ALearning Material
A core distinction organizes all automated control: is the system open-loop (acting blindly, without checking the result) or closed-loop (measuring the result and correcting)? This choice (whether to use feedback) is fundamental to how any automated process behaves, determining whether it can hold a target despite disturbances or just runs a fixed action and hopes.
An open-loop system acts without measuring the outcome. It applies a predetermined action and assumes the result is correct, with no feedback. A timer-run sprinkler (water for 20 minutes regardless of whether the ground is wet), a toaster on a timer, a conveyor at a fixed speed: they do their action blindly. Open-loop is simple and cheap, and fine when the relationship between action and result is reliable and disturbances are small: but it can't compensate for anything unexpected.
A closed-loop system measures the output and feeds it back to correct the action. The feedback that lets it hold a target despite disturbances. It compares the measured result to the desired setpoint (the error) and adjusts (a thermostat measuring room temperature and switching heat; a cruise control holding speed up hills):
OPEN-loop: command -> actuator -> output (no checking)
CLOSED-loop: setpoint -> [compare to measurement] -> actuator -> output -> sensor -> (back to compare)
The decisive trade-off: closed-loop rejects disturbances and holds accuracy (it sees and corrects errors) but needs a sensor and is more complex; open-loop is simpler/cheaper but is blind to disturbances and errors. Most process control is closed-loop (you must hold variables despite disturbances), while some simple timed/sequenced actions are open-loop. The disciplines: open-loop acts without feedback (simple, blind), closed-loop measures and corrects (handles disturbances, needs a sensor), and choose based on whether you must reject disturbances / hold accuracy. This open-vs-closed distinction underlies every control decision in automation.
Why it exists. Whether a system checks its result and corrects (feedback) fundamentally determines its behaviour: closed-loop control can hold a target despite disturbances and errors, while open-loop simply executes a fixed action blindly. Recognizing the distinction, and when each is appropriate (closed-loop where you must reject disturbances/ hold accuracy, open-loop for simple reliable actions), is foundational to designing any automated system.
Mental model. Open-loop is driving with your eyes closed for a few seconds. You steer based on what you remember and hope nothing changed. Closed-loop is driving with your eyes open. You constantly see where you are versus where you want to be and correct the wheel. With eyes open you handle the curve and the crosswind; with eyes closed you just hope.
Common misunderstandings.
- "Open-loop and closed-loop are about complexity." They're about feedback: open-loop has none (acts blindly); closed-loop measures the output and corrects. That's the defining difference, regardless of complexity.
- "Closed-loop is always better." Closed-loop handles disturbances but needs a sensor and adds complexity/cost; open-loop is fine (and cheaper) when the action reliably produces the result and disturbances are negligible.
- "Open-loop can still correct errors somehow." No, with no feedback it can't know there's an error; it just executes its fixed action. Only closed-loop sees and corrects errors.
Connections. This is the automation-level view of the feedback/PID control (PID topic) and regulatory control (DCS): closed-loop is feedback control; open vs closed underlies the process-type control approaches (continuous needs closed-loop regulation), the instrumentation choices (closed-loop needs sensors), and the disturbance-rejection theme throughout.
BImmediate Active Recall
QUERYWhat is the difference between an open-loop and a closed-loop system?
REVEAL
An open-loop system acts without measuring the outcome. It applies a predetermined action and assumes the result is right, with no feedback (e.g. a timer-run sprinkler). A closed-loop system measures the output and feeds it back to correct the action. Comparing the result to the setpoint (the error) and adjusting (e.g. a thermostat). The defining difference is feedback: open-loop has none; closed-loop measures and corrects.
QUERYWhat can a closed-loop system do that an open-loop one cannot?
REVEAL
Reject disturbances and hold accuracy. Because it measures the output and corrects, a closed-loop system can detect when the result deviates from the setpoint (the error) and adjust to bring it back. Holding the target despite disturbances and model errors. An open-loop system, with no feedback, is blind to disturbances and errors and just executes its fixed action regardless.
QUERYWhat is the trade-off between open-loop and closed-loop?
REVEAL
Closed-loop rejects disturbances and holds accuracy (sees and corrects errors) but needs a sensor and is more complex/costly. Open-loop is simpler and cheaper but is blind to disturbances and errors. So you use closed-loop where you must hold a target despite disturbances/maintain accuracy, and open-loop for simple actions where the result reliably follows the command and disturbances are negligible.
QUERYWhen is open-loop control appropriate?
REVEAL
When the relationship between action and result is reliable and disturbances are small/negligible, so a predetermined action dependably produces the desired result without needing to check, e.g. a timed toaster, a fixed-speed conveyor, a simple sequenced action. Its simplicity and lower cost (no sensor) make it the right choice there; you don't need feedback if nothing unexpected will throw off the result.
CConceptual Questions
Answer each in your own words in the box, then reveal the model answer to compare. These ask why, not how, and your answers are saved.
Why is the presence or absence of feedback (closed- vs open-loop) the most fundamental property determining how an automated system behaves, and what specifically can closed-loop do that open-loop fundamentally cannot?
REVEAL MODEL ANSWER
Feedback is the most fundamental property because it determines whether a system can respond to reality or only execute a plan blindly, which shapes everything about how it behaves in the face of an imperfect, changing world. An open-loop system applies a predetermined action and assumes the result follows; it never looks at the actual output, so it is structurally incapable of knowing whether it achieved its goal. This is fine only in an idealized situation where the action reliably produces the result and nothing disturbs it, but the real world has disturbances (a load on a motor, heat loss from a room, wind on a car), model inaccuracies, and drift, none of which an open-loop system can see or counter. A closed-loop system adds the one thing that changes this completely: it measures the output and compares it to the desired setpoint, computing the error, and adjusts its action to drive that error to zero. That single addition, feedback, gives it a capability open-loop fundamentally lacks: the ability to reject disturbances and hold accuracy despite the unexpected. Because it continuously checks the result, it automatically compensates for whatever it never explicitly anticipated, since any deviation shows up in the error and gets corrected. The thermostat holds the room temperature whether or not someone opens a window; cruise control holds speed up hills and down; an open-loop equivalent would just run a fixed action and let reality push the result wherever it likes, with no awareness. So the feedback distinction isn't about complexity or sophistication. It's about whether the system is aware of and responsive to its actual outcome (closed-loop) or blind to it (open-loop). That awareness is what lets automated systems achieve and maintain targets in a messy world, which is why open-vs-closed is the foundational classification underlying every control decision.
Why isn't closed-loop control simply always the better choice, and how do you decide between open- and closed-loop for a given task?
REVEAL MODEL ANSWER
Closed-loop isn't automatically better because its power, feedback, comes at a real cost, and that cost is only worth paying when the task actually needs disturbance rejection or accuracy. To close the loop you must measure the output, which means adding a sensor (and its wiring, calibration, and potential failure), comparing to a setpoint, and implementing correcting logic: more hardware, more complexity, more things that can go wrong, and more expense. If the situation doesn't demand it, all of that is wasted: where the relationship between action and result is reliable and disturbances are negligible, an open-loop action will dependably produce the right result without any of the feedback machinery, so the simpler, cheaper, more robust open-loop solution is the better engineering choice. A toaster on a timer, a fixed-speed conveyor, a sprinkler that runs for a set time. These don't need to measure anything because the open-loop action is reliable enough for the purpose, and adding sensors and feedback would only add cost and failure modes for no benefit. The decision therefore hinges on two questions: does the task require holding a target despite disturbances, and does it require accuracy that the open-loop action can't guarantee? If yes to either: the room temperature must stay steady though heat is lost, the speed must hold though loads vary, the level must be maintained though inflow changes. You need closed-loop, because only feedback can sense and correct the resulting errors. If no (the action reliably produces an acceptable result and nothing significant disturbs it) open-loop is appropriate and preferable for its simplicity and economy. So you match the approach to the requirement: closed-loop where disturbance rejection or accuracy is essential, open-loop where the action is reliable and feedback would be needless cost. This fit-to-purpose judgement, rather than 'always close the loop', is the mark of good automation design.
DPractice Problems
P1 (easy). Classify each of these as open-loop or closed-loop, and for each give the measured variable (or state that there is none) and the disturbance it cannot reject: (a) a thermostat heating a room, (b) a sprinkler on a 20-minute timer, (c) a toaster with a browning dial, (d) a car's cruise control, (e) a stepper motor moving a gantry to a commanded position.
Then pick the two that are most often misclassified and explain why.
P2 (medium). A car holds 100 km/h on the flat at a fixed throttle of 35 percent. It then climbs a hill needing 48 percent, and descends one needing 22 percent.
(a) Compute what happens to the speed in the fixed-throttle case if the relationship between throttle and steady speed is roughly linear at 2.9 km/h per percent of throttle. (b) Describe what a proportional cruise control does at each point, and compute the steady-state error with a gain of 1.5 percent of throttle per km/h of error. (c) Add integral action and say what changes, plus the one new problem it introduces on a long hill.
P3 (harder). Give three real scenarios where open-loop control is the better engineering choice than closed-loop, each for a different reason, and state the reason precisely.
Then give the design pattern that gets most of the benefit of both, with an example, and state the one question that decides between open and closed loop in a case you have not seen before.
Solutionsclick to reveal
P1.
| System | Type | Measured variable | Disturbance it cannot reject |
|---|---|---|---|
| (a) Thermostat | Closed loop | Room temperature | Very few. It rejects outside temperature, open windows, occupancy and solar gain, because it measures the thing it cares about |
| (b) Sprinkler on a timer | Open loop | None | Rain, soil moisture, a blocked nozzle, a burst pipe. It waters for 20 minutes whether the ground is parched or flooded |
| (c) Toaster with a browning dial | Open loop (in most toasters) | None: the dial sets a time, not a colour | Bread thickness, moisture content, whether the toaster is already warm from the previous slice. The second slice is always darker |
| (d) Cruise control | Closed loop | Vehicle speed | Very few on speed. It cannot reject a disturbance it does not measure, such as wheel slip on ice, because wheel speed and ground speed are then different things |
| (e) Stepper moving a gantry | Open loop | None. The controller counts steps it has issued, not steps the motor has taken | Lost steps. If the load exceeds the available torque, the motor slips, the controller does not know, and the position is wrong from then on, permanently |
The two most often misclassified: the toaster and the stepper.
The toaster gets called closed-loop because it has a dial and a mechanism that stops it. It looks like feedback: you set a level and the machine does something until it reaches it. But the dial sets a timer (in a cheap toaster) or a bimetallic strip's trip point (in a better one), and neither measures the browning, which is what the user actually wants. The give-away is the everyday symptom: the second slice comes out darker than the first at the same dial setting, because the machine is warmer and the same time now produces more browning. A closed-loop toaster would produce identical slices regardless, and the ones that do exist measure the bread's surface with an optical sensor.
The general trap: having a sensor does not make a system closed-loop; the loop is closed only if the measured variable is the variable you care about. A toaster that measures its own element temperature and holds it constant has a perfectly good feedback loop around the wrong quantity.
The stepper gets called closed-loop because it moves "to a position" and arrives accurately. It does, and there is no feedback anywhere: the controller issues N pulses, the motor advances N steps if it can, and the controller assumes it did. The accuracy comes from the mechanism, not from measurement: a stepper's step angle is a property of its pole geometry and is repeatable to a fraction of a percent, so open-loop is good enough for a very long time.
Until it is not. Exceed the available torque (a jam, a fast acceleration, a heavier load than expected) and the motor loses steps silently. Nothing reports it. The gantry is now offset by the lost amount, and it stays offset for every subsequent move, because every position is referenced to a count that is wrong. The machine carries on confidently, producing parts in the wrong place.
That silence is the defining property of open-loop control, and it is what the classification is really about: an open-loop system cannot tell you it has failed, because it is not measuring the thing that failed. This is why steppers are homed against a physical switch at start-up (re-establishing truth periodically) and why anything that must not lose position gets an encoder, which is precisely the change from open to closed loop.
P2. (a) Fixed throttle, 35 percent.
The throttle does not change, so the speed settles wherever 35 percent produces equilibrium against the new load.
- On the hill needing 48 percent for 100 km/h: the car is 13 percentage points short. At 2.9 km/h per percent, it loses 37.7 km/h, settling at about 62 km/h
- Descending, needing 22 percent: the car has 13 points too much, gaining 37.7 km/h, settling at about 138 km/h
The open-loop system does not "try" and fail: it has no concept of the target at all. It applies a fixed input and whatever speed results is the speed.
(b) Proportional control, /h.
At steady state the controller's output must equal what the hill requires, and a proportional controller produces output only in proportion to error:
On the hill (48 percent required):
- The controller's contribution above its nominal 35 percent is percentage points
- Error needed to produce that = 8.7 km/h
- The car settles at km/h
Descending (22 percent required):
- Error = 8.7 km/h in the other direction
- The car settles at 108.7 km/h
Compare: open loop gives 62 and 138 km/h; proportional gives 91.3 and 108.7. The error has fallen from 38 km/h to 8.7, a factor of 4.3, which is exactly times the plant gain: . That factor is the loop gain, and it is the whole benefit of closing the loop.
Raising reduces the error further: at the error is 2.6 km/h. It never reaches zero, because a non-zero output requires a non-zero error, and at some the loop overshoots and oscillates.
(c) Adding integral action.
The integral accumulates error over time, so as long as any error remains it keeps increasing the throttle. It stops increasing only when the error is zero, at which point the integral holds whatever value produces exactly 48 percent throttle.
The steady-state error becomes zero, on both the climb and the descent. The car holds 100 km/h. This is what integral action is for and it is the reason real cruise control is PI rather than P.
The new problem on a long hill: integral windup.
Consider a hill steep enough to require more than 100 percent throttle to hold 100 km/h, which is common in a loaded vehicle. The controller commands full throttle, the car slows to 80 km/h anyway, and the error of 20 km/h does not go away. The integral keeps accumulating: after 30 seconds it holds the equivalent of hundreds of percent of throttle.
Nothing appears wrong. The throttle is clamped at 100 percent, which is correct.
Then the hill ends. The error goes negative and the integral begins to unwind, from an enormous value. It takes as long to unwind as it took to accumulate, and for that entire time the commanded throttle stays above 100 percent and therefore clamped at full.
The car accelerates at full throttle down the far side of the hill, for tens of seconds, ignoring the fact that it is now 40 km/h over the setpoint. The driver's experience is a cruise control that seemed fine, then lost its mind at the crest of a hill.
The fix is anti-windup, in one of its standard forms: stop accumulating while the output is saturated and the error would push it further, or clamp the integral term itself, or back-calculate from the difference between the commanded and actual throttle. All three are a handful of lines, none is optional, and the failure is invisible in any test that never saturates the actuator, which is why it survives into production.
P3. Scenario 1: the measurement is harder than the problem.
A domestic microwave oven heats for a set time at a set power. Closing the loop would mean measuring the food's internal temperature, which requires a probe inserted into an item of unknown geometry, in a chamber full of microwave energy, that the user will not do.
The reason: sensing cost and practicality exceed the value of the accuracy gained. The open-loop version is good enough because the user closes the loop themselves, by looking at the food and adding thirty seconds.
Scenario 2: the plant is repeatable and the disturbance is small.
A stepper-driven 3D printer axis moves open loop. The step angle is a geometric property repeatable to a fraction of a percent, the load is predictable and well within the motor's torque, and the machine homes against a switch at the start of every print to re-establish truth.
The reason: the open-loop accuracy already exceeds the requirement, so feedback would add cost, tuning, and new failure modes for no improvement. Note the two conditions that make it work: the plant is repeatable, and there is a periodic re-reference (homing) that bounds the accumulated error. Remove either and the argument collapses, which is why a printer that skips steps mid-print produces a permanently offset object.
Scenario 3: feedback would be dangerous or would destabilise the system.
A welding power supply's arc-start sequence, or a motor's soft-start ramp, runs to a fixed open-loop profile. During the transient the measurement is meaningless (the arc is not established, the motor is not turning, the sensor is reading noise), and a controller acting on that measurement would command something violent.
The reason: during the transient there is no valid measurement to feed back, so an open-loop profile is not a compromise, it is the only correct behaviour. The same argument governs why control loops are started in manual and switched to auto only once the plant is in a sensible state.
The pattern that gets most of both: feedforward plus feedback.
Compute the expected actuator output from a model of the process and apply it directly (open loop), then let a feedback controller correct only the residual error.
Example, the cruise control from the previous problem: the car knows the road gradient from an inclinometer or from the map. Feed the gradient through a model (throttle required = f(gradient, speed, mass)) and apply that immediately, before any speed error has appeared. The PI controller then handles only what the model got wrong: wind, a heavier load than assumed, a mis-calibrated model.
Why it is better than either alone:
- Faster than feedback alone, because the correction is applied when the disturbance occurs rather than after it has produced an error. Feedback is by definition reactive: something must go wrong first
- More accurate than feedforward alone, because the feedback term absorbs every modelling error, and the model does not have to be good
- The feedback controller's job gets easier, so it can be tuned more gently, which means less overshoot and more stability margin
The same structure appears throughout this course: a heater's feedforward from measured inflow, a robot arm's gravity compensation term added to a PD controller, a cascade's inner loop rejecting supply disturbances before the outer loop sees them.
The question that decides an unfamiliar case: what disturbs this process, how large is it, and can I measure the thing I actually care about?
- Disturbances small or absent, and the plant repeatable: open loop. Add a periodic re-reference if error can accumulate
- Disturbances significant, and the controlled variable measurable: closed loop
- Disturbances significant and measurable in advance, but the controlled variable hard to measure: feedforward from the disturbance
- Disturbances significant but the controlled variable not measurable at all: you cannot close the loop on it. Either find a proxy you can measure (a tray temperature standing in for composition, a motor current standing in for torque) and accept that you are controlling the proxy, or accept open loop and manage the consequences
EFeynman Exercise
Explain to a beginner, using driving with your eyes closed for a few seconds versus driving with your eyes open: (1) why with your eyes closed you just steer based on what you remember and hope nothing changed, (2) why with your eyes open you constantly see where you are versus where you want to be and correct, and (3) why eyes-open lets you handle a curve or a gust of wind that eyes-closed cannot.
REVEAL MODEL ANSWER
The difference between open-loop and closed-loop control is like the difference between driving with your eyes closed for a few seconds and driving with your eyes open. With your eyes closed (open-loop), you just steer based on what you remember: 'the road was straight, so I'll hold the wheel here', and you hope nothing changed. You're applying a fixed action with no way to check the result. That's fine for a moment on a straight, empty road where nothing unexpected happens. It's simple, and you don't need to look. But you're blind: you can't tell if you're drifting. With your eyes open (closed-loop), you constantly see where you actually are versus where you want to be, and you correct the wheel to stay in your lane. You're using feedback, comparing reality to your goal and adjusting. The payoff is that eyes-open lets you handle a curve or a gust of wind that eyes-closed cannot: when the road bends or a crosswind pushes you, you see yourself drifting toward the edge and steer back. Whereas with your eyes closed you'd have no idea anything pushed you and you'd drift right off the road. That's exactly why closed-loop control can hold a target despite disturbances (the curve, the wind) while open-loop just executes its plan and hopes: one watches and corrects, the other acts blindly. (And just as you don't need to stare at the road to butter toast, some simple, reliable actions are fine done 'eyes closed' (open-loop) which is cheaper and simpler when nothing will throw them off.)
FError Analysis Framework
- Thinking open- vs closed-loop is about complexity. Why: closed-loop systems seem fancier. Recognise: you miss the real distinction, feedback. Avoid: open-loop has no feedback (acts blindly); closed-loop measures and corrects.
- Assuming closed-loop is always the right choice. Why: feedback handles everything. Recognise: it needs a sensor and adds cost/complexity, sometimes needlessly. Avoid: use closed-loop for disturbance rejection/accuracy; open-loop when the action is reliable.
- Expecting an open-loop system to correct its errors. Why: it's a control system. Recognise: with no feedback it can't know there's an error. Avoid: only closed-loop (with measurement) can detect and correct errors.
- Using open-loop where disturbances are significant. Why: the action usually gives the right result. Recognise: disturbances push the result off and it can't compensate. Avoid: close the loop where disturbances/accuracy demand it.
GMini Challenge
For an automated greenhouse, decide open-loop vs closed-loop for: (a) maintaining the air temperature, (b) running the irrigation for a fixed daily duration, and (c) keeping the soil moisture at a target. Justify each choice and explain the open-vs-closed reasoning.
REVEAL MODEL ANSWER
Decisions:
| Function | Choice | Why |
|---|---|---|
| (a) Maintain air temperature | Closed-loop | Must hold a target despite disturbances (sun, outside cold, door opening): measure temperature, correct heating/venting |
| (b) Fixed daily irrigation duration | Open-loop | A predetermined timed action; if duration reliably gives enough water and disturbances are negligible, no feedback needed. Simple, cheap |
| (c) Keep soil moisture at a target | Closed-loop | Moisture varies with weather/evaporation/plant uptake (disturbances); to hold a target you must measure moisture and correct watering |
Reasoning:
- (a) Air temperature: closed-loop: the temperature is constantly disturbed (sunlight, outside conditions, ventilation), so to hold it at a setpoint you need feedback: measure it, compare to target, and adjust heating/cooling/venting to reject those disturbances. Open-loop (a fixed heater schedule) would be blind to the disturbances and let the temperature wander.
- (b) Timed irrigation: open-loop: if a fixed daily watering duration reliably delivers an adequate amount and there's no significant disturbance to correct, a simple timed action suffices, no sensor needed. Closing the loop here would add cost/complexity for little benefit (the open-loop action is reliable enough).
- (c) Soil moisture: closed-loop: moisture is pushed around by evaporation, weather, and plant uptake (disturbances), so to keep it at a target you must measure soil moisture and correct the watering. Feedback rejects those disturbances and holds the setpoint, which open-loop can't.
The open-vs-closed reasoning: use closed-loop wherever you must hold a target despite disturbances (air temperature, soil moisture). The sensor and feedback are worth their cost because the variable is disturbed and accuracy matters; use open-loop for simple, reliable actions (a timed watering) where a predetermined action dependably gives an acceptable result and disturbances are negligible, simpler and cheaper. Match the approach to whether feedback is actually needed. The foundational control decision.
Quiz Check
A quick auto-graded check, separate from the recall cards above. Your score is pooled with the recall cards into this module's Mastery score, and completing this lesson requires the quiz submitted with pooled mastery at 80% or above.