Obsolescence of electronic components on early computer era traction

[edwin_m]

Most of the electronics used in pre-1990s signalling scheme was for non-vital functions, such as remote control and train describers. These have generally all been replaced by more modern systems as they became obsolete. The underlying interlocking used relays, that are still available.

However, some of these interlockings have suffered from degradation of the wiring - it seems some of what was used in the 70s and 80s tended to have the insulation get crumbly after a few decades. This can make it impossible to modify an interlocking unless all of it is replaced, or even trigger replacement. I assume the recent re-wiring of Gloucester was for this sort of reason.

The increasing use of communications protocols over serial links between signalling equipment, rather than hardwiring each function individually, makes it easier to replace parts of the system instead of replacing it all. The new subsystem generally emulates the protocol to talk to the older one. So now we often see relocking and recontrolling instead of full re-signalling.

 

[montyburns56]

Wasn’t there a preserved 37 that had some of its cards stolen some years ago? I remember the owners appealing for replacements as new ones were prohibitively expensive.

Yes it was one of those Class 37/9 'Slugs' that I believe had some custom made control circuits

37905 & 37906 control cards stolen

Received this tonight. Spread the word. Someone, somewhere must know something. Gents, We need your help to find out who the person (or persons) are who have stolen both sets of GEC 'control cards' from 37905 & 37906. 37905's cards were last seen two weeks ago. 37906 was last used 10th...

www.railforums.co.uk

 

[Bornin1980s]

3D printing does offer a solution for some mechanical components - to over-simplify, an existing one can be scanned and replicated. No use for electronics though...

Haven't there already been experimental 3D printed microcircuits?

As for the 91, I assumed its computer systems had been replaced in the 91/1 upgrade?

 

[Paul AC]

As for the 91, I assumed its computer systems had been replaced in the 91/1 upgrade?

No; from an electronics point of view 91/1 was a number of component and circuit changes to improve reliability, and the software had some bug fixes and improvements to the fault logging system to make fault diagnosis easier. It's still the same 8086 running at 5 MHz clock speed, so about 1 instruction per microsecond. 16- or 32-bit fixed point arithmetic, none of this modern floating point nonsense....

The same processor card and support cards (such as analogue and digital I/O, pulse generators etc.) were standard across a lot of UK and overseas fleets, probably around 30 altogether. Detailed configuration differences, and different 'application' (as we'd call it nowadays) software, with a lot of standard library code.

 

[Anstecker]

This kind of thing seems an obvious use case for AI, assuming you have at least one working system you can inspect. I know Darktrace have worked on systems to "make sense of" complicated industrial control systems (e.g. power grids) to make it possible to spot hacks, and I've seen a case study by a developer named Steve Troughton-Smith where he used AI to make sense of a design programme developed by someone who unfortunately died unexpectedly where the file formats weren't well-documented.

 

[Acathater]

This kind of thing seems an obvious use case for AI, assuming you have at least one working system you can inspect. I know Darktrace have worked on systems to "make sense of" complicated industrial control systems (e.g. power grids) to make it possible to spot hacks, and I've seen a case study by a developer named Steve Troughton-Smith where he used AI to make sense of a design programme developed by someone who unfortunately died unexpectedly where the file formats weren't well-documented.

But who is going to sign off on a safety case for a product interpreted and redesigned by an AI if it is impossible to check manually?

 

[andy33gmail]

But who is going to sign off on a safety case for a product interpreted and redesigned by an AI if it is impossible to check manually?

You’d need to generate human (developer) readable code, in some language, to then compile on your target architecture after reviewing

Possibly you could get some assurance by black box testing identical inputs to an original and replica product - either systematic “unit test” style or “monkey on a typewriter” style

 

[Belperpete]

But who is going to sign off on a safety case for a product interpreted and redesigned by an AI if it is impossible to check manually?

There is a very big difference between using AI to help understand how a system behaves, and using AI to generate a replacement. As you say, nobody is going to sign-off on a safety case for the latter.

 

[Bletchleyite]

There is a very big difference between using AI to help understand how a system behaves, and using AI to generate a replacement. As you say, nobody is going to sign-off on a safety case for the latter.

Testing. Mathematically provable code does exist and is used in places like signalling systems and aircraft control where a minor failure could cause deaths, but there isn't any need for it for most applications. If the effect of a code failure is that the train stops, that's not a significant danger, just annoying.

There are also other strategies, such as double or triple redundancy - have 2 or 3 independently developed and independently operating systems making the same decisions and if any two disagree ask the human what to do instead. That's used in aircraft fly by wire.

 

[webbfan]

IBM used to test their mainframe operating system functions with Mathematically provable code in the 1980's. Didn't stop crashes as users added their own functionality

 

[TheHovisKid]

Yes it was one of those Class 37/9 'Slugs' that I believe had some custom made control circuits

37905 & 37906 control cards stolen

Received this tonight. Spread the word. Someone, somewhere must know something. Gents, We need your help to find out who the person (or persons) are who have stolen both sets of GEC 'control cards' from 37905 & 37906. 37905's cards were last seen two weeks ago. 37906 was last used 10th...

www.railforums.co.uk

Did they turn up?

 

[Hbbz9392]

An interesting side effect of the march of technology:

A 100 year steam locomotive can be kept running,and new parts fabricated locally should they be needed as long as you have access to a skilled machinist and the right tools

A 50 yo engine which is incorpoating 1st gen integrated circuits, CPUs etc is going to be far more of a challenge,because local manufacture on a one off basis is not a viable option. So looking ahead one could visualise a situation where fully functional units from this time period could become much rarer than older steam powered and early diesel stuff

necro but if the engine is using off the shelf parts it's parts can be procured off hobbyist websites or get something 3d-printed. after all there is a steady movement and backing towards retro computers.

 

[mike57]

it's parts can be procured off hobbyist websites

My concern would be early integrated circuits, there is a lot of stuff which is just not available today, either through mainstream suppliers or the secondhand market. Passive components are not a problem, I can still buy a 100ohm 1/2 watt resistor or 100uF 16v electrolytic capacitor, it may look different, but its a direct replacement, same for 'simple' semiconductors, diodes, transistors, thyristors, there are alternatives, but once you delve into early integrated ciruits its a lot less easy, early memory chips for examples, voltages and technologies that have not survived into the current era, I have a few memory chips from the late 70s for example which need -5 0 +5 supplies. Yes you could re-engineer, but its a level of difficulty higher than giving a skilled machinist the worn part, some measurements, and possibly a copy of a 100yo+ drawing. Currently you cannot '3d' print a chip, although I could see that technology appearing in some form at some point in the future.

 

[rf_ioliver]

and even if you have the source code, do you have the compiler (etc.) tools, and the correct hardware and operating system to run them on?

The Intel 8086 (etc.) toolset (as used on 90/91 etc. as previously mentioned) was originally designed to run on Intel's own development system hardware (Series 3, with 8" floppy disks, and Series 4 with a hard disk) but later had wrapper code added to allow it to run on MS-DOS. It still runs ok "today" from a Command Prompt or Batch File on 32-bit versions of Windows. A full code rebuild for class 91 (binary file size about 50 kbytes) used to take over 20 minutes in the late 1980s; longer if using floppy disks (temp files written to disk as you didn't have much RAM available). It takes less than 10 seconds on a modern PC.


There's a similar problem with documentation (I've used IBM Displaywrite, then WordPerfect) and CAD systems for drawings, printed circuit boards, ......

There is another issue now that you brought up compilers etc.and that is the reliability of the tools themselves. It used to be a joke that the most reliable piece of software is the compiler itself - if that breaks nothing works.

Going back to the 8080, 8086 debate - these processors are "easy" to work with - their timing characteristics and behaviours are well known. If you compare that with the modern stuff, well, inside every modern Intel CPU is the System management engine, which itself runs on Minix. As for caching and timing characteristics of moden CPUs, that's an area that causes me headaches.

For most critical systems you rarely (if ever) need wonderful stuff like AVX2 or whatever additional instruction sets Intel and AMD have added to their systems. If you do, then....TDK...panic?

 

[Hbbz9392]

My concern would be early integrated circuits, there is a lot of stuff which is just not available today, either through mainstream suppliers or the secondhand market. Passive components are not a problem, I can still buy a 100ohm 1/2 watt resistor or 100uF 16v electrolytic capacitor, it may look different, but its a direct replacement, same for 'simple' semiconductors, diodes, transistors, thyristors, there are alternatives, but once you delve into early integrated ciruits its a lot less easy, early memory chips for examples, voltages and technologies that have not survived into the current era, I have a few memory chips from the late 70s for example which need -5 0 +5 supplies. Yes you could re-engineer, but its a level of difficulty higher than giving a skilled machinist the worn part, some measurements, and possibly a copy of a 100yo+ drawing. Currently you cannot '3d' print a chip, although I could see that technology appearing in some form at some point in the future.

Can’t you adapt those to modern technology? If originality’s not an issue I think solutions for adaptation can be devised.

 

[gazthomas]

I guess FPGA based solutions may have a place

 

[mike57]

I guess FPGA based solutions may have a place

I think the real problem area could be early analogue stuff, digital is easier to re-engineer. As an example from a few years back I replaced 6 very early digital Telmecanique PLCs dating from around 1980 (Programmable Logic Controllers). We had the original ladder diagram printout, on on very yellowed paper, and some flaky spare kit. Because the ladder logic is digital converting to a modern platform was tedious but doable, and interface ins and outs were 24v all DC which is still the industry standard. So we just chose to replicate original logic "warts and all". Had this been an early analogue sytem the task would have been far more complex. So yes there will be work around for a lot of stuff, but its the oddball one offs that will trip people up.

 

[Belperpete]

I

I think the real problem area could be early analogue stuff,

If a circuit was designed around the operational characteristics of a specific component, such as a thyristor, then you have big problems if it is no longer available, and there is no replacement with the exact same characteristics.

With digital chips, the 8086, Z80 etc were supposed to be backwards compatible with the 8080, but if I recall correctly they weren't perfect. And it was not unknown for software developers to make use of "quirks" of these chips not explained in the manuals, which can make it hard to understand what the software is doing if you are not familiar with these things.