Feedback Control Fundamentals

Introduction

Feedback control is the central mechanism by which a mechatronic system measures its behavior, compares that behavior with a desired target, and acts to reduce the difference. It enables a motor to track position, a robot arm to follow a trajectory, a vehicle to maintain speed, and a temperature-controlled process to reject disturbances.

A feedback loop contains four essential functions: measurement, comparison, control computation, and actuation. The controlled variable is measured by a sensor and returned to a controller. The controller compares it with the reference input and generates a command for the actuator. Because the controller responds to the actual output, feedback can compensate for disturbances, parameter variation, and imperfect modeling.

This chapter develops the standard concepts needed to model, analyze, and implement basic feedback systems. It emphasizes transfer-function models, transient and steady-state performance, stability, on–off control, proportional–integral–derivative (PID) control, feedforward compensation, simulation, and experimental tuning.

Learning Objectives

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

1. Represent a mechatronic feedback loop using a block diagram and transfer functions.
2. Derive closed-loop transfer functions and explain the roles of sensitivity and loop gain.
3. Evaluate rise time, settling time, overshoot, steady-state error, and stability.
4. Explain the operation and limitations of on–off, P, PI, PD, PID, and feedforward control.
5. Select and tune a basic controller using simulation and experimental tests.
6. Recognize implementation, safety, ethical, and reliability issues in practical control systems.

1. The Structure of a Feedback System

Let r(t) denote the reference or setpoint, y(t) the measured output, and e(t) the error. For unity negative feedback,

where all three signals must have compatible units. In the Laplace domain, assuming zero initial conditions,

A basic loop consists of a controller C(s), a plant or process P(s), and, in a general case, a sensor H(s). The loop transfer function is

For negative feedback, the closed-loop reference-to-output transfer function is

when the feedback sensor is included explicitly in the loop and the output is taken before H(s). For unity feedback, H(s) = 1, so

The denominator 1 + L(s) is the characteristic equation. Its roots, called closed-loop poles, determine stability and much of the transient response.

A block diagram should distinguish the physical plant from the controller and sensor. For example, in a servo drive, the plant may include the amplifier, motor, gearbox, and load; the controller may be implemented in a microcontroller; and the sensor may be an encoder or tachometer.

Figure: Standard negative-feedback loop with controller, plant, and sensor.

Open-loop and closed-loop control

In open-loop control, the command is generated without measuring the output. A timed washing-machine cycle is an example. Open-loop systems can be simple and inexpensive, but they cannot automatically correct errors caused by load changes or disturbances.

Closed-loop control measures the output and uses the measurement to modify the input. Feedback improves disturbance rejection and reduces sensitivity to plant uncertainty, but introduces sensor noise, delay, cost, and possible instability.

2. Mathematical Models for Mechatronic Plants

A transfer function is the ratio of output to input in the Laplace domain under zero initial conditions:

Here G(s) is the plant transfer function, U(s) is the plant input, and Y(s) is its output. The variable s has units of inverse seconds. Transfer functions are linear time-invariant models; they do not directly represent saturation, backlash, friction discontinuities, or changing parameters unless these effects are approximated or modeled separately.

A common 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 approaches K with time constant τ. At t = τ, the response has reached approximately 63.2% of its final value.

A standard second-order model is

where ω_n is the undamped natural frequency in rad/s and ζ is the damping ratio. For 0 < ζ < 1, the system is underdamped and exhibits oscillation. ζ = 1 gives critical damping, while ζ > 1 gives an overdamped response.

For a unity-feedback system with a proportional controller C(s) = K_p and a first-order plant P(s) = K/(τs + 1),

The closed-loop time constant is τ_cl = τ/(1 + K_pK), provided τ > 0 and 1 + K_pK > 0. Thus, increasing positive proportional gain makes this particular first-order response faster, but real actuators and unmodeled dynamics impose limits.

Worked Example 1: Closed-loop first-order response

A position-like plant has P(s) = 2/(0.5s + 1), and a proportional controller has K_p = 4. With unity feedback,

Dividing numerator and denominator by 9 gives

The steady-state gain is 0.8889, and the closed-loop time constant is approximately 0.0556 s. For a unit step, the final output is 0.8889, so the steady-state error is 1 − 0.8889 = 0.1111, or 11.11%. This example shows that increasing proportional gain reduces error but does not generally eliminate it for a type-0 plant.

3. Performance Measures

Control performance is evaluated in both the time and frequency domains. For a step response, the most common measures are:

• Rise time: time required to move between specified percentages of the final value, commonly 10% to 90%.
• Peak time: time at which the first maximum occurs.
• Maximum overshoot: amount by which the response exceeds its final value, expressed as a percentage.
• Settling time: time after which the response remains within a specified band, commonly ±2% or ±5%.
• Steady-state error: the final difference between reference and output.

Definitions must state the tolerance band and the input type. Different software packages may use different rise-time conventions.

For a standard underdamped second-order system, the percent overshoot for a unit step is

This formula applies only for 0 < ζ < 1 and a standard second-order response without significant zeros or higher-order effects. The approximate 2% settling time is

and the peak time is

These relationships are approximations based on dominant second-order poles.

Steady-state error and system type

The final value theorem states that, when all poles of sF(s) lie in the open left half-plane,

where F(s) is the Laplace transform of f(t). The stability condition is essential; applying the theorem to an unstable system can produce an invalid conclusion.

For unity feedback, define the error transfer function:

For a unit-step reference R(s) = 1/s, the steady-state error is

when the limit exists and the closed loop is stable. Integral action increases the number of integrators in the loop and can eliminate step error, but it may reduce stability margins and cause windup.

4. Stability

A continuous-time linear system is asymptotically stable when all closed-loop poles have negative real parts. Poles on the imaginary axis may produce sustained oscillation and are not asymptotically stable; poles in the right half-plane produce growing responses.

For a second-order characteristic polynomial

with real coefficients, asymptotic stability requires a_1 > 0 and a_0 > 0. Higher-order systems can be assessed using the Routh–Hurwitz criterion, root-locus methods, or frequency-response tools such as gain and phase margins.

In digital controllers, the corresponding stability requirement is that all closed-loop poles lie inside the unit circle in the z-plane: |z_i| < 1. Sampling frequency, computation delay, quantization, and zero-order-hold behavior must therefore be considered.

Feedback is not automatically beneficial. Excessive gain, sensor delay, actuator dynamics, or an incorrect feedback sign can create oscillation or instability. A practical design must preserve adequate gain margin and phase margin rather than merely achieving a fast nominal response.

5. Basic Controllers

On–off control

An on–off controller applies one of two actuator states according to the sign of the error. With hysteresis,

where h is the hysteresis half-band. Within the band, the previous state is retained. Hysteresis prevents rapid switching, or chatter, near the setpoint. On–off control is inexpensive and suitable for thermostats, relays, and simple limit functions, but it normally produces a limit cycle rather than exact continuous regulation.

Proportional control

The proportional gain K_p must have units that convert error units to actuator-command units. Increasing K_p usually reduces error and speeds response, but may increase overshoot, noise sensitivity, and actuator saturation.

Integral control

where K_i is the integral gain. Integral action accumulates error and can remove steady-state error for constant references and disturbances when the closed loop is stable. Its major practical problem is windup: the integral term continues to grow while the actuator is saturated.

Derivative control

Derivative action responds to the rate of error change and can improve damping and reduce overshoot. It amplifies high-frequency measurement noise, so practical implementations use a filtered derivative rather than an ideal differentiator.

PID control

The ideal continuous PID law is

where T_i is the integral time in seconds and T_d is the derivative time in seconds. Equivalently, K_i = K_p/T_i and K_d = K_pT_d. The ideal transfer function is

The ideal derivative is not directly realizable in the presence of noise. A filtered form is commonly written

where N is a dimensionless filter factor, typically selected as a finite positive value. Anti-windup logic, output limits, bumpless transfer, and derivative-on-measurement are important implementation features.

6. Feedforward and Feedback Together

Feedback reacts after an error appears. Feedforward uses a measured disturbance or reference and predicts the actuator command required to maintain the output. If a plant is approximately P(s) and a measurable disturbance enters through D(s), an ideal disturbance feedforward controller is

subject to sign conventions and realizability. Exact cancellation is rarely possible because models are uncertain, delays may make F(s) noncausal, and disturbances may not be measurable. Feedback is therefore retained to correct feedforward imperfections.

A useful architecture is

where u_ff is the predictive command and u_fb is the corrective feedback command. Feedforward is valuable in robot motion, motor torque control, thermal systems, and vehicle control when load or demand measurements are available.

7. Simulation and Experimental Tuning

Simulation should precede hardware testing. A practical workflow is:

1. Define inputs, outputs, units, actuator limits, sensor dynamics, and safety constraints.
2. Obtain a low-order plant model from physics, identification, or manufacturer data.
3. Simulate open-loop and closed-loop step, ramp, and disturbance responses.
4. Examine stability, saturation, noise sensitivity, tracking error, and robustness to parameter variation.
5. Implement conservative gains on hardware with current, speed, position, and emergency limits.
6. Log reference, measured output, error, controller terms, actuator command, and saturation status.
7. Tune one feature at a time and compare measured behavior with the model.

A common manual procedure is to begin with integral and derivative action disabled, increase K_p until the response is acceptably fast but not persistently oscillatory, add integral action gradually to remove offset, and add a small amount of derivative or derivative filtering if damping is needed. This is not a universal recipe: unstable plants, significant delays, nonlinear friction, and safety-critical systems require more formal design.

Empirical methods such as Ziegler–Nichols can provide an initial estimate, but they may deliberately approach oscillation and can yield excessive overshoot. Relay auto-tuning, frequency-response identification, pole placement, and optimization-based tuning may be preferable where repeatability and robustness matter.

8. Worked Design Example: DC Motor Speed Control

Consider an illustrative motor-speed plant approximated by

where the input is a normalized voltage command and the output is normalized speed. First use proportional control with K_p = 4. The closed-loop transfer function is

The response is stable, has a time constant of 0.04 s, and has a steady-state unit-step error of 1 − 0.8 = 0.2. Thus, the motor reaches only 80% of the demanded normalized speed under this simplified model.

Now add integral action with K_p = 4 and K_i = 20. The controller is C(s) = 4 + 20/s. The characteristic equation of unity feedback is

Substitution gives

Multiplying by s(0.20s + 1),

Dividing by 0.20 gives s^2 + 25s + 100 = 0. The roots are

so the poles are −5 and −20 rad/s. The system is stable and, because integral action is present, its ideal steady-state error to a constant step is zero. However, the slower pole at −5 rad/s implies a settling time of roughly 4/5 = 0.8 s using the dominant-pole approximation. This is slower than the proportional-only time constant suggested by the first-order model, illustrating that adding integral action changes the characteristic dynamics and must be tuned rather than assumed to improve every performance measure.

In a real motor, voltage saturation would limit the command. If the motor starts far from the target, the integral term could accumulate while the voltage is pinned at its limit. An anti-windup mechanism should suspend or back-calculate integral accumulation during saturation.

9. Illustrative Case: Conveyor Positioning System

Background

A small automated conveyor moves packages to a pick-up location. An encoder measures belt position, a motor drive supplies torque, and a microcontroller executes a digital controller.

Problem

An initial proportional controller positions accurately for light packages but leaves a noticeable offset under heavy loads. Increasing gain reduces the offset but causes oscillation and audible motor chatter. The actuator also reaches its voltage limit during rapid moves.

Analysis

The load changes the effective plant gain and introduces friction. Encoder quantization produces a small measurement staircase, while sampling and computation introduce delay. A pure derivative term would amplify encoder noise. The observed offset indicates insufficient low-frequency loop gain; the oscillation indicates that simply increasing proportional gain is reducing stability margin.

Approach and solution

The engineering team identifies a second-order-plus-delay approximation from small step tests. A PI controller is selected to eliminate constant-load offset, with a filtered derivative avoided initially because of encoder noise. The integral term includes conditional anti-windup. A velocity feedforward term based on desired belt speed is added to reduce tracking burden on feedback. The command is limited by motor voltage and acceleration constraints. Simulation includes saturation, encoder quantization, sampling, and two representative loads before hardware tests.

Outcome

The final system reaches the pick-up region with negligible steady-state offset, less oscillation than the high-gain proportional design, and reduced settling time during constant-speed moves. Performance remains acceptable across the tested loads, although the controller is not claimed to guarantee stability for every possible package mass.

Learning points

The case demonstrates that controller selection must consider plant variation, sensing quality, actuator limits, and task requirements. Integral action addresses bias but creates windup risk. Feedforward improves predictable motion but does not replace feedback. A model is valuable for design, but measured data and bounded validation are necessary before deployment.

10. Practical Limitations, Safety, and Professional Practice

Real systems are nonlinear. Saturation, dead zones, backlash, Coulomb friction, flexible couplings, thermal limits, and changing loads can invalidate a linear model. A controller that is stable around one operating point may behave poorly elsewhere.

Sensors require calibration, suitable bandwidth, electrical protection, and plausible-value checks. A failed sensor can cause dangerous commands if its signal is interpreted as a valid measurement. Safety-related systems should use independent limits, watchdogs, emergency stops, fault detection, and safe states. Control software should be version-controlled, tested, documented, and reviewed.

Engineers must not tune a system by intentionally allowing unsafe oscillation, especially when the plant contains high-energy motion, pressure, heat, or stored electrical energy. Logged data should be protected when it contains information about workers, production, or proprietary processes. Autonomous control should provide appropriate human oversight and clear responsibility for alarms and intervention.

Contemporary controllers increasingly combine classical feedback with observers, adaptive control, model predictive control, and machine-learning components. AI may assist with anomaly detection, model identification, or gain scheduling, but learned behavior must be validated under distribution shifts, sensor faults, and safety constraints. A black-box predictor should not silently override hard safety limits.

Best-Practice Checklist

• Define the controlled variable, reference, disturbance, units, and acceptable error.
• Use negative feedback with a verified signal polarity.
• Build the simplest model that captures dominant dynamics and delays.
• Check closed-loop poles and actuator limits before hardware tests.
• Report rise-time and settling-time definitions explicitly.
• Include sensor dynamics, sampling, delay, noise, and saturation in simulation.
• Use anti-windup for controllers with integral action.
• Filter derivative action and avoid differentiating noisy raw signals.
• Combine feedforward with feedback when disturbances are measurable and predictable.
• Validate robustness across loads, temperatures, tolerances, and faults.
• Maintain independent safety mechanisms and clear test records.

Chapter Summary

Feedback control compares a measured output with a reference and uses the error to command an actuator. The closed-loop transfer function for negative feedback is governed by the characteristic denominator 1 + L(s), whose poles determine stability and transient behavior. First-order and second-order models provide useful approximations for time constants, damping, overshoot, peak time, and settling time. Steady-state error depends on low-frequency loop gain and can be reduced or eliminated by integral action when the closed loop remains stable.

On–off control is simple but produces switching behavior. Proportional control improves speed and reduces error; integral control removes constant offset but can wind up; derivative control improves damping but is noise-sensitive. PID control combines these actions with practical filtering and anti-windup. Feedforward can compensate predictable demands or disturbances, while feedback corrects model error. Successful mechatronic control requires simulation, cautious experimental tuning, attention to nonlinearities and limits, and independent safety provisions.

Key Terms

Feedback: Using measured output information to influence system input.

Reference: Desired value of the controlled variable.

Error: Difference between reference and measured output.

Plant: Physical process or device being controlled.

Controller: Algorithm or device that converts error into an actuator command.

Sensor: Device that measures a physical variable.

Loop gain: Product of controller, plant, and feedback-path dynamics.

Closed-loop transfer function: Ratio of output to reference with feedback connected.

Pole: Root of a system denominator; its location influences stability and dynamics.

Stability: Ability of a system to remain bounded and converge appropriately after disturbances.

Rise time: Time required for a response to move through a specified percentage range.

Settling time: Time required to remain within a specified tolerance band.

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

Steady-state error: Limiting difference between reference and output.

On–off control: Control that switches between discrete actuator states.

Proportional control: Control action proportional to instantaneous error.

Integral windup: Excessive accumulation of integral action during saturation.

PID control: Combination of proportional, integral, and derivative actions.

Feedforward control: Predictive control based on reference or measured disturbance information.

Anti-windup: Method that limits or corrects integral accumulation during actuator saturation.

Review Questions

Short Answer Questions

1. What is the difference between open-loop and closed-loop control?
2. What physical meaning does the plant transfer function have?
3. Why does integral action reduce steady-state error?
4. What is actuator saturation, and why can it cause integral windup?
5. Why is derivative action usually filtered in a practical controller?

Long Answer Questions

1. Derive the unity-feedback closed-loop transfer function and explain the significance of its characteristic equation.
2. Compare on–off, proportional, PI, and PID control for a temperature-control application.
3. Explain how damping ratio and natural frequency influence a standard second-order step response.
4. Describe a safe simulation-to-hardware workflow for tuning a motor controller.
5. Explain why feedforward and feedback are often used together rather than separately.

Analytical / Application Questions

1. For P(s) = 3/(0.5s + 1) and K_p = 5 under unity feedback, derive the closed-loop transfer function and calculate its steady-state error to a unit step.
2. A position system oscillates after proportional gain is increased, while a constant load produces offset. Propose a controller modification and identify two implementation precautions.
3. Design an experimental test plan to compare a proportional controller with a PI controller on a conveyor, including measured variables, safety limits, and performance metrics.

Multiple Choice Questions with Answers

Question 1

For a unity-feedback system, which expression represents the closed-loop transfer function?

A. C(s)P(s)

B. C(s)P(s)/(1 + C(s)P(s))

C. 1/(C(s)P(s))

D. 1 − C(s)P(s)

Correct Answer: B. C(s)P(s)/(1 + C(s)P(s))

Question 2

For the first-order model G(s) = K/(τs + 1), τ represents the:

A. Static gain

B. Damping ratio

C. Time constant

D. Natural frequency

Correct Answer: C. Time constant

Question 3

Which controller action is primarily used to eliminate steady-state error for a constant reference in a stable loop?

A. Integral

B. Derivative

C. On–off

D. Sensor filtering

Correct Answer: A. Integral

Question 4

What is the principal practical danger of an unfiltered ideal derivative term?

A. It eliminates actuator saturation

B. It amplifies high-frequency measurement noise

C. It always creates steady-state error

D. It removes all feedback delay

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

Question 5

A continuous-time closed-loop system is asymptotically stable when its poles are located:

A. In the right half-plane

B. On the imaginary axis only

C. In the open left half-plane

D. Outside the unit circle

Correct Answer: C. In the open left half-plane

Question 6

For P(s) = 2/(0.5s + 1) and proportional gain K_p = 4 with unity feedback, what is the unit-step steady-state error?

A. 0.1111

B. 0.2000

C. 0.5000

D. 0.8889

Correct Answer: A. 0.1111

Question 7

What is the main purpose of hysteresis in an on–off controller?

A. To increase sensor bandwidth

B. To prevent rapid switching near the setpoint

C. To guarantee zero overshoot

D. To eliminate the need for feedback

Correct Answer: B. To prevent rapid switching near the setpoint

Question 8

Feedforward control differs from feedback control because feedforward:

A. Uses only the measured error after a disturbance occurs

B. Predicts a command using reference or disturbance information

C. Cannot be combined with feedback

D. Always requires an unstable plant

Correct Answer: B. Predicts a command using reference or disturbance information

Question 9

For a standard underdamped second-order system, increasing damping ratio generally reduces:

A. The number of sensors

B. Percent overshoot

C. The need for a reference

D. The plant’s physical inertia

Correct Answer: B. Percent overshoot

Question 10

Which feature is specifically intended to limit integral accumulation during actuator saturation?

A. Anti-windup

B. Open-loop gain

C. Encoder quantization

D. Natural frequency

Correct Answer: A. Anti-windup