Weishi MTG-3000 'Multi Function Timegrapher'

[icon]

Hi

I've had this device for a while, but haven't taken it apart until now. I doubt they're very common, so I thought a few photos and a description of its function might be of interest.

It's a device for checking the accuracy of mechanical watches. I'll show the guts of the thing and then explain the information it provides.

This is the front of the unit. Unless my eyes deceive me, they've mispeled 'TMEGRAPHER'. The display is just a monochrome LCD, but that's perfectly adequate for the job. The casing is pretty cheap, but at least it's easy to get into! There's a separate power supply and microphone which we'll get to.



This is a general view of the main board, accessed round the back of the device. More detailed photos are below, but notice that there's inputs for microphone, power, printer and 'control'. I have no information about what control does - printer I can guess, but I don't have any idea how one would interface a printer to it.




This is the microphone input and the area adjacent to it - I'd like to confidently rap it with a screwdriver like mikeselectricstuff and say what it's all for, but I can't. My assumption (based on some waveforms below) is that it takes the analogue signal from the microphone and converts it into digital pulses that the microcontroller can do something with. Notice the lettered test points. I have probed A and E. Believe it or not I've only just noticed that there's a B, C and D looking at the photo. I might have another play. There's a very helpful ground pin.



This is a sideways view near the power socket. That's an Atmel controller, apparently 8051 based.



Nearby we have another microcontroller, a Philips this time, also 8051 based. This device is snuggled up to an (RAiO) LCD controller and some Issi RAM - I not sure what that leaves for the Atmel to do. There's also the main timing oscillator, an 18MHz device with a trim pot labelled 'Freguency Callbration'. I'll come back to that.



This is the power supply. Not at all the same quality as the main unit. It provides +5V and +/-6V. It's held together with hot snot and good luck. The board is just glued into the case. No fuse, and the mains cable is secured with an overhand knot. I can't tell what's at mains voltage and what's not from the placement of the devices. There's an LED on the board , but you can't see it with the lid on, unless you peer through the bars.



This is what comes out of the power supply. Does this look unreasonably noisy? There's a discernible frequency to the noise, with bursts of even worse noise that I didn't capture.  I wonder whether it would be worth knocking up a linear supply - I've got  all the bits.



This is the microphone. It's really quite nicely made. The base and the watch holder are made from some kind of machined, not moulded, engineering plastic. It's designed to rotate and swivel so as to place the watch under test in six different orientations. Watch movements behave differently depending on the orientation of the balance wheel, and sometimes you just have to try to average the errors. The watch is held in place by a sprung slide.



Here it is in another orientation. You can see I've put the crown of the watch against a metal stop which couples to the microphone.



This is the underside of the microphone with the cover removed. All quite solid and nicely machined. Some amplification going on here I presume, before the signal is sent down the cable. That's a TL071 op amp.



I presume those tiny wires are connecting to the microphone itself.



And finally for this post, here's a teaser of what the device shows you. I'll start another post for part 2 - How it Does What it Does. Meanwhile, any questions, fire away.



Cheers
John

[mikeselectricstuff]

The Xtal says VCXO on it - voltage controlled crystal oscillator. Often used in PLL circuits, but here just to allow some manual trim. A bit surprised they didn't use a TCXO in something like this.
The stuff near the mic socket is probably just amplifiers, filtering and thresholding to get a clean tick pulse.

[SeanB]

There was an older model on sale near me on Gumtree a while ago. I wonder if it is still available......

Nice unit, and extremely useful if you are doing clocks and watches, to set the unit to reduce the long term drift and get the accuracy down to a few seconds a day in a quicker method than wearing it for a week or putting it in a autowinder cage for the same time.

Even though I am an Instrument tech by trade, i find wristwatches very finicky to work on, not enough fine motor control on my part. i prefer bigger items that only need a magnifier to work on, not a microscope and manipulator probes with reduction.

As to the power supply, it probably would work better with a different power supply, you can use an old ATX unit, take the 5V straight through and use linear regulators on the + and - 12V rails to provide 6V for the analogue side. I suspect the noise on the supply would be reduced considerably if you removed the mains lead by cutting it off about 5cm on the outside, remove the knot and cut a place to put an IEC socket there, and connect the earth lead to the earth pin on the power supply. Would get rid of the common mode noise and remove a lot of the switching noise as well.

[Hypernova]

Interesting, these things just use a mic to listen to the internal movements of the watch. If you got a noisy factory environment how well would it preform?

[icon]

So, part 2 of this 'review'. There's a little bit more going on with this machine than is apparent at first glance, but I just need to divert into the function of watch escapements to explain it.

A watch escapement is the mechanism that controls the rate at which the energy contained in a coiled spring is released. After the world had experimented with loads of variations on the escapement theme in past times, an English chap by the name of Thomas Mudge came up with the Lever Escapement in around 1755. This escapement caught on pretty quickly and remains in use in mechanical watches to the present day. Only in the latter years of the last century did anyone come up with anything (arguably) better - George Daniels (another English watchmaker) produced the Co-Axial Escapement which was eventually adopted by Omega.

However, back to the plot. The key feature of the Lever Escapement that ensures accurately metered doses of energy is the balance wheel, which is free to rotate about its axis but is constrained by a fine spirally wound spring. Set moving, it will vibrate to and fro, alternately winding and unwinding the hairspring until friction brings it to a halt. The clever bit is that the period of this vibration is relatively independent of the amplitude - provided that the balance wheel is free to rotate without being influenced by the rest of the watch mechanism.

Meanwhile the force of the spring, which is trying to unwind through the gear train, is restrained by one of two tiny jewels (usually) called pallets, which take it in turn to lock a tooth of the escape wheel.

Mudge's genius was in finding a way to allow the balance to rotate unrestrained for most of its period of rotation, except for a brief moment where at the centre point of its travel, at maximum velocity, it both unlocks one tooth of the escape wheel and, vitally, receives a tiny kick of energy which speeds it on its way, making up for frictional losses. The microphone distinguishes three events, but for simplicity I've mentioned the first and last. You can see the raw trace below, and the pulses that the timing machine constructs from it. In 10ms, it's all over and the balance wheel is on its way.



Mostly, the electronics do an amazing job of constructing the pulses, but occasionally they get it wrong.



If you zoom out, you can see the 'duty cycle', which graphically shows in the silent portions how much of the time the balance wheel is detached from the rest of the watch movement, only briefly diving into contact to unlock a tooth and receive a kick.



Now the clever bit. For any given period, a greater amplitude means a faster 'zero crossing'. By timing the interval between the first and last pulses in a group, you can estimate the amplitude. I had an equation for this once upon a time, but I can't find it now. (There's a complication around something called 'lift angle' which I'll spare you) Amplitude is important as an indicator of the health of the watch. As oils degrade and thicken the little kick of energy received by the balance wheel finds it harder to overcome friction.

Finally, there's a measurement of the absolute rate - 'rate error', and of the variation in timing of the alternate directions of rotation - the 'beat error'. Both of these are corrected by altering the geometry of the hairspring attachment. The photo below shows a general view of the balance wheel of an ETA 2824 (the one in the photos in the first thread). The balance is spinning, so you can't see the spokes.



And this is a close up of the same area, showing the bits where adjustments are made. The rate error is corrected by adjusting the length of the hairspring and the beat error by adjusting the position of the attachment point. The little jewel with the trefoil spring is the shock absorbing mechanism that allows the pivots of the balance wheel to survive a knock - they are commonly thinner than a human hair.



Anyway, there you go; more than you could possibly want to know about watch timing.

Regards
John

[SeanB]

Bear in mind it is using conduction to listen, pretty well I would guess, you just need any reasonably quiet office space. Most watchmakers and watch technicians work inn a very quiet environment, almost all you wil hear is the ticking of hundreds of mechanisms. Different at the hour and half hour though, a hundred clocks all strike at nearly the same time.

[amyk]

The Xtal says VCXO on it - voltage controlled crystal oscillator. Often used in PLL circuits, but here just to allow some manual trim. A bit surprised they didn't use a TCXO in something like this.

The accuracy of the mechanical watches that this is designed to calibrate is much worse than even an uncompensated crystal.

As for why there are 2 microcontrollers: The big one has its clock input connected directly to the 18MHz VCXO, LCD controller, and microphone input circuitry. The 18MHz frequency is distinctive because it is divisible by all the common mechanical watch frequencies (5, 6, 8, and 10Hz) --- that MCU code was very likely written with exact cycle counts in mind, as its being used as a timer. The small one has a separate 12MHz crystal and runs non-time-critical parts of the system like the buttons and I/O: printer and control being 3 pins, they look like serial ports for outputting data to a printer and receiving commands.

(Also noticed it has a speaker next to the small one, does it beep every time you press a button? )

[icon]

(Also noticed it has a speaker next to the small one, does it beep every time you press a button? )

No - you can turn on an audible 'tick' which gives you acoustical assurance that you're capturing data accurately. You can set the sensitivity of the microphone from the front panel and when it's wrong you hear a jumble of irregular clicks.

John

[baljemmett]

Pretty cool to see the guts of one of these -- I remember seeing something similar in the quartz watch episode of Tim Hunkin's excellent The Secret Life of Machines series many years ago, and thinking it looked like a neat piece of kit.  Thanks also for the description of how it works!

Wonder what it'd make of the Bulova Accutron I've got on my wrist at the moment? 

[Aleph]

The Xtal says VCXO on it - voltage controlled crystal oscillator. Often used in PLL circuits, but here just to allow some manual trim. A bit surprised they didn't use a TCXO in something like this.

The accuracy of the mechanical watches that this is designed to calibrate is much worse than even an uncompensated crystal.

You are right. This instrument has not been definitely designed to calibrate mechanical watches with COSC certificate, which dictates average daily rate to be in range between -4/+6 s/d. For COSC certificated mechanical watches the accuracy of the instrument should be about 0.1 s/d.

Very nice topic by the way. As an horology enthusiast, I have a strong desire to build such an instrument for my workbench. Unfortunately I still don't have enough knowledge and experience in electronics. I am working on it though..