Control Systems

3.     Control Systems

‘Control Systems’ sounds as though this could be a bit complicated – and believe me, they can be. But they can also be very simple. A very simple control system is often called just a controller. A controller, or control system is used to keep something doing what you want it to do, rather than let it do what it wants. So a dog lead, for example, is a simple kind of controller. A cruise control on a car is a more complex control system. Control systems are used all over the place in these automated days: chemical control processes, aircraft and ship auto pilots, automatic machine tools, satellite-tracking antennas, electronic test systems, power plants and in robotics. But there are also simple controllers in use in every household – thermostats.

Thermostats

Thermostats are used in central heating systems, in fridges, deep freezers, electric ovens, immersion heaters, fan heaters, tumble driers and sometimes in greenhouses. The raison d’être of a thermostat is to keep something at (or around) a constant temperature.  Sometimes the temperature is pre fixed, but more commonly it is selectable by the user (that’s you).

Lets look at how a thermostat works in a hot water central heating system. If the room temperature falls below the setting, the thermostat ‘comes on’. An electrical contact is made which turns on the pump, or a valve, and hot water is pumped into the radiator causing the room temperature to rise. If the room gets too hot, the thermostat turns off. The contact is broken, the hot water stops pumping, and the radiator cools, and eventually so does the room.

Now it is impossible to keep a room (or anything else) at exactly a constant temperature. If the thermostat were to turn on and off at exactly the same temperature, then the poor heating system would be constantly turning on and off like a mad thing, always seeking perfection, but never finding it. So a deliberate gap is made between the turn on and off temperatures. This is known as hysteresis by technical types, it is a bit like the play you get in a steering wheel, only this time it is deliberate. So if, for example, you set the room temperature at 20C, the thermostat may come on at 19C, and go off at 21C.

A thermostat is a classic example of negative feedback. Why? Well, what we are doing here is to try and keep the room temperature within a given distance from the setting. Looking at it another way, it is always trying to reduce the difference between the actual temperature and the setting.

I think it is time for another little picture:

Because the thermostat always tries to reduce the difference between the actual temperature and the setting, this is negative feedback. And as we know by now, negative feedback makes things steady, or stable. Which is exactly what is going on here. The room should be kept at a nice steady temperature.

In practice this is not always quite as steady as we would like for a number of reasons. For one thing, radiators take quite a long time to heat up and cool down. For another, the thermostat only measures the temperature in one place, and that may not be typical of the whole room (or the whole house). And finally, outside weather may be changing which is tending to heat up or cool down the room. The thermostat does not ‘know’ what the weather is doing outside, it is a fairly simple, somewhat ‘dumb’ control system. Modern smart control systems are available which take into account the outside temperature, and what is has been doing for the last day or so, in order to adjust the room temperature more accurately. Needless to say, the building industry (at least in the UK) has not got around to fitting these much in houses, though they are used increasingly in large buildings.

The thermostat and heating system that I have described so far is an example of a closed loop control system. The loop is closed because the temperature is fed back to the controller, forming a closed loop as in the diagram.

Open loop control systems, in contrast, are very basic controls, which do not have any feedback of the effect of the control. This would be like driving your car with your eyes shut – there is not much feedback, and it is not a practice to be recommended. When you open your eyes, you are completing a closed loop control system, because your vision gives you feedback as to the effects of your steering and accelerator control. Similarly, in aircraft blind navigation, there may be no feedback as to position, which is computed using dead reckoning. So this is also an open loop control system. Until you come to land that is, then hopefully you can see the runway.

A general open loop system has several key components:

General Component	Thermostat example
Input	Temperature setting
The process being controlled	The heating
Output	Room temperature
A sensor or sensors	Usually a bi-metallic strip
The controller	The switch contacts

A bimetallic strip sounds technical, but is just two different metal strips joined together. When the temperature rises, the different metals expand at different rates. This makes the strip bend and so make an electrical contact.

Servomechanisms

Thermostats are sometimes regarded as a kind of servomechanism. A servomechanism is a closed loop control system which usually controls something mechanical, though it can be used to describe hydraulic, electrical or optical systems.

An example of a servomechanism is the power steering system in a car. The steering wheel controls the direction of the front wheels, through a mechanical/ hydraulic system. If the wheels turn away from the desired direction, the servomechanism automatically brings the wheels back into line.

In fact, one Joseph Farcot first used the word ‘servo’ in 1873 to describe steam powered steering systems. The first servomechanisms were actually used in gun aiming and firing devices, and marine navigation – once again, the pressures of war speeding up the development of new technology. Nowadays, the term is also used to apply to devices for controlling lots of different things, not just temperature or direction. Other things it may control include speed, position, orientation (as in a satellite say) or size.

One of the characteristics of servomechanisms is that they can allow high-powered devices to be controlled by low power signals. So, for example, in a car power steering system, the guts that actually turns the wheels is much more powerful than the driver (well, it is in my case, I don’t know about you). The ratio between the power of the control signal and that of the device controlled can be huge – as for instance in a control system for a super tanker or cruise ship. One of the memories I cherish from a Mediterranean cruise (apart from the food and the weather that is) is watching the captain precisely position a 100,000 ton vessel with a remote control device that looked not much bigger than a laptop.

Big or small, though, all servomechanisms rely on feedback to tell the control system the results of its actions, and to allow it to adjust accordingly.

Another example of a large servomechanism is the tracking antenna of a communications satellite. These are pretty hefty objects, often being 10 metres or more in diameter. The trick is to keep the antenna aimed directly at the satellite to get the best signal. This can be done by comparing the signals from two adjacent receivers, with two different antennas slightly apart. Any difference in the signals is used to move the antenna to reduce the difference, using a servomotor. This negative feedback enables the antenna to be aimed very accurately.

The Scanning Tunnelling Microscope

An example of control systems at the other end of the scale is the scanning tunnelling microscope. These wondrous devices are used to probe matter at an atomic scale. In effect, they ‘feel’ their way over a surface one atom at a time. They do this by making use of something called the quantum tunnelling effect.  They control a probe consisting of a very sharp glass point positioned a tiny distance over the surface. A voltage is applied to the probe, and according to the mysteries of quantum physics, a small current can ‘tunnel’ across the gap to the surface. It is called tunnelling because the electrons tunnel through a voltage barrier that normally prevents them going through it. The current varies critically with the distance from the surface, so it can be used in a feedback control system to keep it a precise distance away as the probe scans along.

Complex Control Systems

Control systems in the early days were pretty much a matter of trial and error, with an engineer or ‘inventor’ playing about until he got it about right. This worked OK for a long time, but around 1930 things started heating up, and control systems needed to be smarter and faster to control things that were starting to move faster. Then the mathematicians got out their pencils and wrote lots of equations until the process of designing control systems became more, well, controlled and less messy.

Control system theory has got pretty complicated these days, so I am just going to give a brief idea of how these things work. To be brutally honest, most of advanced control theory is beyond my ken anyway.

I have already pointed out that a simple on/off thermostat is not going to be all that good at controlling the temperature of a room, because there is a delay or lag between turning on the pump and heating the room. Other problems arise with control systems when they have to control things that are moving fast (like an aeroplane, or worse, a rocket). Control systems can overshoot, that is, they move something too far, undershoot (not far enough) or oscillate (they can’t make up their mind).

Smart control systems try to overcome these problems in a number of ways.

A proportional controller attempts to perform better than the On-Off type by applying graduated control. For example, in an electric oven, a control could apply power to the heater in proportion to the difference in temperature between the oven and the user selected temperature.

Other systems try and predict the future behaviour of the system so they can actually anticipate changes and so make the control more efficient.

Smart control systems can use some model of the system they are controlling to adjust the control they apply to get the desired result. So if, for instance in a car cruise control, there is a known lag in the speed of the car responding to an increase in fuel injection, this can be taken account by reducing the fuel before the measured speed matches the desired speed.

PID (proportional-integral-derivative) controllers take into account the past, present, and predicted future state of a system. As previously mentioned, smart house heating systems take into account the temperature typically over a 24 hour time period.

Adaptive control systems are designed for situations in a very unpredictable environment, and they adjust the control method according to experience.

All control systems have to strike a balance between speed and stability. If you make a control system respond faster, it tends to be more unstable and prone to overshoot. Think about driving a car. As you drive faster, you are more likely to lose control. Smart control systems however, can be both fast and stable.

 

Robotics

Smart robots have been a long time coming. They have been around in Science fiction for decades, so how come we can’t buy them in the shops? The answer, it turns out, is common sense. Common sense is easy, because we all have it, even the most intellectually challenged. But robots don’t. It is the chief irony of artificial intelligence that the things that seem hard to us (like chess, or large sums) are easy for computers, whereas the things that seem easy to us (like tying a shoelace) are incredibly hard for computer controlled robots. The trouble with common sense is that it isn’t something that is fed into us, it is something we acquire. And robots are not good at that, not good at all. In fact, we probably won’t get really smart robots until they are good at it. Which could be a long time coming. That is not to say that there has not been considerable progress. The latest walking dancing robots from Japan have come a long way in robot mobility, and even robot toys are getting good at natural looking movement.

And robots are getting pretty good at handling delicate objects – which is actually quite tricky, and early robots were very ham fisted in this respect. Now you can get a robotic hand capable of performing the delicate task of picking up and holding an egg without breaking it. A tactile sensor on the gripping mechanism sends information to the robot's control computer about the pressure the robotic hand exerts; it then instructs the hand to loosen, tighten, or maintain the gripping force. This feedback loop enables the robotic hand to stay in between the two extremes of dropping and crushing the egg.

Robots like those we see in sci fi films have an awful lot to learn. They have to be able to see and hear, and what is much harder, to make sense of what they see and hear. They have to be able to move in a wide variety of environments, avoid obstacles, pick up and handle a variety of objects, and not stop because their battery has run out. These problems are why most successful robots are very specialised, like in car assembly lines. They are static, which means they can be mains powered, and don’t have to move around. They only perform a limited number of operations which are pre-programmed. Mind you, they are getting pretty good at doing these things. There is a brilliant demonstration in The Power House Museum in Sydney which has a small robot doing various dance movements to different music – including some heavy head banging. It looks life like – but it isn’t really.

The original approach to robotics was to assume that all the information from various sensors – sound, vision, touch, position, would be fed into one central processor which would analyse all the information and issue commands to control movement. This hasn’t worked out too well, partly because of the sheer complexity of the problems. But it also appears that there are fundamental reasons why this approach is not successful. Animals make use of a lot of distributed processing and decision making rather than keeping it all in one place. The human vision system does far more than simply reporting ‘what it sees’ to the brain; it also performs a lot of processing, as does our hearing system.

These considerations have led to new approaches to robotics, pioneered at Carnegie Mellon University, and discussed by Kevin Kelly in his book ‘Out of Control’. The title refers to the motto of the robotics workers at Carnegie Mellon – ‘Small, fast and out of control’. The idea is to try and produce robots that work more like animals (or insects) – small, fast, and not controlled by a central processor, but with several independent processors controlling different movements. It turns out that by allowing each ‘limb’ of a small robot to be controlled very simply by its own processor, a movement arises which is akin to the way insects move.

This is an example of a general phenomenon that is now being recognised as very important in many areas of human understanding, and is known as emergent behaviour. The point is that the behaviour arises out of a system following some often very simple rules; the strange thing is that the behaviour that arises can be very complex – far more so than one would expect from the simplicity of the rules. There is one somewhat unorthodox scientist (Stephen Wolfram) who has written a massive book called ‘A New Kind of Science’ which tries to explain how the whole way the world works as an expression of emergent behaviour.

On a more relevant level, Steve Grand in his book ‘Creation’ describes how his friend (and an old colleague of mine) Owen Holland has persuaded a collection of robots to tidy up a scattered collection of frisbees by following a very simple set of rules. The rules merely tell them to turn and move away if they come up against something bigger than one frisbee. This has the effect of getting them to move Frisbees around until they stack up into neat piles.

 The point is that the robots are not told to stack anything up – this behaviour just emerges. This kind of behaviour is often compared with insect colonies such as bees or ants, where complex structures can be built and maintained by a community of insects all following simple rules.

There have been other moves towards small simple robots – military research is pursuing the idea of small information gathering agents that could be ground or airborne. These agents could talk to each other as well as to any central information gathering command and control. Small robots have a lot of advantages – they are cheap to make in volume, need low power, are hard to detect and can act in concert as a powerful agent. The new field of nano technology is also heading towards the utilisation of large numbers of very small robotic devices, though as yet this is still in its infancy.

All this has taken us quite a long way from feedback, but you can be sure that any system of robots will need feedback both to control their individual actions, and also on a wider scale to control armies of small robots.