Dynamic Modeling of Mechatronic Systems

Introduction and Context

A mechatronic system combines mechanical, electrical, electronic, sensing, computing, and actuation components. Intelligent control depends on a model that predicts how inputs, internal energy-storage variables, disturbances, and outputs are related over time. Dynamic modeling is therefore not merely a mathematical exercise: it is the basis for simulation, controller design, fault diagnosis, state estimation, safety analysis, and performance verification.

A useful model must balance fidelity and simplicity. A highly detailed model may represent friction, saturation, backlash, electrical parasitics, structural flexibility, and thermal effects, but it may be difficult to identify and computationally expensive. A simplified model is easier to analyze but may fail when operating conditions change. The engineering task is to select a model whose assumptions are explicit and whose accuracy is sufficient for the intended control objective.

This chapter develops continuous-time, discrete-time, state-space, and nonlinear models of electromechanical systems. It also examines equilibria, linearization, controllability, observability, uncertainty, and model limitations.

Learning Objectives

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

1. Identify system variables, energy-storage elements, inputs, outputs, and disturbances.
2. Derive differential-equation models of basic electromechanical systems.
3. Convert differential equations into continuous-time state-space form.
4. Obtain discrete-time models and explain the role of the sampling period.
5. Formulate nonlinear models and determine equilibrium points.
6. Linearize nonlinear systems around an operating point.
7. Test local controllability and observability using matrix criteria.
8. Distinguish parameter uncertainty, disturbances, unmodeled dynamics, and measurement noise.
9. Evaluate model limitations and select an appropriate model for an application.

1. Modeling Principles for Mechatronic Systems

1.1 System boundaries and variables

Modeling begins by defining a system boundary. Components outside the boundary are represented by inputs, disturbances, or environmental conditions. Components inside the boundary contribute states and outputs.

Typical variables include:

A state vector is written as x(t) = [x₁(t), x₂(t), …, x_n(t)]^T. It is a minimal collection of variables such that knowledge of the state and future input determines the future state, subject to the model assumptions.

1.2 Conservation laws

Most physical models are derived from conservation principles:

• Newton’s second law for translational and rotational mechanics.
• Kirchhoff’s voltage and current laws for electrical circuits.
• Conservation of mass and energy for fluid and thermal systems.
• Constitutive relations for springs, dampers, resistors, capacitors, and inductors.

For a translational mass, Newton’s law is

where m is mass in kilograms, q is displacement in metres, and ΣF is the sum of forces in newtons. For rotational motion,

where J is moment of inertia in kg·m², θ is angular position in radians, and Στ is the net torque in N·m.

For an ideal resistor, inductor, and capacitor,

where voltage is measured in volts, current in amperes, resistance in ohms, inductance in henries, and capacitance in farads. These relations assume lumped parameters and ideal component behavior.

1.3 Causality and model structure

A causal model calculates present or future outputs using current and past inputs, states, and disturbances. Physical systems generally contain energy-storage elements, which create dynamic behavior. Resistive elements dissipate energy but do not independently create state variables. An ideal mass, inertia, capacitor, or inductor normally contributes one state variable; coupled systems may require additional states.

2. Continuous-Time Modeling

A general ordinary differential-equation model is

where x is the state vector, u is the control input, d is a disturbance, y is the output, and t is time. When the equations can be solved explicitly for the state derivative,

For a time-invariant system, the functions do not explicitly depend on t.

2.1 Example: mass–spring–damper system

Consider a mass connected to a spring and viscous damper. Let q be displacement, u an applied force, m mass, b damping coefficient, and k spring stiffness. Taking positive displacement and force in the same direction,

The terms have units of newtons: mq̈ is inertial force, bq̇ is damping force, and kq is spring force. Define

Then

If position is measured, y = x₁. The model is linear provided that mass, damping, and stiffness are constant and the spring and damper obey the stated constitutive laws.

2.2 Example: armature-controlled DC motor

For an idealized separately excited DC motor, the electrical equation is

and the mechanical equation is

where v is armature voltage, i armature current, ω angular speed, τ_L load torque, R resistance, L inductance, J rotor inertia, b viscous friction coefficient, K_e back-emf constant, and K_t torque constant. In SI units, K_e and K_t have numerically compatible values when expressed consistently.

Choose the state vector

with θ̇ = ω. The state equations become

If the controlled input is voltage and the load torque is a disturbance, the model separates the two effects explicitly. This is important for disturbance rejection and feedforward compensation.

3. Linear State-Space Models

A continuous-time linear time-invariant model has the form

Here, A is the system matrix, B the input matrix, E the disturbance matrix, C the output matrix, D the direct-feedthrough matrix, and F the disturbance-output matrix. Matrix dimensions must be compatible. If there are n states, m inputs, and p outputs, then A ∈ ℝ^n×n, B ∈ ℝ^n×m, and C ∈ ℝ^p×n.

For the mass–spring–damper system with position output,

The state-space representation is not unique. Different state choices can describe the same input–output behavior, although some choices are more physically meaningful or numerically better conditioned.

3.1 Transfer functions and state-space models

For zero initial conditions, the transfer function of a linear time-invariant system is

where s is the Laplace variable and I is the identity matrix. For the mass–spring–damper system,

The transfer function is convenient for single-input, single-output frequency analysis. State-space models are generally more suitable for multiple-input, multiple-output systems, initial-condition analysis, nonlinear extensions, observers, and modern control design.

4. Discrete-Time Models

Digital controllers operate at sampling instants t_k = kT_s, where T_s is the sampling period in seconds. A discrete state-space model is

A zero-order hold assumes that the input remains constant between samples. For a continuous linear model without disturbances,

These formulas are exact under the linear model and zero-order-hold assumption. If A is nonsingular, an equivalent expression is B_d = A⁻¹(A_d − I)B, although the integral or a numerically stable matrix-exponential method is preferable when A is singular or poorly conditioned.

A simple forward-Euler approximation is

It is easy to implement but introduces approximation error and can be unstable for an otherwise stable continuous system if the sampling period is too large. Sampling should be sufficiently fast relative to the dominant system dynamics, while also considering computation, sensor delay, actuator bandwidth, and noise.

Worked Example: first-order actuator model

Consider

with constant input over a sampling period T_s = 0.1 s. The exact discrete coefficient is

For the input coefficient,

Thus

Both coefficients are dimensionally consistent: x and u have the model’s assigned units, while the numerical coefficients account for the time interval and actuator gain.

5. Nonlinear Models and Equilibria

Many mechatronic systems are nonlinear because of trigonometric geometry, friction, saturation, backlash, aerodynamic forces, magnetic effects, or variable stiffness. A nonlinear state model is

An equilibrium pair (x_e,u_e) satisfies

For an autonomous system with fixed input, the state remains at x_e if initialized there. Equilibria may be stable, unstable, or neutrally stable. Multiple equilibria can exist.

Example: inverted pendulum form

For a pendulum of length l and mass m, with angle θ measured from the upright position and applied torque u, an ideal model is

where g is gravitational acceleration and b is viscous damping. With [[MATH{x₁ = θ]] and x₂ = θ̇,

Equilibria with zero velocity satisfy x₂,e = 0 and u_e + mgl sin x_1,e = 0. If u_e = 0, upright and downward orientations are equilibrium positions, depending on the angular convention. The equilibrium equation must always be interpreted with the chosen sign convention.

6. Linearization Around an Operating Point

A nonlinear model can be approximated locally by expanding it around an equilibrium. Define perturbations

The first-order Taylor approximation gives

where

The approximation neglects second- and higher-order terms. It is valid locally when perturbations are sufficiently small and the functions are differentiable near the operating point. It does not describe behavior accurately near discontinuities, hard saturation, impacts, or far from the equilibrium.

For the pendulum model, at zero input and angle θ_e,

The sign of cos θ_e changes between upright and downward configurations, producing different local dynamics. This illustrates why a controller designed around one operating point may fail around another.

7. Stability, Controllability, and Observability

7.1 Stability of linear models

For a continuous-time linear system, asymptotic stability requires all eigenvalues of A to have strictly negative real parts. For a discrete-time system, all eigenvalues must lie strictly inside the unit circle. These are model-based local or global conclusions only within the assumptions of the representation.

7.2 Controllability

A state-space pair (A,B) is controllable if an admissible input can move the system from any initial state to any final state in finite time under the linear model. The Kalman controllability matrix is

The system is controllable if rank(𝒞) = n. Lack of controllability means that some internal modes cannot be influenced by the chosen actuator. In practice, a mathematically controllable system may still be difficult to control if it is poorly conditioned, actuator-limited, or highly sensitive to uncertainty.

7.3 Observability

A pair (A,C) is observable if the initial state can be reconstructed from measured outputs over a finite interval. The observability matrix is

where semicolons denote vertical stacking. The system is observable if rank(𝒪) = n. Unobservable modes cannot be reconstructed by an ideal observer using the selected sensors. Sensor placement therefore affects not only measurement quality but also the feasibility of state estimation.

For a single-input mass–spring–damper system with position measurement,

Its determinant is −1/m², which is nonzero for finite nonzero mass; hence the pair is controllable. The observability matrix is

and is full rank. Thus position measurements, together with the known model, are theoretically sufficient to reconstruct both position and velocity.

8. Uncertainty, Disturbances, and Model Limitations

No engineering model is exact. Important distinctions are:

• Parameter uncertainty: mass, resistance, friction, or stiffness is not known precisely.
• Disturbance: an external influence such as load torque or vibration.
• Measurement noise: random error in a sensor signal.
• Unmodeled dynamics: neglected flexible modes, delays, quantization, or parasitic effects.
• Structural uncertainty: the assumed equations do not represent the true physical structure.

A parameterized model may be written as

where ρ is a vector of uncertain parameters and w represents process disturbance. Measurement may be modeled as

where v is measurement noise. These terms should not be used interchangeably: process disturbance changes the physical state, whereas measurement noise corrupts the reported observation.

Common limitations include:

1. Linearization is local and may ignore large-signal nonlinearities.
2. Constant-parameter assumptions may fail with temperature, wear, or changing payload.
3. Rigid-body assumptions may neglect resonances and structural flexibility.
4. Coulomb friction and backlash can be nonsmooth, making derivative-based linearization inadequate at zero velocity.
5. Sensors and actuators introduce delays, saturation, dead zones, resolution limits, and rate limits.
6. Identification data may cover only a narrow operating range.

Best practice is to validate a model against independent experimental data. Useful tests include step responses, frequency-response measurements, transient trajectories, residual analysis, and tests at multiple loads and operating points. A model should be judged by its intended use, not by complexity alone.

9. Applied Case: Servo Positioning Axis

Background

Consider an illustrative industrial servo axis consisting of a motor, gearbox, lead screw, carriage, encoder, and digital drive. The objective is to position the carriage accurately despite payload variation and friction.

Problem

A rigid-body model predicts the position response well at low speed but the actual axis exhibits oscillation near a particular frequency. The controller performs acceptably with no payload but becomes sluggish and overshoots when the payload changes.

Analysis

A first model uses motor position, motor speed, and armature current as states. It includes motor resistance, inductance, torque constant, inertia, and viscous friction. The gearbox and screw convert motor rotation into carriage translation. This model supports initial controller design and current-loop analysis.

Comparison with experimental data reveals a resonance. The discrepancy indicates that the rigid-body assumption is insufficient. A refined model adds a flexible shaft or coupling, producing an additional relative displacement and velocity state. Payload uncertainty changes the effective inertia and resonance frequency. Encoder quantization and sampling delay are added to the measurement model.

Controllability analysis confirms that the actuator can influence the dominant rigid and flexible modes, but the flexible mode is weakly actuated at some frequencies. Observability analysis shows that the motor encoder does not directly measure carriage deformation; an additional load-side sensor would improve estimation if the application requires high bandwidth.

Approach and solution

The engineering workflow is:

1. Identify the control objective and acceptable error.
2. Derive a low-order physical model.
3. Estimate parameters from datasheets and experiments.
4. Validate the model using unloaded and loaded tests.
5. Add flexible dynamics only when residuals show systematic error.
6. Discretize the validated model using the actual sampling period.
7. Include saturation, delay, quantization, and sensor noise in simulation.
8. Design and test the controller first in simulation, then at reduced speed and finally under full operating conditions.

A gain schedule or robust controller may be required when payload changes substantially. Feedforward can compensate predictable motion demands, while feedback rejects disturbances and modeling errors. Safety limits must independently constrain position, speed, current, and acceleration.

Outcome and learning points

The refined model predicts the resonance and explains why a controller tuned for one payload is unsatisfactory for another. The case demonstrates that model development is iterative: begin with physics-based structure, compare predictions with measurements, identify the source of error, and increase complexity only when justified. It also demonstrates that controllability and observability are practical design questions involving actuator and sensor placement, not merely abstract matrix tests.

10. Professional Practice, Safety, and Contemporary Developments

Models are used in digital twins, predictive maintenance, autonomous machines, collaborative robots, and model-predictive control. Data-driven and machine-learning models can complement physics-based models, especially when friction, wear, or complex contact is difficult to formulate. However, data-driven models may extrapolate poorly outside the training region and may obscure physical constraints. Hybrid modeling combines conservation laws with learned residuals or uncertain parameters.

For safety-critical systems, the model must support hazard analysis, traceability, verification, and validation. Engineers should document assumptions, parameter sources, uncertainty ranges, operating domains, and software versions. A simulation result is not evidence of safety unless the model, numerical method, test coverage, and validation process are appropriate. Ethical practice requires avoiding unjustified claims of accuracy and clearly communicating conditions under which the model may fail.

Chapter Summary

Dynamic modeling converts physical relationships into equations that support analysis and control. Continuous-time models are derived from conservation laws and constitutive relations. State-space form represents internal energy-storage variables and is especially useful for multivariable systems, observers, and nonlinear extensions. Discrete-time models describe the behavior seen by digital controllers; exact zero-order-hold conversion uses the matrix exponential, while Euler methods are approximations.

Nonlinear models are expressed as ẋ = f(x,u). Equilibria satisfy f(x_e,u_e)=0, and local linear models are obtained from Jacobian matrices evaluated at the operating point. Controllability concerns whether inputs can influence all states, while observability concerns whether states can be reconstructed from outputs. Both are tested using rank conditions for linear models.

Uncertainty, disturbances, noise, delays, saturation, flexible modes, and changing parameters limit model validity. Reliable engineering practice combines physical derivation, parameter identification, independent validation, uncertainty representation, and careful documentation. The best model is not necessarily the most detailed one; it is the simplest model that is sufficiently accurate for the intended decision or control task.

Key Terms

State: A variable set containing the information needed to predict future system behavior.

State vector: An ordered vector containing the system states.

Input: A manipulated quantity applied to a system.

Output: A measured or regulated system quantity.

Disturbance: An external influence not directly selected by the controller.

Parameter: A quantity characterizing a component or physical relationship.

Continuous-time model: A model described by differential equations.

Discrete-time model: A model defined at separated sampling instants.

State-space model: A first-order vector representation of system dynamics and outputs.

Equilibrium: A state and input pair for which the state derivative is zero.

Linearization: A local first-order approximation of a nonlinear model.

Jacobian: A matrix of first partial derivatives used in linearization.

Controllability: The ability to move system states using the available inputs.

Observability: The ability to infer system states from measured outputs.

Zero-order hold: An assumption that an input remains constant between samples.

Model uncertainty: Imperfect knowledge of parameters or system structure.

Unmodeled dynamics: Physical behavior omitted from the chosen model.

Validation: Comparison of model predictions with independent measurements.

Review Questions

5 Short Answer Questions

1. What distinguishes a state variable from an output variable?
2. State the continuous-time linear state-space equations.
3. What condition defines an equilibrium of a nonlinear system?
4. Why is a sampling period required in a digital control system?
5. Distinguish measurement noise from process disturbance.

5 Long Answer Questions

1. Derive the state-space model of a mass–spring–damper system and explain the physical meaning of each state equation.
2. Derive the state equations of an armature-controlled DC motor, identifying the control input and a possible disturbance input.
3. Explain exact zero-order-hold discretization and compare it with forward-Euler approximation.
4. Explain how Jacobian linearization is performed around an equilibrium and discuss its limitations.
5. Explain controllability and observability, including the rank tests and their practical significance in actuator and sensor selection.

3 Analytical / Application Questions

1. A motor-driven positioning system has a payload that changes substantially during operation. Propose a modeling and validation strategy that addresses parameter uncertainty, sampling, actuator saturation, and changing operating points.
2. A nonlinear pendulum model has two equilibrium positions. Explain how you would determine local stability and construct separate linear models for controller design around each position.
3. A measured servo response contains a resonance absent from a rigid-body model. Diagnose possible causes, propose model refinements, and explain how you would decide whether an additional sensor is required.

Multiple Choice Questions with Answers

Question 1

Which quantity is normally selected as a state variable because it stores energy in an ideal electrical component?

A. Resistance

B. Capacitor voltage

C. Switch command

D. Output reference

Correct Answer: B. Capacitor voltage

Question 2

For a mass–spring–damper system, which equation is correct under the usual linear viscous-damping assumptions?

A. m q̈ + b q̇ + kq = u

B. m q̇ + bq̈ + kq = u

C. m q + b q̇ + kq̈ = u

D. m q̈ + k q̇ + bq = u

Correct Answer: A. m q̈ + b q̇ + kq = u

Question 3

Under a zero-order hold, what does the input do between two sampling instants?

A. It becomes zero

B. It varies sinusoidally

C. It remains constant

D. It is replaced by noise

Correct Answer: C. It remains constant

Question 4

For a continuous-time linear system, asymptotic stability requires which eigenvalue condition?

A. All eigenvalues have positive real parts

B. All eigenvalues have zero real parts

C. All eigenvalues have strictly negative real parts

D. All eigenvalues have magnitude greater than one

Correct Answer: C. All eigenvalues have strictly negative real parts

Question 5

A nonlinear equilibrium pair (x_e,u_e) satisfies:

A. f(x_e,u_e)=1

B. f(x_e,u_e)=0

C. h(x_e,u_e)=0 only

D. x_e=u_e

Correct Answer: B. f(x_e,u_e)=0

Question 6

The linearization matrix A of ẋ=f(x,u) is obtained from:

A. ∂f/∂u evaluated at the operating point

B. ∂h/∂x evaluated at the operating point

C. ∂f/∂x evaluated at the operating point

D. The inverse of f

Correct Answer: C. ∂f/∂x evaluated at the operating point

Question 7

A pair (A,B) is controllable when the Kalman controllability matrix has:

A. Rank zero

B. Rank equal to the number of inputs

C. Rank equal to the number of outputs

D. Full rank equal to the number of states

Correct Answer: D. Full rank equal to the number of states

Question 8

Observability concerns whether:

A. Inputs can be limited by an actuator

B. States can be reconstructed from outputs

C. A model has no disturbances

D. A system has no nonlinearities

Correct Answer: B. States can be reconstructed from outputs

Question 9

For the scalar model ẋ=−2x+4u with T_s=0.1 s, the exact zero-order-hold state coefficient is closest to:

A. 0.1353

B. 0.5000

C. 0.8187

D. 1.2000

Correct Answer: C. 0.8187

Question 10

Which statement best describes model validation?

A. Increasing model complexity without measurements

B. Comparing model predictions with independent experimental data

C. Removing all disturbances from the physical system

D. Assuming manufacturer parameters are exact

Correct Answer: B. Comparing model predictions with independent experimental data