Harwell Dekatron / WITCH restoration during 2012
Post this page to popular social media
Each TNMOC project has either a working group or project team assigned to do the work. Working groups are either managed in association with the CCS (Computer Conservation Society) or solely within the Museum.
The latest updates on the Harwell Dekatron / WITCH Computer restoration project during 2012.
For more detailed information on The Harwell Dekatron / WITCH computer, please visit the CCS Harwell Dekatron / WITCH page.
19/05/2012 update from Johan Iversen
Work has now started on the first of the two accumulators. The heater transformer passed its insulation test and the unit with no sub-assemblies (the carry and transfer modules) plugged in, powered up. After an open-circuit 1M ohm resistor was replaced on one of the dekatrons along with a few flying leads from the pulse generator, oh! and a helping hand, we were able to get the dekatrons to spin (Gallery Image 2) Meanwhile Eddie is cleaning the contacts on all the plug-in units and then giving them a good polish. Also this week Dick Barnes (one of the original designers) paid us a visit. Delwyn, Eddie and I had a great chat with him picked his brains about various things and got a few stories too. Dick could see the progress being made and left very happy. (Gallery Image 1)
18/03/2012 update from Delwyn Holroyd
In the last few weeks work has continued on the pulse generator. The major problem has proven to be trigger tubes. These are cold cathode neon filled tubes that strike when a certain voltage is applied to the trigger input. The type used in the pulse generator is a specialized high speed part (a relative term!) and are consequently now rare and difficult to obtain.
We have a large number of spares which we set about testing. Tony F had built a trigger tube tester for another type used in the WITCH, and it was a simple operation to add an additional socket to test the high speed G1/371K. Gallery Image 3 shows the inside of the tester (the glowing parts are voltage stabilizers) and the right picture below shows a working trigger tube on the tester. The glow comes from the priming gap, which is always lit and creates enough ionization in the tube to enable very fast triggering. To our dismay we found that the majority of our spares do not work, and we have only seven working ones in total! The machine requires six, of which two could be considered optional.
This problem has come as a surprise since neon tubes of the era generally last well - the most likely scenario is that the gas has escaped or reacted with some contaminant left over from the manufacturing process. We are now actively seeking more of these parts in the hope of finding enough working ones to give adequate spares. Compatible types are G1/370K, G1/371K and CV2224. Please get in touch with the museum if you are able to help.
Two faults developed in the oscillator section whilst work was ongoing - a valve failure and a gradually worsening instability which was traced to a timing capacitor. These parts have been replaced and the oscillator is now working well. Gallery Image 5 shows the three waveforms +A, -B and -A pulses. The negative A and B pulses are supplied to the guide inputs of the many Dekatron counting tubes used in the machine, and these pulses overlap slightly to ensure reliable stepping of the Dekatrons. The spikes on the +B waveform shown in the previous update are now not thought to be significant as they are effectively removed by the following stage in the circuit.
A number of pulse trains are derived from the oscillator using a Dekatron to count out ten pulses and a switching circuit to chop these into 1B and 9B pulse trains. Following the first ten pulses, further pulses are generated for the carry units which continue until all carry over is complete. For testing purposes we connected the carry pulse output directly back into the check input. We were then able to operate the unit in continuous mode which makes it easy to examine all the signals on an oscilloscope.
We found there were additional narrow pulses on the switched B pulse outputs. This problem was traced to a pair of germanium diodes which instead of clamping pulses were differentiating them, causing the switching circuits to switch in the middle of a pulse. These diodes are too modern for the age of the machine so must already have been replacements for what would have been metal rectifiers originally. We used a modern silicon diode replacement. We did debate whether to use contemporary components, but given the extremely long operational life of the machine it isn't clear what age would be most appropriate to use! The use of modern components means replacements are obvious to the trained eye, but to the casual observer overall appearance is not affected. All original components are being retained and modifications documented to aid any future research.
Another function of the pulse generator is to generate signals used in the stabilizer to switch anode voltage to the trigger tubes in the carry and transfer units, thus extinguishing them at appropriate times in the operation sequence. We found these signals were overlaid with very narrow spikes which would possibly cause the trigger tubes to extinguish at the wrong time. After a lot of head scratching we realised the root cause was a diode used to mix a pulse train with a ramp signal which was causing ringing on the pulse edges. On inspection the diode was clearly not an original part - a section of lead from the original component was still in place! The original metal rectifier would have had enough inherent capacitance to prevent the ringing, unlike the much faster replacement diode. We solved the problem by adding a small amount of capacitance to the mixing node to damp the ringing. The pictures show the additional component (Gallery Image 6) and the waveforms (Gallery Image 7) after this modification (the top anode switching waveform is formed by subtracting the lower two).
Work is now almost complete on the pulse generator. Our confidence in the power supplies has also increased following many hours of operation during work on the pulse generator. One of the final tasks is to measure the current consumed on the various supply rails so that we can eventually calculate a figure for the whole machine.
Eddie has tested all the high voltage wiring on the racks in preparation for powering up the machine when bench testing of the individual units is complete.
29/01/2012 update from Delwyn Holroyd
Restoration work on the WITCH recommenced late last year after a rather longer delay than anticipated...
With the relay side of the machine mostly operational, the next job was the arithmetic rack containing all the valve electronics. However first we needed a working HT power supply for the valves and this has proven quite challenging.
The WITCH HT supply is in two separate rack units - the rectifier unit which converts AC mains voltage to several high DC voltages, and the stabilizer unit which generates the large variety of exact DC voltages required by the machine.
At the start of the restoration in 2009, Tony F found the rectifier unit to be in quite a state, with one of the chokes having melted and leaked pitch over everything else, and a variety of overheated and burnt out components. Due to the extent of the mess he entirely stripped and rebuilt the unit, re-using as many of the original components as possible.
Upon resuming work last year, Johan and I discovered the oil filled transformer in the rectifier unit had started to leak. Our options were to replace the transformer with a modern one, remove all the oil and have the case opened, cleaned, re-sealed and re-filled, or attempt to stop the leak with an externally applied filler. We chose the latter since it didn't preclude going back to either of the first two options, and the second option presented significant risks both to health due to the likely poisonous nature of the oil, and of irreversible damage to the transformer. We quickly discovered there were multiple leaks, most probably due to fissures opening up in the original soldered joints. Re-soldering wasn't an option without first completely removing the highly flammable oil, and we eventually used an oil resistant Araldite to seal all the soldered joints. This has been 99% successful, but a couple of areas still require another coat.
The next job was to diagnose a suspected fault in the stabiliser unit with one of the valves glowing cherry red without any load on the supply - an indication of the valve being over-stressed. This was traced back to a failed metal rectifier in the rectifier unit. This unit contains quite a tour of rectifier technology, with other metal rectifiers already having been replaced with more modern components throughout the working life of the machine.
The problem with replacing old components with modern equivalents is that the modern ones tend to be much more efficient, which can have knock-on effects elsewhere in the circuit. To get around this resistors have been added to compensate, but final sizing of these can only be done when we know the load the machine will draw from the supply. We also added a load resistor to one of the supply outputs that is often unloaded in the normal operation of the machine, both to safely discharge the filter capacitors and to prevent the voltage across them rising to a level exceeding their rating, this latter problem being compounded by the more efficient modern rectifiers.
The two power supply units have now been bench tested without issue into dummy loads for a number of hours over several weeks. When we looked at the stability of the outputs in more detail we discovered a small wander on one of the outputs, which will probably prove to be due to a dirty connection or pot.
With working supplies we moved on to the Pulse Generator unit. Tony F had already cleaned this unit and done a lot of component testing, so after some more checks and safety tests on the heater transformer the next step was to apply power. We wired a spare connector to enable the unit to be connected to the main power supply on the bench. This has been nicknamed the "Connector of Doom" due to the 13 high voltages between -385V and +370V DC it carries, and AC mains as well!
The pulse generator (Gallery Image 8) as its name implies generates various pulse trains needed to step the Dekatron tubes and drive other electronics in the machine. We had expected the timing may have drifted due to component values changing with age - many of them being over 60 years old - but to our surprise the oscillator is running and generating 1.5ms pulses with a cycle time of 4.5ms, just as it is supposed to. However the spikes on the bottom waveform on the scope (see picture) indicate a problem we will need to investigate. This is the first time it's been powered since the mid-1970s when the machine was on show at Birmingham Museum of Science and Industry.
Last week Eddie cleaned and adjusted several relays which had been causing intermittent problems, so the relay side of the machine is now running reliably again.