[TC28] Pendulum Clock

[ord]

Nice work! Congratulations on getting it so accurate.

[JunkstyleGio]

Lovely work. I really like the looks! Nice video by the way!

About the accuracy:  0,12 per 5 minutes slower/faster is about 35 seconds per day.  (about 3h33m per year.)

I really would lke to see the result of a 24 hour test!

 

Edited February 9 by JunkstyleGio

[Berthil]

I see the clock more as demo of what can be achieved with Technic. It is possible to get it accurate with a lot of testing and shifting the upper and lower pendulum weights, but in the end it is still LEGO. I expect ambient temperature will have the most influence on the clock, as with real clocks.

[Davidz90]

Outstanding work!

2 hours ago, Berthil said:

I see the clock more as demo of what can be achieved with Technic. It is possible to get it accurate with a lot of testing and shifting the upper and lower pendulum weights, but in the end it is still LEGO. I expect ambient temperature will have the most influence on the clock, as with real clocks.

True. Thermal expansion results in about 3 seconds/day error per 1 degree C change of temperature. The current 35 seconds/day is extremely good for such a small and light pendulum. 

Out of curiosity, if it isn't too much of a trouble, could you record the sound of ticking (5-10 minutes is enough) and post it somewhere online? I could run some statistics on it, find the average period, standard deviation, difference between left and right swing time etc...

Edited February 9 by Davidz90

[aeh5040]

Congratulations - this is a very nice and interesting piece of work. Impressive that you managed to get it so accurate. Some might question the fact that getting on for half the time is spent (auto) winding! But this seems a perfectly reasonable compromise, of the kind that every Lego model inevitably involves.

Edited February 10 by aeh5040

[Saruzeufel]

Very impressive. Not only the very complex mechanism, but also the looks are great! Love the scientific approach to making it most accurate and measuring drift

[Berthil]

Thank you all! Next week I'm home-alone and hope to run the clock for 12 hours with a timelapse video.

On 2/9/2025 at 3:33 PM, Davidz90 said:

Out of curiosity, if it isn't too much of a trouble, could you record the sound of ticking (5-10 minutes is enough) and post it somewhere online? I could run some statistics on it, find the average period, standard deviation, difference between left and right swing time etc...

I will have a look if I can provide as I already have recorded the clock for 5 minutes, I should be able to export the sound only and make it available to you. I'll let you know via PM. I expect the standard deviation is not that good because when the driver weight is lifted, the clocks runs noticeably faster.

[Berthil]

The 12h Timelapse is done and I'm pretty happy with it as this was my first Timelapse video ever. I had to do it in 3 takes because the camera battery life does not go further than 4 hours and a bit. 1 photo every minute so the position of the seconds hand shows the deviation to the one minute. There is a sunset and sundown but unfortunately not visible in the video light. On the watch next to the clock it is visible that the clock is very accurate. I had to crawl back from the kitchen to make sure I wasn't in the background 

A building instruction is in preparation and will be published after the competition.

 

[Davidz90]

Wow, the timelapse looks so cool! Maximum error of 40 seconds in 12 hour run is really good!

Thank you for the data, I'm back with some interesting stats. The activations of autorewinder are easily visible on the audio recording. When it runs, the clock is considerably faster; for a moment, double period drops from 2.05 seconds to about 1.9 seconds (explanation why I'm using twice the period is below). Interestingly, in times between rewinds, period is slowly increasing. This means that driving torque is decreasing as the weight goes down. One potential cause is the string pooling up on the spool; the more string there is on the spool, the larger its effective radius becomes, resulting in more torque,

Next, we can look in detail at the audio signature:

There is a pronounced asymmetry in the escapement, with every second tick being much louder than the other; this is also audible in the video. Moreover, times of the left/right swing differ significantly, with 0.6:0.4 ratio. Apart from ticks caused by escape wheel impacting one of the the pallets, there are also weaker sound signatures of escape wheel slipping out of the pallet. Due to that louder/quieter tick phenomenon, it was easier for the program to catch only the loud ticks, so all periods are doubled. 

Next, period histogram:

There are two main peaks corresponding to autorewinder being on and off. The mean is remarkably close to two seconds, with 0.2% error. This corresponds to about 3 minutes/day. This is more or less in agreement with the 12 hour timelapse. Interestingly, the clock is actualy better in timelapse, with about 1.5 minutes/day.

Finally, the total error. Here, the impact of autorewinder is the most visible. Between rewinds, clock is consistently too slow, accumulating error. During rewinds, the error is reduced.

 

Edited February 21 by Davidz90

[Berthil]

Thanks for the stats! In the 12h video I've put the rewinder on max power on the 9V train speed controller. In the 5 min. video the controller is on half power which of course make the rewinder go slower but also causes the PowerFuctions switch into a lingering half on / half off mode which makes the rewinder even go slower. So that could also be visible in your stats. With the 9V train controller this happens less often but when it does, the rewind period is still shorter.  When a battery box is used as the power source the power is always maximum so that is good.

9 hours ago, Davidz90 said:

Maximum error of 40 seconds in 12 hour run is really good

In the 12 hours I had to stop and restart the clock two times because of the battery life limit on the camera, so I think the error is even smaller than that because the main error in the 40 seconds is the restart and I can't control the seconds hand and time of Timelapse picture that precise on a restart. When watching the three runs of 4 hours separately there is on average no more than 25 seconds over the 3 stints, that's 0,17% deviation over 4 hours.

9 hours ago, Davidz90 said:

This means that driving torque is decreasing as the weight goes down

This must indeed be because the circumference on the string drum is getting less when the weight is going down because there is less string on the drum, and as a result, so is the torque on the mechanism via the drum axle. You are correct with your analysis. When using a chain this effect would no be there, but the direction of the mechanism doesn't allow for a chain. Very nice to see in the data that it is that sensitive, thanks again!

Edited February 21 by Berthil
typos, chain remark added

[Berthil]

The Pendulum clock has been added to the entry topic including a final video with better clarification of the 12h Timelapse.

 

[msk6003]

How did you solve the problem of the watch speeding up when rewinding the weight? In the video, it seems to speed up slightly on the first rewind, and it was the biggest problem I couldn't solve while making the watch. That's why in the two watches I posted here, I solved this problem by attaching the weight to a chain instead of the usual thread and weight combination, and the weight goes up on that chain.

(I know second one is not match 1sec but that is because I misplaced pendular weight.)

[Davidz90]

7 hours ago, msk6003 said:

How did you solve the problem of the watch speeding up when rewinding the weight?

He did not, it does indeed speed up quite a bit when rewinding, as shown on my plots. However, that is not a problem in itself, as long as speedup is consistent.

In general, there are two solutions to avoid speedup:

1. Make the rewinder as slow as possible. That reduces the extra force exerted by rewinder when it runs.

2. Use an escapement that doesn't care about extra force. 

You are correct that weight climbing up a chain is less problematic than other rewinder types. The chain acts as a separation between motor and escapement, making the extra force smaller. That's the main reason why I used this type of rewinder in my contest entry.

Edited February 27 by Davidz90

[Berthil]

It seems the extra speed averages out well, even in the 5 min. run in the video but certainly in the three four hour timelapse runs. A weight climbing up a chain is not feasible in this very compact table design.

Edited February 27 by Berthil
typos