Modeling, Simulation, and Digital Engineering Research

Introduction

Mechatronic systems combine mechanical structures, electrical power, sensors, actuators, embedded computation, communication networks, and control algorithms. Their behavior is therefore difficult to predict from intuition or from isolated component tests. Modeling and simulation provide a systematic way to represent such systems before, during, and after physical development.

In engineering research, a model is not merely a drawing or a computer program. It is a purposeful representation of a real system, developed for a defined question. A model may be used to predict motion, estimate energy consumption, design a controller, identify faults, compare alternatives, or determine whether a prototype satisfies requirements. Simulation is the execution of that model under specified inputs, parameters, initial conditions, and numerical methods.

Digital engineering extends this process across the product life cycle. It links requirements, computer-aided design, finite-element analysis, control models, test data, manufacturing information, and operational data. A digital twin is a particularly important form of digital engineering: it is a digital representation connected to a specific physical asset or process through data exchange and maintained for a defined purpose.

This chapter develops the research methods needed to create, calibrate, test, and responsibly use mechatronic models.

Learning Objectives

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

1. Formulate mathematical models of mechanical, electrical, and electromechanical systems.
2. Distinguish first-principles modeling, system identification, and hybrid modeling.
3. Select suitable numerical simulation methods and interpret their limitations.
4. Design hardware-in-the-loop (HIL) experiments.
5. Explain the architecture and lifecycle of digital twins.
6. Perform sensitivity analysis and use it to prioritize measurements and design changes.
7. Verify computational implementation and validate a model against physical evidence.
8. Evaluate uncertainty, reproducibility, ethics, cybersecurity, and practical research risks.

1. The Role of Models in Mechatronics Research

A useful model has four characteristics: a stated purpose, a defined boundary, assumptions that can be examined, and a connection to evidence. No model represents every feature of reality. A model is adequate only relative to a task and an acceptable error tolerance.

A common engineering representation is the state-space model:

Here, x is the state vector, u is the input vector, y is the measured-output vector, p is the parameter vector, and t is time. The function f describes state evolution and g maps states and inputs to outputs. For a linear time-invariant approximation, these become:

The matrices A, B, C, and D must have compatible dimensions. Their numerical values and units depend on how states, inputs, and outputs are defined. Linear models are local approximations unless the underlying physical system is genuinely linear over the operating range.

The modeling process normally includes:

1. Define the research question and performance measures.
2. Establish the system boundary and interfaces.
3. Identify physical laws, parameters, disturbances, and measurable variables.
4. Select the required fidelity.
5. Implement the model and numerical solver.
6. Verify the implementation.
7. Validate predictions using independent evidence.
8. Quantify uncertainty and document assumptions.

Levels of Model Fidelity

A low-order model may be suitable for controller design, while a high-fidelity finite-element model may be needed to investigate structural stress. Increasing fidelity generally increases computational cost, parameter requirements, and opportunities for modeling error. A model should therefore be as simple as possible but sufficiently accurate for its intended decision.

2. Mathematical Modeling of Mechatronic Systems

2.1 Mechanical Systems

For a translational mass–damper–spring system, Newton’s second law gives:

where m is mass in kilograms, c 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 model assumes small displacement, linear spring behavior, viscous damping, and one-dimensional motion.

For a rotational system:

where J is rotational inertia in kilogram-metres squared, b is rotational damping in newton-metres per radian per second, k_θ is torsional stiffness in newton-metres per radian, θ is angular displacement in radians, and τ is applied torque in newton-metres.

Introducing states x₁ = x and x₂ = ẋ gives:

This first-order form is convenient for numerical integration and control analysis.

2.2 Electrical Systems

For a series resistor–inductor–capacitor circuit, Kirchhoff’s voltage law gives:

and the capacitor relation is:

where L is inductance in henries, R is resistance in ohms, C is capacitance in farads, i is current in amperes, v_C is capacitor voltage in volts, and v_s is source voltage in volts. These equations assume lumped components and negligible parasitic effects unless such effects are explicitly added.

2.3 Electromechanical Coupling

A simplified DC motor model illustrates coupling between electrical and mechanical domains:

Here, L_a and R_a are armature inductance and resistance, i_a is armature current, v_a is applied armature voltage, K_e is back-emf constant, ω is angular speed, J is rotor inertia, b is viscous friction, K_t is torque constant, and τ_L is load torque. In SI units, K_e and K_t have numerically compatible values when the motor conventions are consistently defined. Friction, saturation, dead zones, gear ratios, and current limits may need to be added for a realistic model.

2.4 Conservation Laws and Bonding Domains

A robust model often begins with conservation of mass, momentum, charge, and energy. Power consistency is especially important in mechatronics. Mechanical power is P_m = τω, while electrical power is P_e = vi. A model that creates energy without a source, or dissipates energy without a physical mechanism, may contain an implementation error or an invalid assumption.

Dimensional analysis is a simple verification tool. In the mass–spring equation, mẍ, cẋ, kx, and F must all have units of newtons. If one term has units of joules or metres, the formulation is inconsistent.

3. System Identification and Parameter Estimation

First-principles models use physical laws and parameters such as mass, resistance, and stiffness. System identification estimates a model from measured inputs and outputs. It is valuable when parameters are uncertain, nonlinearities are difficult to describe, or the complete physical structure is unknown.

A typical identification experiment requires:

• an informative input signal;
• synchronized and sufficiently sampled measurements;
• known sensor characteristics and calibration;
• safe operating limits;
• separation of estimation and validation data;
• documentation of filtering, preprocessing, and discarded observations.

For a discrete linear model, one common representation is:

where y_k is the measured output, φ_k is a vector of measured regressors, θ is the unknown parameter vector, and ε_k is measurement and modeling error. Ordinary least squares estimates θ by minimizing the sum of squared residuals:

If the matrix Φ, whose rows are φ_k^T, has full column rank, the least-squares estimate is:

The inverse formula applies only when Φ^TΦ is nonsingular and the data are appropriately scaled. In numerical software, QR or singular-value decomposition is generally more stable than explicitly computing the inverse.

Persistent Excitation

An input must excite the dynamics of interest. A constant input cannot reliably identify a dynamic system. Steps, pseudorandom binary sequences, chirps, and multisine signals can be useful, but the choice depends on bandwidth, safety, actuator limits, and the expected model structure. The experiment must cover the operating conditions in which the model will be used.

Model Structure and Overfitting

A model with too few parameters has high bias and cannot represent important behavior. A model with too many parameters may fit noise and perform poorly on new data. Validation data must not be used to select parameters repeatedly without accounting for this dependence. Regularization, cross-validation, and physically meaningful parameter constraints can improve generalization.

4. Numerical Simulation

A continuous model must be converted into numerical values at discrete time points. For the initial-value problem:

the forward Euler method is:

where h is the time step. Euler’s method is easy to implement but may be inaccurate or unstable for stiff systems.

A fourth-order Runge–Kutta method uses four slope estimates:

For smooth problems with adequate step size, this method has fourth-order global accuracy. Its error advantage does not eliminate the need for convergence checks or physical validation.

Numerical Issues

A simulation can fail because of poor step size, discontinuities, algebraic constraints, stiff dynamics, ill-conditioned equations, or incorrect initial conditions. Adaptive solvers vary the time step according to an error estimate. A stiff system may require an implicit solver because explicit methods would need impractically small steps for stability.

A convergence study repeats a simulation with successively smaller step sizes or finer spatial meshes. If the reported quantity changes significantly, the original discretization is not yet adequate. Numerical convergence is verification; it does not prove that the physical model is correct.

5. Verification, Validation, and Uncertainty

Verification asks, “Was the model implemented correctly?” Validation asks, “Is the model an adequate representation of the real system for the intended use?” These are different questions.

Verification Methods

Useful verification practices include:

• unit and dimensional checks;
• conservation-law checks;
• comparison with analytical solutions or limiting cases;
• code review and automated tests;
• mesh and time-step convergence studies;
• solver tolerance studies;
• independent implementation of selected equations;
• tests using known benchmark problems.

Validation Methods

Validation requires comparison with physical or trusted reference evidence. A validation plan should specify the operating conditions, measured outputs, uncertainty, acceptance criteria, and whether the data were used during calibration. Agreement at one operating point does not establish validity across all speeds, loads, temperatures, or environmental conditions.

A basic error metric is the root-mean-square error:

where y_i is the measured value, ŷ_i is the predicted value, and N is the number of observations. RMSE has the same units as the output and penalizes large errors. It should be accompanied by plots, residual analysis, and engineering tolerances because a single metric can hide systematic error.

If measurements have known standard uncertainties u_i, a normalized residual may be written as:

The interpretation depends on how u_i was defined and whether model discrepancy is included. Uncertainty should not be treated as a purely statistical quantity: sensor calibration, unmodeled friction, parameter variation, and environmental effects are also relevant.

6. Sensitivity Analysis

Sensitivity analysis determines how changes in inputs or parameters affect outputs. It helps researchers identify influential parameters, design experiments, simplify models, and allocate measurement effort.

A local normalized sensitivity of output y to parameter p is:

This dimensionless measure is meaningful when y and p are nonzero and the response is sufficiently smooth near the operating point. A large magnitude indicates that relative uncertainty in p can strongly influence relative uncertainty in y.

For a finite-difference estimate:

The perturbation ∆p should be large enough to exceed numerical noise but small enough to represent a local derivative. For nonlinear systems, local sensitivity may vary substantially with time and operating condition.

Global methods vary several uncertain inputs over prescribed ranges. Screening methods, variance-based methods, and Monte Carlo sampling can reveal interactions that local one-at-a-time analysis misses. Correlated parameters require special care; independently varying physically correlated quantities can produce impossible system states.

7. Hardware-in-the-Loop Testing

Hardware-in-the-loop testing connects real hardware to a real-time simulation. For example, a motor controller may receive simulated position and speed signals while its commanded voltage is applied to a real or simulated power interface. The real-time plant model must execute within each fixed sampling interval.

A HIL setup commonly contains:

1. a real-time computational target;
2. a plant model;
3. input/output interfaces;
4. the hardware under test;
5. safety interlocks and fault injection;
6. data acquisition and time synchronization.

HIL testing exposes interface problems, timing faults, sensor scaling errors, saturation behavior, and controller responses to faults without requiring a complete physical plant. However, latency, quantization, signal conditioning, and electrical interface dynamics must be represented or characterized. A HIL result is not automatically equivalent to a full physical test.

The test plan should define timing deadlines, sample rates, tolerable latency, fault conditions, pass/fail criteria, and a safe shutdown procedure. Real-time overruns must be logged because a model that is accurate but cannot meet its deadline is unsuitable for real-time use.

8. Digital Engineering and Digital Twins

Digital engineering creates a connected technical information environment rather than a collection of disconnected files. A typical workflow links requirements to architecture, CAD geometry, simulation models, software versions, test records, and operational data.

A digital twin contains at least:

• a physical asset or process;
• a digital representation;
• a data connection between them;
• a defined purpose and lifecycle;
• mechanisms for version control, uncertainty management, and access control.

A static design model is not necessarily a digital twin. The term should be used carefully, because a twin’s credibility depends on traceability, update frequency, model scope, and validation evidence. Digital twins may support predictive maintenance, virtual commissioning, energy optimization, operator training, and fleet-level analysis.

Figure: A digital twin links a physical asset, a digital model, evidence, and engineering decisions in a feedback cycle.

Interoperability is a major challenge. Data schemas, units, coordinate frames, timestamps, model versions, and interface definitions must be explicit. A twin that combines incompatible timestamps or undocumented coordinate systems can produce confident but incorrect conclusions.

9. Worked Example: Identifying a Motor–Load Model

Consider a motor driving a rotational load. Suppose the electrical dynamics are much faster than the mechanical dynamics, so the armature inductance is neglected for a low-speed mechanical study. The reduced model is:

Assume a measured constant current of 1.5 A produces a steady speed of 30 rad/s, and the torque constant is K_t = 0.20 N·m/A. At steady state, dω/dt = 0, so:

Therefore:

The units are consistent: torque divided by angular speed gives (N·m)/(rad/s) = N·m·s/rad; radians are dimensionless in SI conventions. This estimate is valid only if load torque is negligible or separately accounted for, friction is approximately viscous, and the speed is truly at steady state. If a constant external load torque exists, the equation becomes bω = K_ti_a − τ_L, and ignoring τ_L would overestimate or underestimate b depending on its direction.

A second experiment can estimate J from the initial acceleration. If i_a = 2 A, ω = 0 at the initial instant, b = 0.010 N·m·s/rad, and the measured initial acceleration is 20 rad/s²:

The researcher should then validate the model using a different current profile and compare the predicted speed trajectory with measured speed. Agreement in the steady-state points alone would not validate the transient inertia estimate.

10. Illustrative Case: Digital Twin for a Conveyor Servo Axis

Background

The following is a fictional illustrative case based on common industrial practice. A manufacturing laboratory operates a servo-driven conveyor axis used to position components for robotic assembly. Intermittent positioning errors occur after extended operation, but replacing the drive and motor without diagnosis would be expensive.

Problem

The research team must determine whether the errors arise from mechanical compliance, encoder noise, controller timing, temperature-dependent friction, or communication delay. The team also needs a safe method to test fault responses.

Analysis and Approach

The team defines the model boundary from the motor shaft to the conveyor carriage. A multibody model represents inertia, belt compliance, and damping. A motor model represents torque production and current limits. Temperature-dependent friction is represented by a parameter identified from experiments at three temperatures. Encoder measurements are time-stamped and synchronized with drive commands.

First, unit tests verify the model against known spring–mass responses. A time-step convergence study shows that reducing the step from 1 ms to 0.5 ms changes peak positioning error by less than the pre-established 1% numerical tolerance. Next, parameters are estimated from unloaded motion and validated under a different payload.

A HIL test connects the real motion controller to the real-time model. Artificial encoder dropouts and communication delays are injected within safe limits. The controller’s fault handling is assessed using predefined pass/fail criteria. Finally, operational data are used to update the digital twin, but only after checks for sensor plausibility and timestamp integrity.

Outcome

The analysis indicates that thermal changes increase friction and that a timing delay amplifies the error during rapid reversals. The team modifies the compensation schedule, improves timestamp monitoring, and changes a maintenance threshold. A physical confirmation test shows improved positioning over the required temperature range.

Learning Points

1. A digital twin is useful only when its purpose, interfaces, and validation conditions are explicit.
2. Identification experiments should separate calibration data from validation data.
3. HIL testing can expose timing and fault-response problems before destructive physical tests.
4. Model updates must be traceable; an unexplained parameter change reduces confidence.
5. A successful model is not necessarily universally valid outside its tested load, speed, and temperature envelope.

11. Research Quality, Ethics, and Governance

Model-based research can create an illusion of precision. Graphs with smooth curves do not prove accuracy, and machine-learning predictions do not remove the need for physical reasoning. Researchers should report assumptions, parameter sources, software versions, solver settings, calibration procedures, validation conditions, uncertainty, and known failure modes.

Safety is central when models influence actuators. A digital model should not be allowed to command hazardous equipment without independent interlocks, bounded outputs, emergency shutdown, and staged testing. Cybersecurity is also a safety issue: unauthorized alteration of a model, firmware, sensor stream, or calibration file can produce unsafe decisions.

Operational data may contain personal, proprietary, or safety-sensitive information. Access controls, retention policies, anonymization where appropriate, and audit trails should be established. If artificial intelligence is used to identify faults or estimate parameters, its training data, performance across operating conditions, false-alarm costs, and human oversight must be documented. A model should support engineering judgment rather than conceal uncertainty.

Reproducibility requires more than publishing a final plot. A credible research package includes data dictionaries, preprocessing scripts, model equations, parameter files, test protocols, and version identifiers. Units and coordinate conventions should be machine-readable where possible.

12. Best-Practice Workflow

A disciplined modeling study can follow this sequence:

1. State the decision or research question.
2. Define outputs, acceptable errors, and operating envelope.
3. Select the simplest model capable of answering the question.
4. Record assumptions, units, boundaries, and parameter sources.
5. Design informative and safe experiments.
6. Implement automated verification tests.
7. Estimate parameters using calibration data.
8. Validate against independent measurements.
9. Perform sensitivity and uncertainty analysis.
10. Use HIL or staged physical tests before full deployment.
11. Document limitations and update triggers.
12. Archive data, code, model versions, and evidence.

Contemporary Developments

Current practice increasingly combines reduced-order physics models with machine-learning components. Physics-informed learning can improve prediction where data are limited, while residual learning can model effects omitted from a first-principles model. Edge computing enables local digital-twin updates with lower latency, and cloud platforms support fleet-level analysis. These developments do not remove fundamental requirements for observability, calibration, validation, cybersecurity, and explainability.

The emerging research emphasis is therefore not simply on building larger models. It is on trustworthy models: models whose assumptions, uncertainty, provenance, and domain of validity are visible to the engineers who use them.

Chapter Summary

Modeling and simulation provide controlled representations of mechatronic systems for analysis, design, prediction, and research. State-space equations offer a general framework, while mechanical, electrical, and electromechanical laws provide physically grounded models. System identification complements first-principles modeling when parameters or mechanisms are uncertain.

Numerical methods convert continuous equations into computable predictions, but step-size, stiffness, discretization, and solver errors must be examined. Verification checks implementation correctness; validation evaluates adequacy against physical evidence. Sensitivity analysis identifies influential parameters and supports experiment design and uncertainty management.

Hardware-in-the-loop testing connects real hardware to a real-time simulation and is valuable for timing, fault, and controller testing. Digital twins connect physical assets, digital representations, data, and engineering decisions throughout the lifecycle. Their usefulness depends on traceability, interoperability, cybersecurity, and evidence-based updating. Professional practice requires explicit assumptions, independent validation, safe experimentation, ethical data governance, and reproducible documentation.

Key Terms

Model: A purposeful representation of a real system.

Simulation: Execution of a model under specified conditions.

State variable: A variable whose value, together with inputs, describes system evolution.

State-space model: A first-order differential or difference-equation representation using states and outputs.

System identification: Estimation of model structure or parameters from measured data.

Parameter: A quantity that characterizes a model, such as mass, stiffness, or resistance.

Persistent excitation: Input variation sufficient to reveal the dynamics being identified.

Numerical integration: Approximation of a continuous-time solution using discrete computations.

Stiff system: A system requiring special numerical treatment because stability restricts explicit step size.

Verification: Demonstration that a model or implementation has been constructed correctly.

Validation: Evaluation of whether a model adequately represents a real system for a defined purpose.

Calibration: Adjustment of model parameters using selected evidence.

Sensitivity analysis: Study of how outputs change with parameters or inputs.

Hardware-in-the-loop: Real hardware operating with a real-time simulated environment.

Digital engineering: Connected management of engineering information, models, data, and decisions.

Digital twin: A data-connected digital representation of a specific physical asset or process.

Model discrepancy: Error arising because the model cannot represent all relevant physical behavior.

Uncertainty: Imperfect knowledge of parameters, measurements, conditions, or model form.

Interoperability: Ability of systems and models to exchange and correctly interpret information.

Traceability: Ability to link requirements, data, model versions, tests, and decisions.

Review Questions

Short Answer Questions

1. What distinguishes model verification from model validation?
2. Why must a system-identification experiment use an informative input?
3. State two limitations of a low-order lumped-parameter model.
4. What is the purpose of a time-step convergence study?
5. Name four essential elements of a digital twin.

Long Answer Questions

1. Explain how a mechanical mass–spring–damper system can be converted into state-space form, defining all variables and assumptions.
2. Compare first-principles modeling, system identification, and hybrid modeling in a mechatronic research project.
3. Describe a complete verification and validation plan for a motor-control simulation.
4. Explain the architecture, benefits, and limitations of hardware-in-the-loop testing.
5. Discuss how uncertainty, cybersecurity, and traceability affect the credibility of a digital twin.

Analytical / Application Questions

1. A motor produces 0.30 N·m of torque at 25 rad/s in steady operation with negligible external load. Estimate the viscous damping coefficient and state the assumptions required.
2. A model agrees with measured output at one speed but fails during rapid reversals. Propose a diagnostic sequence using sensitivity analysis, new experiments, and validation data.
3. Design a HIL experiment for a robotic actuator that must detect encoder loss and communication delay. Specify the hardware, simulated elements, timing requirements, safety measures, and pass/fail evidence.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes verification from validation?

A. Verification checks implementation correctness; validation checks adequacy against the real system.

B. Verification uses physical tests; validation uses only source-code inspection.

C. Verification estimates parameters; validation selects the numerical solver.

D. Verification and validation are identical terms.

Correct Answer: A. Verification checks implementation correctness; validation checks adequacy against the real system.

Question 2

For the model mẍ + cẋ + kx = F, what are the units of the stiffness k?

A. kg

B. N·s/m

C. N/m

D. N·m

Correct Answer: C. N/m

Question 3

In ordinary least squares, why should QR or singular-value decomposition generally be preferred to explicitly computing (Φ^TΦ)^-1?

A. They remove the need for measured data.

B. They are usually more numerically stable.

C. They guarantee a physically correct model.

D. They make all parameters dimensionless.

Correct Answer: B. They are usually more numerically stable.

Question 4

What is the main purpose of persistent excitation in system identification?

A. To keep the actuator completely stationary

B. To reveal the dynamics of interest in the measured data

C. To eliminate all sensor noise

D. To guarantee that the system is linear

Correct Answer: B. To reveal the dynamics of interest in the measured data

Question 5

A digital twin is best described as:

A. Any CAD drawing of a product

B. A simulation that has never been compared with measurements

C. A data-connected digital representation of a specific physical asset or process

D. A replacement for all physical testing

Correct Answer: C. A data-connected digital representation of a specific physical asset or process

Question 6

If K_t = 0.20 N·m/A, i = 1.5 A, and steady speed is 30 rad/s, with negligible load torque, what is b in bω = K_ti?

A. 0.001 N·m·s/rad

B. 0.010 N·m·s/rad

C. 0.10 N·m·s/rad

D. 10 N·m·s/rad

Correct Answer: B. 0.010 N·m·s/rad

Question 7

Which activity is primarily a verification activity?

A. Comparing predicted temperature with an independent physical test

B. Checking model dimensions and units

C. Testing a controller on a new payload

D. Measuring positioning error in production

Correct Answer: B. Checking model dimensions and units

Question 8

What is a major advantage of hardware-in-the-loop testing?

A. It eliminates all interface latency.

B. It permits safe testing of timing and fault responses before full physical testing.

C. It proves that the plant model is universally valid.

D. It removes the need for safety interlocks.

Correct Answer: B. It permits safe testing of timing and fault responses before full physical testing.

Question 9

A large magnitude of local normalized sensitivity generally indicates that:

A. The output is unaffected by the parameter.

B. Relative uncertainty in the parameter may strongly affect relative output uncertainty.

C. The numerical solver is definitely unstable.

D. The parameter must be measured in metres.

Correct Answer: B. Relative uncertainty in the parameter may strongly affect relative output uncertainty.

Question 10

Why should calibration data and validation data normally be separated?

A. To prevent the validation from being used as independent evidence after fitting

B. To increase the number of fitted parameters automatically

C. To make all residuals equal to zero

D. To avoid defining model assumptions

Correct Answer: A. To prevent the validation from being used as independent evidence after fitting