Control-System Theory and Design

Introduction

Industrial automation depends on control systems that make machines and processes behave predictably. A conveyor must maintain speed, a robot joint must reach a commanded position, and a thermal process must hold temperature despite disturbances. These tasks require more than sensing and actuation: they require a mathematical model of the process, a control objective, a controller, and a method for evaluating performance.

A control system compares a desired value, or setpoint, with a measured process value and uses the difference to determine corrective action. In mechatronics, the controlled plant may contain mechanical, electrical, hydraulic, pneumatic, thermal, or software elements. Control-system theory provides a common language for all of them through differential equations, transfer functions, state-space models, stability criteria, and frequency-response methods.

This chapter develops the principles needed to model dynamic systems, assess stability and transient response, and design open-loop, closed-loop, PID, feedforward, cascade, ratio, and selected advanced control strategies.

Learning objectives

By the end of this chapter, the reader should be able to:

1. Distinguish open-loop and feedback control systems.
2. Derive and interpret transfer-function and state-space models.
3. Analyze stability, steady-state error, and transient response.
4. Select and tune proportional, integral, derivative, and PID controllers.
5. Explain feedforward, cascade, ratio, and advanced control structures.
6. Evaluate practical effects such as saturation, noise, delay, sampling, and actuator limitations.
7. Apply a systematic design procedure to an industrial automation problem.

1. Fundamental control concepts

A process or plant is the physical system being controlled. An actuator supplies manipulated input, while sensors measure process variables. The setpoint is the desired value, the controlled variable is the variable regulated by the controller, and the manipulated variable is the input adjusted to achieve regulation. A disturbance affects the process but is not normally commanded by the controller.

For example, in a motor-speed system, speed is the controlled variable, motor voltage is the manipulated variable, load torque is a disturbance, and the desired speed is the setpoint.

Open-loop control

An open-loop controller produces an action without measuring the resulting output. A timer-based washing machine cycle is a familiar example. Open-loop control is simple and inexpensive, but it cannot automatically correct errors caused by changing loads, friction, supply voltage, or disturbances.

Closed-loop control

A closed-loop, or feedback, system measures the output and compares it with the reference. For negative feedback, the error is

where e(t) is the error, r(t) is the reference, and y_m(t) is the measured output. If the sensor is ideal, y_m(t) equals the output y(t). Negative feedback tends to reduce tracking error and reject disturbances, but it introduces sensor noise, possible instability, and additional cost.

Figure: Basic negative-feedback control loop.

For a linear system with controller transfer function C(s), plant transfer function G(s), and sensor transfer function H(s), the closed-loop reference-to-output transfer function is

Here s is the Laplace variable. The expression assumes linear time-invariant components, zero initial conditions, and negative feedback. The product L(s) = C(s)G(s)H(s) is the loop transfer function. The characteristic equation is 1 + L(s) = 0; its roots determine closed-loop stability.

2. Dynamic-system modeling

Differential-equation models

Physical laws are usually the starting point for a model. For a translational mass–damper–spring system,

where M is mass in kilograms, B is viscous damping in newton-seconds per metre, K is stiffness in newtons per metre, x is displacement in metres, and F is applied force in newtons. The equation assumes linear viscous damping and a linear spring, with lumped parameters.

Taking the Laplace transform under zero initial conditions gives

This is the plant transfer function from force to displacement. Transfer functions are useful for interconnecting linear systems and examining poles and zeros, but they do not directly represent nonlinearities or arbitrary initial conditions.

First-order systems

A standard first-order model is

where K is the static gain and τ is the time constant in seconds. For a unit-step input, the output is

The response reaches approximately 63.2% of its final value at t = τ, 86.5% at 2τ, 95.0% at 3τ, and 98.2% at 4τ. These percentages apply to a stable first-order system with zero initial output.

Second-order systems

The standard normalized second-order transfer function is

where K is the DC gain, ω_n is the undamped natural frequency in radians per second, and ζ is the damping ratio. For 0 < ζ < 1, the system is underdamped and oscillatory. ζ = 1 gives critical damping; ζ > 1 gives overdamping; ζ = 0 gives undamped oscillation. Stability requires positive ω_n and positive ζ for this standard form.

For a unit-step response with 0 < ζ < 1, the percent overshoot is

The 2% settling-time approximation is

These relations assume a dominant second-order response and a 2% settling band. They should not be used blindly when higher-order poles, zeros, saturation, or delay substantially influence the response.

State-space representation

A general continuous-time linear model is

where x is the state vector, u is the input vector, y is the output vector, and A, B, C, and D are system matrices of compatible dimensions. The matrix A describes internal dynamics, B input coupling, C output selection, and D direct feedthrough. State-space models are especially useful for multiple-input multiple-output systems, time-varying systems, digital control, and modern control design.

3. Stability and performance

A continuous-time linear system is asymptotically stable when all closed-loop poles lie strictly in the left half of the complex s-plane. Poles on the imaginary axis may produce marginal stability only under restricted conditions; repeated imaginary-axis poles cause instability. Any right-half-plane pole causes instability.

Routh–Hurwitz criterion

For a polynomial characteristic equation

with real coefficients and a_n > 0, the Routh–Hurwitz test determines the number of right-half-plane roots from the signs in the first column of the Routh array. A stable polynomial requires all first-column entries to have the same sign, assuming no special zero-row cases. The test is valuable because it assesses stability without explicitly calculating roots.

Root locus and frequency response

The root-locus method shows how closed-loop poles move as a real controller gain changes. It supports gain selection for desired damping and speed. Frequency-response methods describe magnitude and phase as functions of sinusoidal frequency. Important measures include gain margin, phase margin, bandwidth, and crossover frequency.

A positive phase margin generally indicates useful robustness, although the required value depends on plant uncertainty, delay, noise, and performance requirements. A larger bandwidth often improves response speed but can amplify high-frequency measurement noise and unmodeled dynamics.

Steady-state error

For unity feedback, the error transfer function is

where G_c is the controller and G_p is the plant. If the closed loop is stable, the final-value theorem gives

The theorem applies only when all poles of sE(s) are in the open left half-plane, except possibly a simple pole at the origin. Integral action increases the system type and can eliminate steady-state error for a constant reference, but it may reduce stability margins and create windup.

4. Controller structures

Proportional control

A proportional controller is

where K_p is dimensionless only when input and output units have been appropriately scaled. Increasing K_p generally reduces error and speeds response, but excessive gain can cause oscillation or instability. Proportional control alone cannot usually eliminate offset caused by a constant disturbance.

Integral control

Integral action accumulates error:

where K_i has units of controller-output per error-unit per second. It removes constant offset in a stable loop, but introduces an additional pole and can cause sluggish or oscillatory behavior. The integral term should be limited or stopped when the actuator saturates and the error would continue driving the integral in the same direction.

Derivative control

Derivative action is

where K_d has units of controller-output times seconds per error-unit. It responds to the rate of error change and can improve damping. Pure differentiation amplifies high-frequency noise, so industrial controllers use filtered derivative action. Derivative action is often applied to the measured process variable rather than the error to avoid a sharp “derivative kick” when the setpoint changes.

PID control

The ideal continuous PID law is

where T_i is integral time in seconds and T_d is derivative time in seconds. Equivalently, K_i = K_p/T_i and K_d = K_pT_d. A practical filtered form is

where N is a positive filter factor, commonly selected between about 5 and 20 depending on noise and implementation. The exact tuning must consider actuator limits, sensor noise, delay, and sampling.

Common tuning approaches include process-reaction-curve methods, frequency-response methods, relay autotuning, and model-based optimization. Rules such as Ziegler–Nichols are useful starting points but may produce excessive overshoot in demanding industrial applications. Modern practice often tunes against explicit objectives such as settling time, overshoot, robustness, energy use, and constraint violations.

Digital implementation

A digital controller samples signals at interval T_s seconds. Sampling frequency should be sufficiently higher than the dominant closed-loop bandwidth, and anti-alias filtering is required when analog signals contain higher-frequency content. A simple rectangular approximation to the integral is

where k is the sample index. Derivative estimates are sensitive to quantization and noise; filtered differences are preferred. Digital implementation must also address computational delay, task scheduling, numerical overflow, and fail-safe behavior.

5. Feedforward, cascade, and ratio control

Feedforward control

Feedback corrects error after it appears. Feedforward measures a disturbance and changes the manipulated variable before the disturbance significantly affects the output. If a process is approximately

where d is a disturbance, an ideal feedforward controller is

provided the required models are known and the inverse is physically realizable. In practice, model mismatch, unmeasured disturbances, actuator limits, and delays prevent perfect cancellation. Feedforward is therefore normally combined with feedback.

Cascade control

Cascade control uses an inner feedback loop nested within an outer loop. The inner loop regulates a fast secondary variable, while the outer loop regulates the primary variable. The inner loop must be faster than the outer loop, commonly by a factor of three to ten, and its sensor must detect important disturbances before they affect the primary variable.

Examples include temperature control with an inner flow loop, or motor-position control with inner velocity and current loops. Proper commissioning tunes the inner loop first, closes it, and then tunes the outer loop.

Ratio control

Ratio control maintains a desired relationship between two flow rates or other process variables:

where F₁ is the wild or reference flow and F₂ is the controlled flow. The setpoint for F₂ is generated from the measured F₁ and desired ratio R_d:

Units must be compatible, and density, composition, sensor bias, and signal dynamics may need compensation. Ratio control is common in blending, combustion, and chemical dosing.

6. Advanced control strategies

State feedback

If all relevant states are available, a state-feedback law can be written as

where K is a feedback-gain matrix and N is a reference precompensator. Gains may be selected by pole placement or linear-quadratic regulation. If states are not measured, an observer estimates them. A linear system is controllable when its states can be driven by inputs, and observable when its internal states can be reconstructed from measured outputs. These properties are assessed from the controllability and observability matrices.

Model predictive control

Model predictive control (MPC) uses a dynamic model to predict future outputs over a finite horizon. At each sample, it solves an optimization problem involving tracking error, control effort, and constraints, applies only the first calculated move, and repeats the calculation with new measurements. MPC is valuable for multivariable systems and processes with hard constraints, but it requires a sufficiently accurate model, reliable computation, and careful treatment of infeasibility.

Adaptive and robust control

Adaptive control changes controller parameters as plant characteristics vary. It is useful when payload, friction, or process gain changes substantially, but adaptation can be unsafe if parameter estimates become unreliable. Robust control explicitly seeks acceptable performance over a specified uncertainty range. Both approaches require validation beyond nominal simulation.

Nonlinear and intelligent methods

Many industrial plants contain dead zones, hysteresis, friction, saturation, and nonlinear flow relationships. Gain scheduling uses different controller parameters in different operating regions. Fuzzy control and neural-network methods can represent nonlinear relationships, but they require careful validation, explainability, monitoring, and fallback behavior. AI does not remove the need for physical safety systems, formal testing, or engineering accountability.

7. Worked example: position control of a servo axis

Consider an illustrative servo axis approximated by a first-order plant

with unity feedback and proportional controller C(s) = K_p. The closed-loop transfer function is

The closed-loop time constant is therefore

For K_p = 4,

The final output for a unit step is

Thus the steady-state error is 0.2, or 20% of the unit command. The 2% settling-time approximation for a first-order response is approximately 4τ_cl = 0.16 s. Increasing proportional gain would reduce the error and time constant, but actuator saturation, noise, unmodeled resonances, and robustness limits would eventually constrain the gain. Adding integral action could remove the constant error, provided the resulting loop remains stable and anti-windup is implemented.

8. Applied case: temperature control in a batch process

Background

This illustrative case concerns a jacketed mixing vessel used to heat a chemical batch. The controlled variable is vessel temperature. Steam flow through a control valve is the manipulated variable. Feed temperature and batch composition act as disturbances.

Problem

A single temperature PID loop produced slow recovery after cold-feed disturbances. Operators increased gain, but the valve frequently saturated and the temperature overshot. The process also exhibited measurement noise caused by a long sensor cable.

Analysis

A step test identified an approximate process model with significant delay:

where K is process gain, τ is the time constant, and θ is dead time. The delay limits achievable bandwidth; aggressive feedback cannot respond before the disturbance signal reaches the sensor. Valve saturation caused integral windup, and the noisy measurement made derivative action unreliable.

Approach and solution

The design team implemented a cascade structure. The outer temperature controller generated a steam-flow setpoint, while a fast inner flow controller regulated the valve and detected supply-pressure disturbances. A feedforward signal based on measured cold-feed flow and temperature was added to anticipate thermal-load changes. The temperature PID used filtered derivative action sparingly, and conditional integration prevented further accumulation while the valve was saturated in the wrong direction.

The design was commissioned in stages: sensor verification, manual inner-loop tuning, automatic inner-loop testing, outer-loop tuning, feedforward commissioning, and disturbance tests. High-temperature alarms and an independent shutdown system were retained outside the regulatory controller.

Outcome

The cascade loop reduced the effect of steam-pressure variation, feedforward reduced the initial temperature deviation, and anti-windup shortened recovery after valve saturation. The final design was not judged solely by response speed: product-quality variation, valve travel, alarm frequency, and safe operating limits were also evaluated.

Learning points

1. A controller cannot compensate immediately for an unmeasured disturbance.
2. The fastest useful loop should normally be placed inside a cascade structure.
3. Integral action requires explicit treatment of actuator saturation.
4. Safety instrumented functions should not depend exclusively on a complex control algorithm.
5. Commissioning and validation are as important as mathematical tuning.

9. Practical design workflow and limitations

A reliable control project normally follows these steps:

1. Define controlled variables, manipulated variables, disturbances, constraints, safety requirements, and performance criteria.
2. Establish units, signal ranges, sensor locations, actuator characteristics, and data quality.
3. Derive or identify a model, documenting assumptions and operating range.
4. Choose the simplest control structure that meets requirements.
5. Check open-loop behavior, controllability, observability, delay, and actuator authority.
6. Design and simulate the controller, including noise, saturation, quantization, and failures.
7. Commission conservatively, beginning with sensors and inner loops.
8. Validate tracking, disturbance rejection, robustness, energy use, alarms, and fallback modes.
9. Monitor performance and retune only under controlled change management.

Important limitations include nonlinear operation, parameter drift, dead time, sensor bias, actuator hysteresis, quantization, network delays, and interactions among loops. A mathematically stable linear model does not guarantee safe behavior in a real plant. Stability margins can be consumed by neglected dynamics, and a controller that performs well at one operating point may fail elsewhere.

Safety, governance, and professional responsibility

Automation engineers must separate basic regulatory control from independent protective functions where risk warrants it. Emergency stops, interlocks, permissives, and safety instrumented systems should be designed according to applicable organizational and industry standards. Cybersecurity is also a control concern: unauthorized changes to setpoints, logic, or firmware can create physical hazards. Access control, audit trails, secure configuration management, network segmentation, and tested recovery procedures are essential.

Engineers should document model assumptions, tuning parameters, alarm rationales, test evidence, and limits of AI-assisted design tools. AI-generated controller settings or code require independent review, simulation, traceability, and plant-level validation.

Chapter Summary

Control-system design connects physical modeling with purposeful manipulation of process behavior. Open-loop systems are simple but cannot correct unknown disturbances; closed-loop systems use feedback to reduce error but introduce stability and robustness concerns. Differential equations, transfer functions, and state-space models represent dynamic behavior under stated assumptions. Poles determine stability, while damping ratio, natural frequency, settling time, overshoot, bandwidth, and steady-state error describe performance.

P, I, and D actions address present error, accumulated error, and error rate, respectively. Practical PID design must account for filtering, sampling, saturation, delay, and anti-windup. Feedforward anticipates measured disturbances, cascade control separates fast and slow dynamics, and ratio control maintains a specified relationship between variables. State feedback, MPC, adaptive control, robust control, and nonlinear methods extend control capability for multivariable, constrained, or changing systems. Successful industrial design combines mathematical analysis, simulation, commissioning, safety engineering, cybersecurity, and continuous validation.

Key Terms

Actuator: Device that converts a control signal into physical action.

Anti-windup: Method that prevents integral action from accumulating excessively during actuator saturation.

Bandwidth: Frequency range over which a system responds effectively.

Cascade control: Nested control structure with an inner loop and an outer loop.

Closed-loop control: Control using measured output feedback.

Controlled variable: Process quantity maintained near its desired value.

Disturbance: Input that affects a process but is not normally commanded.

Feedforward control: Control based on measured disturbances or operating conditions.

Gain margin: Frequency-domain measure of allowable gain increase before instability.

Integral action: Controller action that accumulates error over time.

Loop transfer function: Product of transfer functions around a feedback loop.

Manipulated variable: Process input adjusted by the controller.

Model predictive control: Optimization-based control using predictions and constraints.

Overshoot: Amount by which a response exceeds its final value.

Plant: Physical process or system being controlled.

Pole: Value of s that makes a transfer-function denominator zero.

Ratio control: Control that maintains a desired ratio between process variables.

Settling time: Time required for a response to remain within a specified final-value band.

Steady-state error: Difference between desired and actual output after transients vanish.

Transfer function: Ratio of output to input transforms under zero initial conditions.

Review Questions

5 Short Answer Questions

1. What are the principal differences between open-loop and closed-loop control?
2. Define a plant, controlled variable, manipulated variable, and disturbance.
3. What information is conveyed by the poles of a closed-loop transfer function?
4. Why can integral action eliminate constant steady-state error?
5. Why is derivative action usually filtered in industrial controllers?

5 Long Answer Questions

1. Derive the closed-loop transfer function for a negative-feedback system and explain the significance of its characteristic equation.
2. Explain first-order and second-order system responses, including the roles of time constant, damping ratio, and natural frequency.
3. Describe a systematic PID tuning and commissioning procedure for an industrial process.
4. Compare feedforward, cascade, and ratio control, including suitable applications and limitations.
5. Explain how sampling, actuator saturation, sensor noise, and time delay affect digital control performance.

3 Analytical / Application Questions

1. A unity-feedback plant is G_p(s)=1/(0.5s+1) with proportional gain K_p. Derive the closed-loop time constant and steady-state error for a unit-step input, then discuss the trade-off involved in increasing K_p.
2. Design a control structure for a motor position system with current, speed, and position measurements. Justify the ordering of any cascade loops and identify likely disturbances.
3. A heating process has a long dead time and a valve that saturates. Recommend a controller structure and commissioning safeguards, explaining why a high-gain PID alone may be unsuitable.

Multiple Choice Questions with Answers

Question 1

Which condition indicates asymptotic stability for a continuous-time linear system?

A. All closed-loop poles lie strictly in the left half of the s-plane

B. At least one pole lies in the right half of the s-plane

C. All poles lie on the imaginary axis

D. The system has no zeros

Correct Answer: A. All closed-loop poles lie strictly in the left half of the s-plane

Question 2

For a first-order system with time constant τ, the step response reaches approximately what percentage of its final value at t = τ?

A. 25.0%

B. 50.0%

C. 63.2%

D. 95.0%

Correct Answer: C. 63.2%

Question 3

What is the principal purpose of integral action in a feedback controller?

A. To amplify high-frequency noise

B. To eliminate constant steady-state error in a stable loop

C. To measure the plant output

D. To create an open-loop system

Correct Answer: B. To eliminate constant steady-state error in a stable loop

Question 4

In cascade control, which loop is normally tuned first?

A. The outer, slower loop

B. The inner, faster loop

C. Both loops must be tuned simultaneously

D. Neither loop requires tuning

Correct Answer: B. The inner, faster loop

Question 5

For a unity-feedback system, the closed-loop transfer function is generally

A. C(s)G(s)H(s)

B. C(s)G(s) / [1 + C(s)G(s)H(s)]

C. 1 + C(s)G(s)H(s)

D. H(s) / [1 − C(s)G(s)]

Correct Answer: B. C(s)G(s) / [1 + C(s)G(s)H(s)]

Question 6

A proportional controller with K_p = 4 is applied to the plant G_p(s)=1/(0.2s+1) with unity feedback. What is the unit-step steady-state output?

A. 0.20

B. 0.50

C. 0.80

D. 1.25

Correct Answer: C. 0.80

Question 7

What is the main limitation of pure derivative control in an industrial loop?

A. It cannot respond to changing error

B. It amplifies high-frequency measurement noise

C. It always eliminates offset

D. It causes actuator saturation in every application

Correct Answer: B. It amplifies high-frequency measurement noise

Question 8

What is the main purpose of feedforward control?

A. To anticipate the effect of a measured disturbance

B. To remove all sensors from a control system

C. To guarantee stability without a model

D. To replace every feedback loop

Correct Answer: A. To anticipate the effect of a measured disturbance

Question 9

In ratio control, the controlled relationship is commonly expressed as

A. F₁ + F₂ = 0

B. F₂/F₁ = R

C. F₁F₂ = 1

D. F₂ − F₁ = τ

Correct Answer: B. F₂/F₁ = R

Question 10

Which feature is especially associated with model predictive control?

A. Use of future-output predictions and explicit constraints

B. Operation without any process model

C. Exclusive use of derivative action

D. Guaranteed performance without computation

Correct Answer: A. Use of future-output predictions and explicit constraints