Autonomous Locomotives

On a lesser part of the internet we were discussing autonomous ships and aircraft. The technology of course is still not there yet. Arguably, no real self-thinking device of this sort has yet been built.

But what about an automatic steam locomotive? Here the problem of steering is non-existant. You could imagine a locomotive that could have worked. You would fill it with coal, tie on the empty boxcar and let her pull the empties the thirty miles up to the quarry (or whatever). Some sort of device would detect switches and a clockwork gadget would govern speed.

**Was such a device ever built? If not, why not? **

Steam, by its nature, will make this a bit harder.

If you’ll accept electric locomotion, these are common - for example, numerous airports have automatic trains that shuttle passengers between and around terminals.

Surprisingly enough, I operated a steam traction engine a few years ago. That is rather a more primitive form of the technology, and I am doubtful it could be safely automated. The steam valve, while cleverly designed so as to neutralize flow and pressure forces from the steam, required near constant fiddling, and with the annoying complication that you needed to anticipate what was coming up, like an incline, or a decline. The engine also could not be allowed to ‘coast’ downhill at over 2 MPH, or the mechanism would not generate sufficient drag force to prevent a ‘runaway’. The situation would be deemed dire at 4MPH downhill, and the engine would be out of control and a menace to life and limb at 6 MPH.

The fault control logic, vision assisted auto-nav, and AI capable of accurately anticipating future motoring conditions would seem to be an intractable problem.

Stoking/refueling the fire and regulating the airflow in anticipation of the expected loading conditions on the engine would be a further complex and time dependent variable.

There are also some safety concerns. Regulating the water level in the water jacket would seem to be a simple task for a float switch, but it is not. The replacement water is ‘mysteriously’ induced into the device via a voodoo type widget that somehow jets steam through itself to entrain water and blast it into the tank. This gizmo has several valves that have to be set by ‘feel’. Hell if I know how that thing works, and it is evident to me after listening to all the old-timers descriptions of how it works, they don’t have a clue either. The water level is critical, if it falls too low, the steel making up the engine will overheat above the firebox, fracture, and explode. Putting more water in needs to be timed to an increase in the fire in the firebox, above and beyond the anticipated load to be placed on the engine.

As impressed as I am with Watson on Jeopardy, I would be terrified to turn him loose on a steam engine.

For the programing of a steam engine would be very complex. Stoking (adding the proper amount of coal at the proper place) a coal fire on a HRT boiler almost takes the mark 1 mod 0 eye ball. Not all coal is the same. Seeing if a clinker is forming would be hard to sense. And clearing it by auto would requite complex mechanical equipment.

Controling water level with an inductor (I am trying to remember back 40 years ago for proper clasification) will be tricky. But if we are spending massive amounts of money to automate it could be solved by adding a feed water pump and an automatic water level controll system.

Working the Johnson bar (direction control and linking the engine in or out) would be tricky. On trains I believe it was done by sound and feel.

And now the major problem.

Safety!!! No high pressure boiler (over 15 psi) should be operated with out an attendant. If you have ever seen the effects of a boiler failure it is big.

A computer operated steam engine would be more safe, efficient, and reliable, than one operated by a human, if such a machine was ever put into use. It would not fall asleep, have a heart attack, or get distracted by the chicks bathing in the water tower at Petticoat Junction. The computers in cars currently control much more complex systems with failure rates much lower than those of human operated steam engines.

However, a computer controlled steam engine would be far more complex and sophisticated than human controlled ones. The sensing of heat, noise, and vibration would have to be incorporated into the steam engine hardware instead of being considered an intrinsic feature of the computer, where those are already intrinsic features of humans.

If there are practical, productive, steam engines now, I’m sure they are already computer controlled.

I could not disagree more. Today most stationary boilers are fully automatic, and they do blow up sometine when the operators does not pay attention. Mechanical equipment will fail sooner or later. And if no operator is around it can be extreamly dangerous, and do a lot of damage. In normal operation a boiler will not blow with only one failure. It takes multiply failures, depending on the plant, from 4 to 10 failures. But every now and then one blows and takes down a wall or two or a building.

There are no practical, productive, steam engines being made today. The efficiencys are too low, Diesel is king today.

Mechanical systems (including electronics) can be made redundant more readily and economically than human systems. Machine failure is the equivaent case of a person not paying attention, and is much less likely for modern control systems. No human could launch or land a space shuttle unaided, but the only failures have been due to construction, not the control systems.

Which is why they are not automated.

They are not really the same. Machiery will fail, that is why it needs someone watching it to correct the failures. Your are right no human could launch a space shuttle, but humans are necessary to oversee the the computers. Huston is not a unmaned center during takeoffs.

The link below is a story about a plant with mechanical failures that happened and a operator who did not double check his systems. The plant had two scotch marine boilers. Both had feed water pump failures. Both had low feed water pressure alarm failures. Both had low water cutout failures. Both had low water limit failures. One had low water alarm failures. The operator was working in another part of the plant because his boiler room was fully automated with all the safeties and cutouts and he thought it was ok to operate unmaned. Mid shife he went into the plant to get some tools heard the low water on one boiler. Found he had no feed water pressure. So he started the feed water pump. The other boiler that the alarm was not working on water level was below the crown sheet (Fire on one side water normally on other. The crown sheet was red hot. When the water hit the red hot metal it became steam very rapidly over pressurizing the weekened metal and the boiler went flying.

The OP is asking about an unmaned steam engine and I say it is not safe.

But you keep providing evidence that manned steam engines are unsafe, not unmanned ones.

Maybe on one of those trams in a US airport like Toronto.

This was a fully automatic plant. An unmanned plant would be fully automatic. And a unmanned plant would be unsafe because of the potiential for failure. An unmanned plant is about as safe as a poorly manned plant because both would depend on the redundant safeties and past experience tells me there can be multiply failures at the same time.

And If you have a unmanned engine that is diesel or electric that goes through a failure and the safeties do not work the damage is less dangerous. If a diesel engine overspeeds and blows or a electic engine motor over amps and burns anyone that is around the engine with the failure is less likely to get hit by the flying parts that if they are next to a steam engine that the boiler blows.

What I am saying is I do not believe you can make a safe unmaned steam engine. Do not get me wrong, it not that I think automation is unsafe. But risk benifit needs to be factored. And the risk is high.

Remember steam engines are fire tube boilers not water tube boilers. There is a whole lot more water that will flash into steam fith a boiler failure.

Several major cities (or, in the case of Singapore, an entire country) have automated electric train lines, too- things like the Docklands Light Rail in London, for example.

As for the OP: I’m absolutely positive some enterprising Victorian Engineer doubtless attempted something like what you describe (some of the stuff people came up with back then is fascinating!), but by the time it became practical, technology had moved beyond steam engines to diesel/electric ones, I’d say.

I rather feel that the supporters of full automation show a touching faith in the abilities of computer control systems that is not born out in reality.

As has been noted, no human could reasonably launch a Space Shuttle, but there is also no possible way that an automatic system would be allowed full authority to launch one either. (Nitpick, the SS launch control is done at the Cape, control of a mission is handed over to Houston at the point the vehicle clears the launch tower.)

The problem with automatic control systems is that they are brittle. If you assume that all the sensors and actuators are working perfectly, you can make a reasonable fist of making a safe design. But you can’t assume this. Any sensor or actuator can fail. Multiple redundant systems only gets you so far. First up, you can’t assume that a sensor or actuator fails to an inactive state. A water level sensor may develop a fault that causes it to read 20% high, or to stick at various times. An actuator may fail on, not off. It may fail with an inability to work across its full range, or may stick, or respond slowly etc. Redundant sensors and controllers can suffer from common mode failures, especially if they are of identical design.

Faced with such issues even multiply redundant systems have hard to manage problems. The control software will have to condition the inputs and the conditioning may have to decide if a sensor had failed, and in the face of some failures it may not be able to reasonably decide which sensor is the bad one. In the face of more than one failure, of perhaps failure of both a sensor and an actuator, the software may actually decide that the good sensor is the one with the fault.

Perhaps the best example of current systems that meet some of these needs are aircraft control systems. Airbus especially have bet on large amounts of software control. And they have lost planes due to flaws in the design of this software. And that was despite having human pilots overseeing things. Sometimes the humans trusted the software too much, othertimes the software designers actually created a system where the human pilots were not allowed to override the software.

There is no software engineering regime that can cope with the level of complexity here. There can exist instabilities, corner cases, and just outright wrong actions inherent in the best crafted systems that will never be detected until a system fails for real.

Faced with a complex set of faults, you need human judgement. You are not going to be able to construct any software system with current computer capabilities and technology that can problem solve a messy and complex sensor fault. You are certainly not going to trust a machine as dangerous as a steam locomotive to such a controller.

This doesn’t sound like a formidable problem. You put in a bunch of redundant and independently controlled sensors, of several types. When the indications of one disagrees with all the others, you assume it’s bad and ignore what it says. When more than one disagree, the system is deemed unreliable and shut down.

And yet we trust them to humans, who quite regularly make mistakes that automatic systems never would.

As you point out, there have been some notable failures of complex automatic control systems on aircraft. Yet these continue to be used in increasing numbers - in large part because it’s easy to show that they do a better job than humans.

Indeed, autopilots consistently give excellent results - and they and continually improve. It’s now quite normal for the takeoff, flight, pattern, approach and landing to be done entirely “hands off” - which is surely as complex a set of tasks as running a steam engine. In the “cockpit of the future” it’s widely held that the pilot will actually be a backup system, with few to no specific duties during a routine flight (except probably to make occasional reassuring announcements on the PA system).

Forgive me; my unclear writing has misled you all.

**Why were no old-days clockwork steampunk autonomous steam locomotives ever built? ** Not retro steam engines with modern computer controls.

Because in those days, automation cost more than labor and was less reliable. In todays world, (first world) labor costs more than automation and is in many ways less reliable.

[Diversion which I’ll connect at the end]

In aviation today the big push is on pilotless aircraft. Not simple drones or overgrown remote-controlled models, but largely autonomous fighters & bombers. Google [UCAV], [UCAS], or [UCLASS] for more. These are research projects, but the expectation is that they will produce real functional front-line equipment in 20-30 years.

As **Francis Vaugn **said the challenge is not normal operations; computers can do a decent job of that today. The challenge is entirely building in enough of the right kind of redundancy & degraded modes for the computer to see and understand the true state of the machine, malfunctions and all.

In aviation today, we have layers of equipment & instruments ranging from real fancy to real basic. The critical thing the humans provide is the ability to effectively diagnose failures in real time and cut over to at least partial reliance on backup systems. In the face of ambiguity, the humans don’t always get it right immediately, but they can adjust their plan as the situation develops.

The jet I fly has (no exageration) 120 idiot lights, 40 aural warnings, and umpteen indicators. When half that stuff lights up at once, sorting out what’s important and what’s not is a challenging problem for AI. A simple fault diagnosis tree is too rigid.

Eventually the state of the art in AI will get to where this is doable enough for sorta-expendable military aircraft. I doubt I’ll live to see it for machines where the customers riding on board want near-total assurance of arriving alive. But eventually airline pilots will go the way of elevator operators and subway drivers.

A big difference between trains and airplanes is the impact of weather. If we could wave our magic wand and make weather disappear, about 90% of the headaches & instability in the current aviation / airline system would disappear. Except in heavy snow country, trains don’t much care about weather.

Remote control locomotives are commonplace now. Mostly-autonomous ones would be comparatively trivial versus the mostly-automonous airplanes being built now. Societal concerns about having that much concentrated power rolling through towns controlled only by “blind” machines will hold back development far more than technical hurdles will.

In the day of steam AI was very limited. Just the feed water regulator would be a major problem. Adding water by use of the inductor would be a major problem to complex. To automate I think would have require changing over to a feed water pump. Ans with the rapid changes in steaming rates it would require a multi element feed water regulator. It was not until the late 60 or early 70’s a two element feed water regulator was delveloped. And by then the age of steam for trains was over.

Feeding the rate of coal could be easily automated. But the spreading of the coal on the fire to keep a consistant even fire would take a complex AI system with a powerful computer and sensors. A fireman on a train has to stay infrom of the engineer at all times.

Sitting idle in the station the fireman keeps the water level lower in the boiler and the fires at a low rate. Before the engineer grabs the throttle he lets the level drop a little begins increasing the amout of coal going into the boiler and opens his draft. As the engineer opens the throttle he has to watch the water level to make sure that there is not too much swell (with increased steam flow the level will go up). As steam is used he will need to check his fire to be ssure it is stoked properly for a good fire. And as the water level drops add more water keeping the level at a proper level inresponce to the throttle and being ready for the engineer to incrrease the throttle as they clear the station.

Then after they are up to speed maintaining water level and firing rate. If they come to a hill he will need to lead the steam demand going up, and cut fires and besure the water level is a little higher before the engineer cutss the throttle coming down the other side. He does all this while helping the engineer maintain the other systems on the engine.

Quoth Francis Vaughan:

Quite the contrary: nothing other than a fully automated system would ever be allowed control of the Shuttle on launch.

I don’t think we disagree. I regard most of the launch as the lead up to the point about 8 seconds before the SRBs light up. At that point full authority is passed to the automatic systems. Up until then the entire launch system passes through humans that have authority to abort or continue the launch. Once the vehicle is in flight it does indeed run almost entirely under automatic control. Almost. The pilot has the ability to override the control program to initiate various abort modes (which will largely continue under automatic control.) The point is that decisions that affect ultimate safety of the craft are kept under human control as much as possible. On the ground the range safety officer is the guy with responsibility and authority to blow the shuttle up if needed. During flight the mission control guys in the pit watch the shuttle and will make abort or continue decisions in the face of possible failures. These decisions re not delegated to the flight control software. Again, it is the same story. If everything is going well the automatic systems work well, and are left to do their work. If something is going wrong, or just appears to be going wrong, the critical decision is still made by a human being. A good example is the the decision to go for abort to orbit in the face of a main engine failure. This is not made by the flight control software. The couple of times it has happened it was made by the flight controllers in Houston. They inform the pilot who commands the flight control software to switch to that profile.

Which makes a critical assumption. That you can safely shut it down in the face of sensors or actuators that have failed enough to no longer be useful. Indeed it assumes that shutting down is an option at all. A steam locomotive does not just “shut down”. There is no power switch. Or, since we we mentioning the Space Shuttle, you don’t get to decide that half way though powered flight that the shuttle is no longer reliable, just turn it off, just pack up and go home. Your ability to manage the system in a safe manner includes the ability to bring it back to a safe quiescent state. A runaway steam locomotive with a jammed shut water feed can’t be just “shut down”.