GPS receiver (which I already have) Roger
     
  38. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    Hi Tinker..Sorry for the delay in replying but I have been chatting with the writer of the program I've been using on the raspberry Pi computer. I ran a test for a week using a 1pps GPS pulse divided by 30 into the Pi instead of the clock. I'm not able to pass on all he told me because I don't follow all the technicalities but basically he doesn't disagree with anything you said.
    However we did decide that in spite of the points you raise (which he was already aware of ) the system is perfectly good for monitoring a pendulum clock running at atmospheric pressure and domestic temps.
    The graphical results we use are done with adapted open source software called MRTG.
    The graphical display resulting from the weeks trial of the GPS source was a perfectly flat line right on GMT. The resolution of the line does depend on the PC monitor resolution but we reckon it probably takes an error of 10 milliseconds to step up or down a level on my screen.
    The short term errors seemed to be within +/- 3 milliseconds and they generally occur quite slowly. The graph I've attached shows the previous 7 days of the actual clock rate which we think is a good enough indication of how it's going. The temp was not monitored and the scale of the graph is +/- 0.5 seconds.



    [​IMG]
     
  39. Tinker Dwight

    Tinker Dwight Registered User

    Oct 11, 2010
    13,680
    28
    0
    Calif. USA
    If these are absolute numbers, the clock is not doing well.
    One wonders what is the source of the variation. The numbers
    span about 3%.
    It would have been good to get a reading of the environment
    variables for this time span.
    Tinker Dwight
     
  40. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    ooch ! And I thought it was not going too badly. High pressure predominated until the middle of the period when the low arrived overnight. The lowest I saw was was 991 Mb but it may have been lower at some time..Today it's 996. Overall it's only 0.2 secs ahead of where it was a week ago (although it's still gaining) I suspect the rating is a tad out and may be inherently fast but I thought I'd wait another week before fiddling. Unfortunately I don't have a convenient way of recording ongoing temp and pressure. The monitoring continues...Roger
     
  41. Tinker Dwight

    Tinker Dwight Registered User

    Oct 11, 2010
    13,680
    28
    0
    Calif. USA
    Is this a pressure measurement? I was thinking it was the pendulum.
    Tinker Dwight
     
  42. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    Yes Tinker..they are approximate pressure measurements added to the graph with Photoshop..Should have added the "Mb"..So, not so bad after all ? The middle line is GMT...Roger
     
  43. Phil Burman

    Phil Burman Registered User

    Mar 8, 2014
    92
    0
    6
    Suffolk England
    Country Flag:
    Which direction is +0.5s and which is -0.5s.

    Phil

    PS: I assume it is s/day.

    With out ambient temperature I think yo are in the long grass.
     
  44. Phil Burman

    Phil Burman Registered User

    Mar 8, 2014
    92
    0
    6
    Suffolk England
    Country Flag:
    A pendulum will lose time with higher atmospheric pressure due to flotation. If I read your graph correctly yours is apparently gaining at higher pressure. Bob friction in air and air mass movement both also have the possibility to alter the rate due to a change in air density (read pressure).

    There are several books that will give you insight into this somewhat complicated issue.

    Phil
     
  45. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    #45 rogerj, May 13, 2017
    Last edited: May 13, 2017
    Phil..The graph has lost it references after cropping the full image. The centre horizontal line is GMT. The graph shows the error between the clock and GMT , plotted against time. Trending downward, the clock is losing and vice versa. The vertical height of the displayed section is 1 second, centred on GMT. The dotted verticals are midnight of successive days. The rightmost one is the start of May12th.
    So....at the end of the period shown the clock was ahead of GMT by about 0.25 sec.
    And during the period of high pressure the clock was losing and when the low arrived overnight 9th/10th it began to gain..following the law as expected.
    I hope this clears up the confusion I have inadvertently created. If I ever post another graph I won't make those mistakes again !
    There is a fair degree of temp compensation (the subject of the original post) and the actual temperatures prevailing, with the central heating off (broken), are around 20C with no great variation night or day at present.
     
  46. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    To mark my 100th post, here is a graph from the test we did earlier (re.#38) of the validity of using the computer time service for monitoring clocks. The point "A" on the graph is the point at which the input to the system was swapped from my clock to the NTP/GPS derived 30 sec pulses.
    The blue line is the offset error as described in my previous post and is calculated as an 5 minute average of the raw data as collected every 30 secs
    The red line is ( as I understand it) the rate of change of the raw data ie. NTP/computer clock errors).
    For my purposes I think this is as good as is necessary for this type of clock. I do have an expanded scale version that shows that the blue error line is actually about 4 to 5 Milliseconds later (minus) than GMT due to the inherent delay in the network paths etc. We do wonder if anyone else has used NTP for this purpose..Roger

    [​IMG]
     
  47. Phil Burman

    Phil Burman Registered User

    Mar 8, 2014
    92
    0
    6
    Suffolk England
    Country Flag:
    Hi Roger. As was previously pointed out the rate of temperature equalisation between the bob, the invar rod and the (buried) steel rod section will not be the same due to the difference in relative location, geometry and thermal mass. Without careful measurement of rate versus temperature you cannot just dismiss this aspect as job done. For your configuration even a change in the rate of temperature change will produce different results. If your method of temperature compensation is good then all (invar rod) pendulums would already be designed this way.

    Quantify "fair degree" and " with no great variation"

    The rate change appears to be almost twice what it should be for 34 millibar change in barometric pressure.

    What was the barometric pressure at the end of the 7 days.

    A photo of you pendulum, bob and suspension arrangement would be useful. As previously said, pendulum behaviour is a complex subject. The usefulness of the comments you get will be directly proportional to the amount of information you provide.

    Plotting only the rate drift hides a multitude of issues that can be revealed by a boxcar average plot. Can you not download the raw data to Excel, then you would be able to graph the hell out of the data.

    Phil:)
     
  48. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    Hi Phil..With such a knowledgeable audience I'm a bit reluctant to post detailed pictures and information here about my DIY clock as it is definitely not a work of art..basically it's a hobby electronics project which came about after buying a copy of Electrical Timekeeping by Hope-Jones.
    I bought the book following the purchase of a MK2 Synchronome movement and its subsequent finishing off as a "marriage". After I retired I acquired 4 more 'nomes plus a PO36 and a Gents C7.
    However I have written a blog on the Chipp Toggle clock as I went along and it can be easily found by Googling "Hipp Toggle"
    The idea has never been more than a voyage of discovery rather than an attempt build a near perfect clock. I just hoped it would be as good a timekeeper as a Synchronome - and It has achieved that in spite of its shortcomings.
    Specific answers: temperature compensation is what it is..1 metre of invar plus 1.5 inches of steel plus plus about 3 inches of steel 4ba rating thread. The bottom supported steel bob is 8 inches tall. There is about 1.5 inches of brass in the top chops plus 3/8 inch suspension spring.
    I have no doubt the compensation is not right - but insufficient maths to calculate by how much.
    Since my heating failed a couple weeks ago, A max/min thermometer over the last week shows 16.7 to 20.6 and that would cover the period of the graph in #38
    I completely accept that the variations in rate are twice what one would expect for those barometric changes but I have no idea why.
    At the end of the period shown in #38 the pressure was 996Mb I think. As I write the pressure is showing a 1006 Mb and the clock is neither gaining or losing.
    As I've already said the clock is was not meant to be a "regulator" or anything similar. All this discussion has come about as a result of monitoring it with a computer :) Before that, but excluding the ridiculous period when I tried heating in the case, the clock was - and still is a reasonable timekeeper and has run for several weeks at a time before I need to press the buttons that add or subtract seconds to the count.
    Re Excel. my maths is pretty poor and I don't know how to use Excel. But my friend does and I just send him the data and that's exactly what he does with it..The graphs I get back are from Excel.
    This monitoring system is cheap to set up but not as comprehensive as the commercial equivalent. It would indeed be great if temp and pressure could also be monitored at the same time. Then the observations might be far easier to interpret. But is has been fun..Roger
     
  49. Phil Burman

    Phil Burman Registered User

    Mar 8, 2014
    92
    0
    6
    Suffolk England
    Country Flag:
    Hi again Roger, The rate at the end of the run, with a pressure of 996mb, is slower than the apparent rate at 1025mb (this is the wrong way round to be explained by barometric changes). If your invar rod does not come with a certificate for coefficient of expansion then calculation is useless. For example several people (including Philip Woodward) have reported that their invar pendulums do not require any compensation, this list would now include me. Temperature compensation should not be buried inside the bob and the bob should be supported from its centre. Anything else and you could be chasing your tail for a life time.You can only determine the correct temperature compensation by actual measurement. With a temperature variation of 4 degrees C you are wasting your time considering rate variations related to barometric pressure, they will be swamped by the temperature issue.

    Good Luck

    Phil:)
     
  50. rogerj

    rogerj Registered User

    Dec 21, 2014
    112
    0
    16
    retired broadcast engineer
    Cornwall
    Country Flag:
    Phil..I have taken to heart your comments on temperature effects and will revisit that topic. I have ordered a temp logging thermometer with remote sensor and will install it in the clock.
    Re the central support of the bob in precision clocks (a la Woodward etc) I (and my friend) have long wondered about the compensation of a standard Synchronome master clock with its bottom supported bob. The thermal inertia of the cast 16lb bob must be considerable and we pondered on what, if any, heat treatment of the Invar might have been done by HJ.
    Any thoughts on that would be appreciated.. Roger
    BTW My invar did not come with any certificate...it was bought from a steel stochholder.
     

Share This Page