Control systems at work

A1.3.6 took a control system apart into its components. This section puts them back together in the systems you meet every day. The skill exam questions test is the same each time: name the sensors, say what the controller decides, name the actuators, and explain how feedback keeps the system on track.

• Almost all are closed loop. Anything that must cope with a changing world (traffic, weather, people walking through a door) needs to measure what is actually happening.
• Many mix open and closed loop. A washing machine follows a fixed sequence of stages (open loop), but each stage ends when a sensor says it is done (closed loop). School heating often runs on a timetable, with a thermostat in control during those hours.
• Safety is designed in. Lifts, doors and cars have extra sensors (light curtains, door edges, emergency brakes) whose only job is to stop the system harming someone, and these override everything else.
• Most are embedded systems. The controller is a small computer built into the device to do that one job (a washing machine's, a lift's), running a dedicated program, often from ROM or flash, rather than a general-purpose computer.
• Centralised or distributed. A city can run every junction's lights from one central computer, which sees the whole network and can coordinate it, but if it fails every junction fails with it. Or each junction can have its own controller (distributed), which keeps working on its own and is simpler to extend, but sees only its own traffic unless the controllers talk to each other. Large systems usually do both: local controllers that a central system coordinates.
• Complexity varies enormously. A thermostat compares one number with another; a self-driving car fuses data from dozens of sensors and runs machine-learning models to predict what other road users will do. The loop is the same.

• Bad input: a dirty camera, a sensor in direct sunlight or a faulty reading leads to wrong decisions. Important systems use several different sensors so one can check another.
• Situations nobody planned for: rules and trained models handle what their designers expected. An autonomous car meeting something truly unusual is the classic example.
• Security: connected systems (smart homes, traffic networks, factories) can be attacked over the network, so they must be protected like any other computer.
• Responsibility: when an automatic system makes a harmful decision, who is responsible: the owner, the manufacturer or the programmer? These are ethical questions as well as technical ones.