Control Systems

Introduction

Control Systems studies how dynamic systems can be made to behave in a desired way. You encounter control whenever a system measures what is happening, compares it with what should happen, and changes its input accordingly. Examples include aircraft autopilots, motor drives, robotic manipulators, chemical process plants, power converters, autonomous vehicles, climate control, biomedical devices, and networked cyber-physical systems.

At university level, control engineering connects Differential equations, Linear algebra, Signals and systems, Dynamical systems, Optimization, Probability, Electronics, Mechanical engineering, and Computer engineering. The central challenge is not merely to make a system respond, but to make it respond stably, accurately, quickly, robustly, and within physical constraints.

By the end of this aiMOOC, you should be able to model linear dynamic systems, interpret transfer functions and state-space models, analyze stability and transient response, explain feedback, design and tune basic controllers, reason with Bode and root-locus methods, assess controllability and observability, and connect classical control ideas with digital and modern state-space design.

The diagram above shows the essential closed-loop idea: the plant output is measured and fed back so that the controller can act on the error between the reference and the measured output.

Core Concepts of Control

Open-Loop and Closed-Loop Control

An open-loop system chooses an input without using the measured output to correct its action. A timed toaster is a simple example: the heating duration is set in advance, but browning is not directly measured. Open-loop control can be inexpensive and simple, but it cannot automatically compensate for disturbances or model errors.

A closed-loop system uses feedback. The desired output is the reference or setpoint. A sensor measures the actual output, a comparator forms an error, and the controller uses that error to compute a new control input. Feedback can improve tracking, disturbance rejection, and robustness, but poor controller design can also create oscillation or instability.

Important physical elements include the plant to be controlled, the controller, the actuator that converts the controller command into physical action, the sensor that measures relevant variables, and the disturbances and measurement noise that affect operation.

Modeling Dynamic Systems

Control design begins with a model. For many physical systems, you start from conservation laws such as Newton's laws, Kirchhoff's laws, mass balance, or energy balance. These usually lead to differential equations. Around an operating point, a nonlinear model can often be approximated by a linear model through linearization.

For a continuous-time linear time-invariant system with zero initial conditions, the transfer function is the ratio of the Laplace transform of the output to the Laplace transform of the input:

$G (s) = \frac{Y (s)}{U (s)}$

The roots of the denominator are the poles; roots of the numerator are the zeros. Poles strongly influence natural modes and stability, while zeros shape how inputs are transmitted to outputs.

A general state-space model is

$\dot{x} = A x + B u$

$y = C x + D u$

where $x$ is the state vector, $u$ is the input, and $y$ is the output. State-space form is especially useful for multi-input multi-output systems, modern control design, estimation, and digital implementation.

Feedback Algebra and Loop Gain

For a negative-feedback loop with controller $C (s)$ , plant $G (s)$ , and feedback path $H (s)$ , the closed-loop transfer from reference to output is

$\frac{Y (s)}{R (s)} = \frac{C (s) G (s)}{1 + C (s) G (s) H (s)}$

The product around the loop is the loop gain. In a unity-feedback system, let $L (s) = C (s) G (s)$ . The sensitivity function

$S (s) = \frac{1}{1 + L (s)}$

describes how some disturbances and model variations propagate, while the complementary sensitivity

$T (s) = \frac{L (s)}{1 + L (s)}$

describes reference tracking and important noise-transmission effects. Because $S + T = 1$ , improving behavior in one frequency range usually creates trade-offs elsewhere.

Time-Domain Analysis

Stability and Poles

A control system must first be stable. For a continuous-time linear state-space model, asymptotic stability requires all eigenvalues of the system matrix $A$ to have strictly negative real parts. In transfer-function language, a stable minimal continuous-time LTI system has all poles in the open left half of the complex $s$ -plane.

Stability is not enough by itself. A stable system may still be too slow, too oscillatory, too sensitive to disturbances, or too demanding on actuators. Controller design therefore combines stability with quantitative performance requirements.

For discrete-time LTI systems, asymptotic stability requires the relevant poles or eigenvalues to lie strictly inside the unit circle in the complex $z$ -plane.

Step Response and Second-Order Behavior

A unit-step input is a standard test because it reveals how the system moves from one operating condition to another. Useful time-domain measures include rise time, peak time, maximum overshoot, settling time, and steady-state error.

A standard second-order model is

$G (s) = \frac{ω_{n}^{2}}{s^{2} + 2 ζ ω_{n} s + ω_{n}^{2}}$

where $ω_{n}$ is the natural frequency and $ζ$ is the damping ratio. When Fehler beim Parsen (Syntaxfehler): {\displaystyle 0<\zeta<1} , the response is underdamped and may overshoot. Increasing damping generally reduces oscillation, while increasing bandwidth can speed response but may increase sensitivity to noise and unmodeled dynamics.

PID Control

The proportional-integral-derivative controller is one of the most widely used controller structures. In ideal continuous-time form,

$u (t) = K_{p} e (t) + K_{i} \int e (t) d t + K_{d} \frac{d e (t)}{d t}$

Proportional action responds to present error. Increasing proportional gain often speeds response, but excessive gain can increase overshoot or instability risk. Integral action accumulates error and can eliminate constant steady-state offset for many plant types, but it can slow or destabilize the loop and may suffer from integrator windup under actuator saturation. Derivative action reacts to the rate of change of error and can add damping, but pure differentiation amplifies high-frequency noise, so practical derivative action is normally filtered.

A practical tuning process should begin with requirements and a plant model, not with arbitrary gain changes. You should inspect actuator limits, sensor noise, delays, operating-point variation, and safety constraints. Anti-windup logic, derivative filtering, setpoint weighting, and gain scheduling are common practical additions.

Frequency-Domain Analysis

Frequency Response and Bode Plots

The frequency response of a stable LTI system is evaluated by substituting $s = j ω$ . A Bode plot separates this response into magnitude and phase versus logarithmic frequency. Bode plots help you see bandwidth, resonances, low-frequency tracking capability, high-frequency attenuation, and stability margins.

Gain margin and phase margin are classical measures of how far an open-loop frequency response is from an instability boundary. Larger margins often indicate greater tolerance to gain changes and phase lag, but no single margin captures every uncertainty or robustness issue.

Nyquist Stability Criterion

The Nyquist plot maps the complex frequency response of the loop transfer function. The Nyquist stability criterion relates encirclements of the critical point $- 1$ to the number of open-loop and closed-loop unstable poles. It is powerful because it can assess closed-loop stability from open-loop frequency-response information, including systems with time delay.

When you use Nyquist reasoning, you must keep the contour convention and the number of open-loop right-half-plane poles explicit. Memorizing a sign rule without the underlying contour convention is unsafe; derive the relation consistently for the convention used in your course.

Root Locus and Classical Design

The root locus shows how closed-loop pole locations change as a scalar loop gain varies. It gives geometric insight into the relation between gain, damping, natural frequency, and stability. Root-locus design is especially useful when adding lead, lag, proportional-derivative, or proportional-integral compensation.

A typical classical design sequence is to convert performance goals into pole-location or frequency-domain requirements, choose a controller structure, shape the loop, and then validate the full closed-loop response against disturbances, noise, uncertainty, and actuator limits.

State-Space and Modern Control

Controllability and Observability

For the state-space model $\dot{x} = A x + B u$ , a system is controllable if suitable inputs can move the state between arbitrary states in finite time. For an $n$ -state LTI system, the controllability matrix

$𝒞 = [B A B A^{2} B \dots A^{n - 1} B]$

must have rank $n$ for complete state controllability.

A system is observable if its internal state can be reconstructed from the known input and measured output over time. The observability matrix

$𝒪 = [\begin{matrix} C \\ C A \\ C A^{2} \\ ⋮ \\ C A^{n - 1} \end{matrix}]$

must have rank $n$ for complete state observability.

State Feedback, Pole Placement, and LQR

If a linear system is controllable, full-state feedback of the form

$u = - K x + r_{f}$

can be designed to assign the eigenvalues of $A - B K$ to desired locations, subject to practical limitations. Pole placement directly targets closed-loop dynamics but does not by itself encode an optimal trade-off between state error and control effort.

The linear quadratic regulator chooses a state-feedback law to minimize a quadratic performance index such as

$J = \int_{0}^{\infty} (x^{T} Q x + u^{T} R u) d t$

with weighting matrices that express the relative cost of state deviations and control effort. For standard continuous-time LQR assumptions, the solution is obtained from an algebraic Riccati equation.

State Estimation

Full state feedback requires state information, but many states cannot be measured directly. A state observer combines the model with measured outputs to estimate the state. A Luenberger observer places estimator error dynamics through an observer gain. A Kalman filter provides an optimal linear state estimate under its standard stochastic assumptions, including specified process and measurement noise statistics.

The separation principle allows controller and observer design to be treated separately for suitable linear systems: state-feedback dynamics and observer-error dynamics can be assigned independently when the required controllability and observability conditions hold.

Digital Control and Implementation

Real controllers are commonly implemented on microcontrollers, digital signal processors, programmable logic controllers, or industrial computers. Sampling converts a continuous-time problem into a discrete-time one. A typical discrete state model is

$x_{k + 1} = A_{d} x_{k} + B_{d} u_{k}$

$y_{k} = C_{d} x_{k} + D_{d} u_{k}$

You must choose a sampling rate that captures relevant dynamics while allowing enough time for sensing, computation, communication, and actuation. Too-slow sampling can distort dynamics or create aliasing. Computation and communication delays add phase lag and can reduce stability margins.

Digital implementation also introduces quantization, finite word length, saturation, rate limits, scheduling jitter, and software faults. A controller that is stable in an ideal simulation may fail if these effects are ignored. Verification should therefore include worst-case delays, actuator constraints, sensor failures, and realistic discretization.

Robustness, Uncertainty, and Nonlinearity

Every model is incomplete. Parameters vary, neglected modes exist, actuators saturate, sensors have noise, and operating conditions change. Robust control asks whether stability and performance remain acceptable despite these uncertainties.

Useful classical robustness tools include sensitivity functions, gain and phase margins, bandwidth analysis, and loop shaping. More advanced university courses may introduce small-gain reasoning, structured singular value analysis, $H_{\infty}$ design, model predictive control, adaptive control, and nonlinear stability methods such as Lyapunov analysis.

You should distinguish nominal performance from robust performance. A controller that performs beautifully on one exact model may be a poor engineering design if small parameter changes make it unstable.

Case Study: Inverted Pendulum

An inverted pendulum is a classic benchmark because the upright equilibrium is open-loop unstable. A movable base must accelerate so that the pendulum angle remains near vertical. The system couples translational and rotational dynamics, making it useful for studying linearization, state feedback, estimator design, actuator limits, and nonlinear validation.

A typical workflow is to derive nonlinear equations of motion, select the upright operating point, linearize the model, verify controllability, design a controller such as pole placement or LQR, simulate disturbances and saturation, and then test the nonlinear closed-loop model. A real implementation must also address sensor calibration, finite rail length, motor current limits, friction, sampling, and safety shutdowns.

Engineering Workflow

A disciplined control-engineering workflow reduces the risk of designing a controller for the wrong problem.

Requirements engineering: Define tracking accuracy, disturbance rejection, settling time, overshoot, bandwidth, energy use, safety limits, and operating range.
System identification: Derive or estimate a model and document assumptions, uncertainty, delays, nonlinearities, and neglected dynamics.
Stability analysis: Analyze open-loop and candidate closed-loop stability before optimizing performance.
Controller design: Choose a method that matches the model, measurements, constraints, and required robustness.
Simulation: Test nominal and worst-case scenarios, disturbances, noise, saturation, and model uncertainty.
Implementation: Discretize, schedule computation, handle sensors and actuators, and add fault handling.
Validation: Compare measured performance with requirements and revise the model when experiments reveal discrepancies.

Interactive Tasks

Quiz: Test Your Knowledge

What is the main defining feature of closed-loop control? (The controller uses measured output information) (!The input is fixed before operation) (!The plant has no disturbances) (!The sensor is removed from the system)

For a continuous-time asymptotically stable LTI system, where must its poles lie? (In the open left half plane) (!On the positive real axis) (!In the open right half plane) (!Exactly on the imaginary axis)

Which PID term is primarily used to remove persistent steady-state error? (Integral action) (!Derivative action) (!Sensor filtering) (!Feedforward scaling)

What does a Bode plot display? (Magnitude and phase versus frequency) (!State trajectories versus time) (!Only pole locations) (!Only actuator saturation limits)

What does the root locus show? (Closed loop pole motion as gain varies) (!Sensor noise as temperature varies) (!State dimension as time varies) (!Sampling error as memory varies)

What does complete controllability mean? (The input can drive the state throughout the state space) (!Every state is directly measured) (!The plant has no zeros) (!The controller requires no model)

What does complete observability mean? (The state can be reconstructed from input and output information) (!Every actuator is duplicated) (!Every pole is at the origin) (!The reference is always constant)

Where must poles lie for an asymptotically stable discrete-time LTI system? (Inside the unit circle) (!Outside the unit circle) (!Only on the real axis) (!Exactly on the unit circle)

Why is ideal derivative action problematic in practice? (It amplifies high frequency measurement noise) (!It always creates steady state error) (!It removes all damping) (!It makes sampling unnecessary)

What is a central purpose of robust control analysis? (To maintain acceptable behavior despite uncertainty) (!To eliminate the need for sensors) (!To guarantee zero computation time) (!To replace every model with open loop control)

Memory Game

Feedback	Using measured output information to influence the control input
Actuator	Device that converts a control command into physical action
Pole	Complex value associated with a natural mode of an LTI model
Observability	Ability to reconstruct internal state from known input and measured output
Bandwidth	Frequency range over which a closed-loop system responds effectively
Saturation	Limit that prevents an actuator command from increasing indefinitely
Estimator	Algorithm that reconstructs unmeasured state variables

Drag and Drop

Match the correct terms.	Topic
Reference signal	Desired system output
Tracking error	Difference between desired and measured output
Plant model	Mathematical description of the controlled process
Control law	Rule that computes the actuator command
Feedback sensor	Device that measures a variable used by the controller

...

Crossword Puzzle

Feedback	What process returns measured output information to the controller?
Stability	What property means the system does not diverge under the stated conditions?
Actuator	What device turns a controller command into physical action?
Sensor	What device measures a physical variable for feedback?
Controller	What component computes the command applied to the plant?
Observability	What property allows internal state to be reconstructed from measurements?

LearningApps

Cloze Text

A closed-loop controller uses

to compare measured behavior with the desired behavior. A transfer function represents an LTI input-output relation in the

domain. Continuous-time asymptotic stability requires poles to have negative

parts. Integral action can remove persistent steady-state

. A Bode plot shows magnitude and phase as functions of

. Root locus traces the movement of closed-loop

as gain changes. A state-space model describes a system using a vector of internal

. Complete controllability requires the controllability matrix to have full

. Complete observability means the state can be reconstructed from measured

. Robust design checks whether acceptable performance remains despite model

.

Open-Ended Tasks

Easy

Feedback diagram: Draw and label a closed-loop control diagram for a room-temperature system, including reference, controller, actuator, plant, sensor, output, and one disturbance.
Step response interpretation: Find or generate a stable step-response graph and explain rise time, overshoot, settling time, and steady-state error in your own words.
Control application photo study: Produce a one-page visual collection of four real systems that use feedback and annotate what is sensed, what is actuated, and what variable is controlled.
PID concept video: Record a two-minute explainer that uses one practical example to distinguish proportional, integral, and derivative action.

Standard

DC motor modeling: Derive a simplified transfer-function or state-space model of a DC motor and state every physical assumption you make.
PID simulation study: Simulate a second-order plant with several PID settings and compare tracking performance, control effort, overshoot, and disturbance rejection.
Frequency response investigation: Create Bode plots for a chosen plant, estimate bandwidth and stability margins, and explain what changes when controller gain is increased.
Control engineer interview: Interview a practitioner or researcher about one deployed control system and summarize the most important modeling, sensing, safety, and implementation challenges.

Advanced

Inverted pendulum controller: Build a simulation of a linearized inverted pendulum, verify controllability, design state feedback, and test the nonlinear model with actuator saturation.
Observer design: Design a state observer for a system with at least one unmeasured state, compare true and estimated trajectories, and investigate sensitivity to measurement noise.
Robustness experiment: Create a family of uncertain plant models, design one controller, and determine where stability or performance requirements fail as uncertainty increases.
Digital control prototype: Implement a controller on a microcontroller or real-time simulator, measure sampling and computation delay, and compare experimental behavior with the continuous-time design.

Learning Assessment

Model-to-controller reasoning: Given a physical system description, derive a suitable model, justify the chosen states or transfer-function variables, and identify which assumptions could threaten controller validity.
Stability and performance trade-off: Compare two stable controller designs and argue which one is preferable when overshoot, settling time, noise sensitivity, and actuator effort are all considered.
Frequency-domain design decision: Use a Bode or Nyquist representation to identify a stability or robustness weakness and propose a controller modification that addresses it.
State-space feasibility: For a supplied state-space model, determine controllability and observability and explain what those results imply for state feedback and state estimation.
Implementation transfer: Convert a continuous controller concept into a digital implementation plan that addresses sampling, delay, quantization, saturation, and fault handling.
Uncertainty challenge: Evaluate whether a controller designed for one nominal model remains acceptable over a specified family of plant variations and defend your robustness criterion.

Evidence of Learning

Knowledge: You can explain feedback, stability, transient response, transfer functions, state-space models, frequency response, PID control, root locus, controllability, observability, digital control, and robustness.

Analytical skills: You can derive models, calculate poles and eigenvalues, interpret step responses, read Bode and Nyquist plots, test controllability and observability, and connect mathematical results to physical behavior.

Design skills: You can translate requirements into controller objectives, select a controller structure, tune or synthesize gains, and judge trade-offs among speed, damping, accuracy, robustness, noise sensitivity, and control effort.

Technical products: Strong evidence may include a documented simulation, controller code, plots, a model derivation, an experiment report, a design review, or a short technical presentation with reproducible results.

Transfer achievement: You can apply the control-design workflow to an unfamiliar physical system, identify missing information, test uncertainty and constraints, and explain why the resulting design should or should not be trusted.

OERs on the Topic

Wikibooks: Control Systems provides an open engineering textbook covering classical and modern control topics.

MIT OpenCourseWare: Analysis and Design of Feedback Control Systems provides university-level course materials on time-domain analysis, frequency-domain design, root locus, state variables, and digital control.

MIT OpenCourseWare: Feedback Control Systems provides state-space-oriented material on control design, implementation issues, robustness, and nonlinear effects.