🎛️ Control Systems

What you’ll learn

1. The Toaster and the ThermostatDistinguish open-loop from closed-loop control and state the bargain feedback offers.A toaster executes a plan; a thermostat pursues a goal by measuring the difference between the world and its intention. Open loop is cheap, needs no sensor, and can never oscillate — but has no defence against disturbances or model error, and fails silently and confidently. Closed loop buys performance the plant never had, paying with information instead of engineering, and the universe charges for it in stability.
2. Anatomy of a LoopIdentify the five blocks of any feedback loop and the meaning of the loop's sign.Reference, plant, sensor, comparator and controller appear in every feedback system ever built, and the comparator is the interesting one — a control system acts on the difference between the world and its intention. Negative feedback opposes deviation and is the stabilising kind; positive feedback reinforces it. Crucially, delay alone can turn a negative loop into a positive one at some frequency without anyone changing a sign.
3. Feedback Is Ancient, and It BitesTrace feedback from Watt's governor through the hunting problem to Maxwell's founding theory.Watt designed his centrifugal governor in 1788 at Boulton's suggestion — a closed loop made of iron with no intelligence in it. As engines got faster it over-compensated and oscillated between opening and closing the valve, and the Victorians' added weights and springs helped unpredictably because they were treating parts. Maxwell's 'On Governors' (Proc. Roy. Soc. 16, 270–283, 1868) showed stability lives in the roots of the loop's characteristic equation; Routh answered his open problem in 1876.
4. Proportional Control, and the Offset It Cannot FixExplain proportional control and prove why it cannot eliminate steady-state error.u = K_p·e is intuitive and is most of what many loops need, with gain as the only knob. But when a plant has a constant load, a non-zero output mathematically requires a non-zero error — the controller can only hold the furnace on by remaining wrong, so the room settles permanently short of setpoint. Raising K_p shrinks the droop without ever reaching zero, and oscillation arrives first. The fix must be a term built from the error's past.
5. Integral: Killing the Offset, Buying a ProblemShow how the integral term eliminates offset and what it costs in lag and windup.Integrating the error gives a term that keeps its accumulated value when the error reaches zero — so it hunts for whatever constant output the load requires and then holds it, learning the load without ever being told loads exist. The costs are real: the term is slow and backward-looking, so it eats stability margin, and integral windup lets accumulation pile up while the actuator is saturated, causing large overshoots. Anti-windup is inelegant and universal.
6. Derivative: Anticipation, and Its EnemyExplain derivative action, its origin in Minorsky's ship steering, and why noise usually kills it.Derivative acts on the rate of change of error, adding damping and permitting higher gain — it is the anticipation you use when parking a car. Minorsky derived the PID law in 1922 by watching a helmsman, and on the USS New Mexico PI control gave ±2° of sustained yaw while adding D gave ±1/6°, better than most helmsmen. Yet D appears in only ~25% of deployed controllers, because differentiation amplifies high-frequency sensor noise brutally.
7. Reading a Step ResponseRead a step response and choose damping against a specification rather than a formula.Since stability is a property of the loop rather than the schematic, you kick the system and watch: rise time, overshoot, settling time and steady-state error are the loop's fingerprint, and a non-zero offset immediately indicts the integral term. Underdamped rings, overdamped crawls, critically damped is the fastest with no overshoot — but 'critically damped is optimal' is a myth, since a few percent of overshoot is free in most loops and catastrophic in an elevator or an insulin pump.
8. Why Systems OscillateExplain instability as delay plus gain, and define gain and phase margin.A hotel shower oscillates because every correction is exactly right for a situation that existed ten seconds ago: delay plus enough gain to act decisively on stale information, and nothing is broken. A sine shifted half a period is its own negative, so at the frequency where lag reaches 180° a correction reinforces the error — and whether the loop grows or decays depends on whether its gain there exceeds 1. Gain and phase margin measure the room you have; real controllers are deliberately detuned.
9. Beyond PIDSurvey feedforward, cascade, state-space and MPC, and know when each is the answer.PID is entirely reactive — every term is computed from an error, which is something that has already gone wrong — so feedforward acts on a measured disturbance before it reaches the output, and pairs with feedback to clean up model error. Cascade nests a fast inner loop to shorten delay for the fast disturbances; state-space handles coupled multivariable plants, with Kalman's filter estimating the state you cannot measure; MPC optimises a sequence of future moves and handles constraints natively.
10. What Feedback Costs, and Why We Pay ItRecognise feedback loops beyond engineering and restate the course's central bargain.Thermoregulation, pupillary response and glucose control are feedback loops, which is why an artificial pancreas is literally a control-engineering problem complete with windup concerns. Predator–prey cycles hunt like governors, and central banks manage phase margin without using the words. Every idea in the course — integral, derivative, tuning, margins, feedforward, MPC — is a payment on one debt: feedback lets a machine be better than its parts, and the price is always stability.

Questions this course answers

Grounded in trusted sources

• J. C. Maxwell, 'On Governors', Proceedings of the Royal Society of London 16, 270–283 (1868)
• Wikipedia — PID controller (Minorsky 1922; USS New Mexico PI ±2° vs PID ±1/6°; pneumatic PID from ~1932; Ziegler–Nichols 1940s; derivative in ~25% of deployed controllers; integral windup): https://en.wikipedia.org/wiki/PID_controller
• Wikipedia — Centrifugal governor (Watt's governor 1788 at Boulton's suggestion; hunting and over-compensation; Porter, Hartnell): https://en.wikipedia.org/wiki/Centrifugal_governor
• Karl J. Åström and Richard M. Murray, Feedback Systems: An Introduction for Scientists and Engineers (Princeton University Press)
• Otto Mayr, The Origins of Feedback Control (MIT Press, 1970)
• H. Nyquist, 'Regeneration Theory', Bell System Technical Journal 11, 126–147 (1932)
• Wikipedia — Control theory; Negative feedback; Routh–Hurwitz stability criterion; Nyquist stability criterion; Bode plot
• Wikipedia — Step response; Damping ratio; Ziegler–Nichols method

Every Wunder lesson is built from real, reputable sources — never invented.

Related Science courses