11. Digital Engineering, Simulation and Digital Twins

Introduction and Context

Mechanical-system design has progressively moved from drawing boards and isolated calculations toward integrated digital environments. Computer-aided design (CAD), computer-aided engineering (CAE), finite-element analysis (FEA), computational fluid dynamics (CFD), multibody dynamics (MBD), product lifecycle management (PLM), and model-based systems engineering (MBSE) now form an interconnected design and decision framework. The purpose of this framework is not merely to produce a three-dimensional model. It is to represent requirements, geometry, physics, manufacturing information, operating conditions, performance evidence, and lifecycle data in a controlled digital form.

A digital model can be used to predict deformation, stress, vibration, heat transfer, fluid flow, fatigue, motion, energy consumption, maintainability, and system-level behaviour before physical prototypes are built. However, simulation results are not automatically reliable. Their value depends on the quality of assumptions, material data, boundary conditions, discretization, numerical methods, verification, validation, and engineering interpretation.

Digital twins extend this idea by linking a digital representation to a physical asset or process through data, models, and feedback. A digital twin may support design, commissioning, operation, maintenance, diagnosis, optimization, and eventual retirement. It is therefore more than a static CAD model or a simulation file. It is a managed relationship between a physical object, its digital representation, and the data exchanged between them.

For postgraduate mechanical engineers, the central competence is not simply knowing how to operate software. It is the ability to select an appropriate model, establish its credibility, integrate models across disciplines, manage engineering data, and make defensible decisions under uncertainty.

Learning Objectives

After studying this chapter, the learner should be able to:

1. Explain the relationships among CAD, CAE, FEA, CFD, MBD, MBSE, PLM, and digital twins.

2. Formulate a simulation problem by defining objectives, assumptions, physics, inputs, outputs, and acceptance criteria.

3. Distinguish verification from validation and assess the credibility of a computational model.

4. Interpret mesh, convergence, sensitivity, uncertainty, and correlation results.

5. Describe how digital threads and PLM support configuration and lifecycle management.

6. Apply an integrated workflow to a mechanical-system design problem.

7. Evaluate the technical, organizational, cybersecurity, ethical, and governance risks of digital engineering.

8. Identify current developments involving automation, reduced-order models, artificial intelligence, and cloud-based engineering.

From Digital Models to Engineering Decisions

CAD as the geometric and product-definition foundation

CAD represents the geometry, topology, dimensions, tolerances, assemblies, materials, and sometimes manufacturing information associated with a product. Parametric CAD models encode design intent through constraints, features, relations, and design variables. This allows an engineer to change a principal dimension and propagate the change through an assembly or drawing.

A CAD model, however, is not necessarily a valid analysis model. Detailed fillets, fasteners, threads, small holes, gaps, and cosmetic features may create unnecessary computational cost or numerical difficulty. Conversely, removing a feature that controls stress concentration, flow separation, contact, or heat transfer may invalidate the analysis. Model simplification must therefore preserve the physics relevant to the question being asked.

CAE and multiphysics reasoning

CAE includes computational tools used to evaluate engineering behaviour. Structural analysis may address static strength, buckling, fracture, fatigue, impact, or vibration. Thermal analysis may involve conduction, convection, radiation, and phase change. CFD predicts fluid motion, pressure, temperature, turbulence, and species transport. MBD predicts the motion and force transmission of interconnected rigid or flexible bodies.

Many industrial systems are multiphysics systems. A brake involves structural deformation, frictional contact, heat generation, thermal expansion, and possibly thermoelastic instability. A pump involves rotating machinery, fluid pressure, structural stress, vibration, cavitation, and acoustic effects. The engineer must decide whether these effects can be treated sequentially or require two-way coupling. Sequential coupling is computationally simpler, while strongly coupled problems may require iterative exchange of data between solvers.

Model-based systems engineering

MBSE applies formal models to represent system requirements, functions, architecture, interfaces, behaviour, verification methods, and lifecycle relationships. Rather than beginning with component geometry, MBSE begins with the question: what must the system do, under what conditions, and how will compliance be demonstrated?

A typical systems representation moves through a chain such as:

Requirements → functions → logical architecture → physical architecture → component design → analysis and verification → integration and validation.

This chain helps prevent local optimization. A lighter component is not automatically a better design if it increases vibration, reduces maintainability, violates safety requirements, or creates an interface problem. Traceability links a requirement to a design feature, analysis result, test, and approval record.

Product lifecycle management and the digital thread

PLM governs product information throughout its lifecycle, including requirements, CAD files, simulation models, drawings, bills of material, revisions, test results, nonconformances, service records, and retirement data. It provides configuration management, access control, revision history, workflows, and change authorization.

The digital thread is the connected flow of information across the product lifecycle. For example, a requirement for maximum bearing temperature may be linked to a thermal model, a sensor specification, a test procedure, an inspection record, and an operating limit. Without such links, organizations often repeat analyses, use obsolete geometry, or approve a design without knowing whether the analysis corresponds to the released configuration.

Building a Credible Simulation

A simulation is a mathematical and computational representation of a physical problem. Its credibility must be established systematically rather than inferred from attractive colour contours or a solver completion message.

Problem definition and idealization

The first task is to state the engineering decision the simulation will support. “Analyze the bracket” is inadequate. A stronger statement might be: “Determine whether the bracket meets the yield and fatigue requirements under the specified load spectrum and mounting condition, with a target safety margin.”

The analyst should define:

• the physical domain and relevant components;
• loads, constraints, contacts, and initial conditions;
• material properties and constitutive laws;
• operating and environmental conditions;
• output quantities and acceptance criteria;
• assumptions and excluded phenomena; and
• required accuracy, computational cost, and schedule.

Idealization converts the real system into a mathematical model. A component may be represented as a solid, shell, beam, rigid body, porous medium, or lumped parameter. Each choice introduces assumptions. A beam model may be appropriate for a slender shaft but unsuitable near a complex fillet or bearing seat.

Discretization and numerical solution

In FEA, the domain is divided into elements connected at nodes. The governing equations are approximated over this mesh. Element type, order, aspect ratio, distortion, and mesh density affect accuracy. A coarse mesh may provide a useful global displacement but fail to resolve a local stress concentration. A very fine mesh may consume resources without improving the quantity of interest.

In CFD, the governing conservation equations are discretized over a mesh or control-volume arrangement. Important choices include turbulence treatment, wall resolution, time step, spatial discretization, convergence criteria, and treatment of moving boundaries. In MBD, the analyst defines bodies, joints, actuators, contacts, friction, flexible components, and numerical integration settings.

Verification and validation

Verification asks whether the equations have been solved correctly. It concerns the mathematical and numerical implementation. Typical verification activities include comparison with analytical solutions, code-to-code comparison, mesh refinement, time-step refinement, conservation checks, and residual monitoring.

Validation asks whether the selected model represents the real physical system adequately for its intended use. It requires comparison with experimental or field data obtained under comparable conditions. A model may be numerically verified but physically invalid if the wrong material law, boundary condition, turbulence model, or contact assumption has been selected.

The distinction can be expressed simply:

• Verification: “Are we solving the equations correctly?”
• Validation: “Are these the appropriate equations and assumptions for reality?”

Neither process proves universal truth. Validation is conditional on a domain of applicability, measurement uncertainty, and the operating conditions used for comparison.

Convergence, sensitivity, and uncertainty

A converged solution is one for which numerical changes become acceptably small under the selected convergence criteria. In nonlinear analysis, convergence may be affected by contact changes, material yielding, large deformation, instability, or abrupt loading. A solver reporting convergence does not guarantee engineering accuracy.

A mesh-convergence study examines how a result changes as the mesh is refined. The quantity of interest should be specified: displacement, reaction force, natural frequency, average stress, or another engineering measure. Singular stresses at sharp re-entrant corners may increase indefinitely with refinement; in such cases, a nominal or averaged stress, local geometry modification, or fracture-mechanics treatment may be more meaningful.

Sensitivity analysis investigates how outputs respond to changes in inputs. It identifies influential parameters such as thickness, friction coefficient, material modulus, damping, inlet turbulence, or heat-transfer coefficient. Uncertainty quantification goes further by representing uncertain inputs through ranges or probability distributions and estimating the resulting output variation. This is essential when decisions depend on reliability rather than nominal performance.

Integrated Digital Engineering Workflow

An effective workflow is iterative rather than strictly linear.

Establish requirements and model purpose

The team first defines performance, safety, regulatory, cost, manufacturing, service, and environmental requirements. Each requirement should have a measurable verification method. The team also decides whether the intended model is for concept screening, detailed design, certification support, real-time control, maintenance, or another purpose.

Create and prepare the design model

The CAD model is developed with design intent and configuration control. An analysis representation is then prepared. Simplification must be documented, because the analyst should be able to explain why a feature was retained, removed, or idealized.

Select physics and model fidelity

Model fidelity should be proportional to the decision risk. A low-order beam model may be appropriate for rapid concept comparison, while a detailed nonlinear contact model may be necessary for final qualification. Using a highly complex model too early can slow exploration; using an oversimplified model for a safety-critical decision can create false confidence.

Define, solve, and inspect

Loads and constraints should represent the actual load path, not merely convenient software selections. The analyst inspects reactions, energy balance, deformation patterns, temperature fields, pressure distributions, contact status, and residuals. Unexpected behaviour is investigated rather than hidden by changing the display scale.

Correlate with tests and update the model

Physical testing should be planned to distinguish competing assumptions. If measured natural frequencies differ from predictions, possible causes include mass distribution, joint stiffness, boundary conditions, damping, or material properties. Updating a model to match one test without understanding the cause can reduce predictive capability elsewhere.

Release and manage evidence

The final package should include the model version, solver version, mesh and element information, material sources, assumptions, loads, boundary conditions, convergence evidence, validation comparisons, results, limitations, and approval status. This information is part of the engineering product, not administrative decoration.

Digital Twins in Mechanical Systems

A digital twin is a lifecycle-oriented digital representation connected to a physical asset through data and an explicit purpose. The connection may be periodic or continuous. A prototype digital twin may support design and testing before production. An operational twin may use sensor data to monitor condition, diagnose faults, estimate remaining useful life, or optimize operation.

The architecture commonly includes:

1. a physical asset or process;

2. sensors, actuators, and data-acquisition systems;

3. communication and data-ingestion services;

4. a digital model containing geometry, physics, rules, and state variables;

5. analytics for estimation, diagnosis, prediction, or optimization; and

6. user interfaces and decision workflows.

The twin may combine first-principles models with data-driven methods. A physics-based model explains behaviour and can operate outside the range of observed data, while a machine-learning model may capture complex patterns efficiently. Hybrid models seek to combine physical constraints with statistical learning.

A digital twin should not be defined solely by the presence of live data. A sensor dashboard is not necessarily a twin if it does not contain a meaningful representation, relationship, or decision function. Conversely, the appropriate data-update frequency depends on the application. A slowly changing industrial structure may require hourly data, whereas a high-speed rotating machine may require much faster sampling for condition monitoring.

Twin applications

In design, a twin can compare design variants using expected operating profiles. During commissioning, it can compare predicted and observed behaviour to identify installation errors. During operation, it can detect deviations from a baseline. In maintenance, it can estimate degradation and support condition-based interventions. In production, it can coordinate machines, tooling, quality data, and process parameters.

The business value depends on decisions improved by the twin. If no maintenance, control, design, or safety decision changes, collecting more data alone creates limited value.

Applied Case: Digital Twin for an Industrial Pumping Unit

Background

The following is an illustrative case. A plant operates several centrifugal pumping units in a process-water service. Unexpected bearing failures cause production interruptions. The existing maintenance practice is calendar-based, although the load varies substantially with production demand.

Situation and problem

The engineering team has CAD data, manufacturer pump curves, vibration measurements, bearing temperatures, motor current, suction pressure, and discharge pressure. The data are stored in separate systems, and the original simulation model does not match the currently installed configuration. The plant wants to reduce unplanned failures without replacing healthy bearings prematurely.

Analysis

The team first establishes the asset configuration: pump model, impeller diameter, motor rating, coupling arrangement, bearing type, installation date, and revision history. A hydraulic model estimates operating point and efficiency. An MBD representation examines shaft and coupling dynamics, while an FEA model evaluates bearing-housing stiffness and thermal expansion. These models identify variables that influence vibration and bearing load.

Historical sensor data are cleaned and synchronized. Fault records are reviewed to distinguish bearing degradation from cavitation, misalignment, imbalance, and operating outside the preferred range. The team finds that high vibration frequently follows operation at low flow, where hydraulic forces and recirculation increase. A purely temperature-based alarm would therefore be insufficient.

Approach and solution

An operational twin is created with a configuration-controlled asset model. Sensor data update estimated flow, pressure, shaft speed, temperature, and vibration state. A hybrid diagnostic model combines pump physics with anomaly detection. The system does not automatically declare failure; it presents a confidence-ranked diagnosis and recommends inspection when vibration trends, hydraulic operating point, and bearing temperature jointly indicate degradation.

The model is validated against archived events and then monitored during a controlled operating period. Alarm thresholds are linked to risk and measurement uncertainty rather than selected arbitrarily. Maintenance personnel are included in the workflow so that inspection findings can be fed back into the model.

Outcome

The principal improvement is not a particular software feature but the integration of configuration data, physical models, sensor measurements, and maintenance decisions. The plant can distinguish operating-condition effects from probable mechanical deterioration and can schedule inspections with stronger evidence. The twin remains subject to periodic recalibration because sensor drift, repairs, and configuration changes alter the asset.

Learning points

1. A digital twin requires a defined operational purpose.

2. Data quality and configuration identity are as important as analytical sophistication.

3. Combining physics and data can be more robust than relying exclusively on either.

4. Human inspection and engineering judgment remain part of the decision loop.

5. Model performance must be monitored after deployment.

Engineering Data, Governance, and Professional Responsibility

Digital engineering introduces risks beyond numerical error. A wrong revision, undocumented assumption, or uncontrolled spreadsheet can propagate through many departments. Configuration management should identify the authoritative model, approved revisions, responsible owners, and change history. Access permissions should protect intellectual property while allowing appropriate collaboration.

Cybersecurity is particularly important when operational twins connect engineering systems to industrial control networks. Threats may include unauthorized changes to models, manipulated sensor data, ransomware, stolen intellectual property, and unsafe remote commands. Secure authentication, network segmentation, encryption, logging, backup, patch management, and incident response are engineering requirements rather than solely information-technology concerns.

Ethical practice requires engineers to report uncertainty and limitations honestly. Colour maps should not be used to exaggerate insignificant differences. A model should not be calibrated selectively to conceal poor agreement. Automated recommendations affecting safety or maintenance must be explainable to an appropriate degree, auditable, and subject to human authority. Responsibility cannot be transferred to software or an artificial-intelligence model.

Regulatory and contractual requirements may govern data retention, traceability, software qualification, safety evidence, export controls, and protection of personal or operational data. The required level of governance should be proportional to consequence and risk, but safety-critical decisions demand particularly strong evidence.

Common Failure Modes and Best Practices

Frequent failures include beginning with software rather than the engineering question, importing geometry without understanding interfaces, applying unrealistic fixed constraints, ignoring preload and contact, using nominal material data, accepting default solver settings, confusing residual convergence with validation, and presenting maximum nodal stress without assessing singularity or relevance.

Further problems occur when simulation specialists work in isolation, when tests are planned after the model is already treated as final, when data lack units or timestamps, and when digital twins are deployed without an owner or maintenance process.

Best practice is to:

• define the decision and quantity of interest before modelling;
• document assumptions and idealizations;
• use progressively increasing model fidelity;
• conduct mesh, time-step, and parameter sensitivity studies;
• verify numerical implementation and validate against independent evidence;
• preserve configuration, revision, and provenance information;
• assess uncertainty and communicate it with results;
• connect simulation, testing, manufacturing, and service data;
• involve domain experts and end users; and
• review models after design changes, repairs, or changes in operating conditions.

Current Developments and Emerging Directions

Cloud-based simulation enables scalable computation and collaboration, although data sovereignty, cybersecurity, licensing, and network dependence must be addressed. Automated design exploration and topology optimization can generate efficient concepts, but manufacturability, fatigue, joining, inspection, and repairability still require engineering review.

Reduced-order models support rapid optimization and real-time digital twins by approximating high-fidelity simulations at much lower computational cost. Surrogate models and machine learning can accelerate prediction, anomaly detection, and inverse identification. Their use is strongest when supported by sound training data, physical constraints, uncertainty estimates, and monitoring for distribution shift.

The broader direction is toward interoperable digital threads in which requirements, geometry, analysis, test, production, operation, and maintenance remain connected. The major challenge is not only technical integration. It is the creation of shared semantics, disciplined processes, capable personnel, and governance that make digital information trustworthy across organizational boundaries.

Chapter Summary

Digital engineering integrates CAD, CAE, FEA, CFD, MBD, MBSE, PLM, and digital twins into a lifecycle approach to mechanical-system design. CAD defines geometry and design intent, while CAE evaluates physical behaviour. MBSE links requirements to architecture and verification; PLM controls product information and configuration; the digital thread connects evidence across the lifecycle. A credible simulation requires clear objectives, appropriate idealization, suitable discretization, verification, validation, sensitivity analysis, and uncertainty assessment. Digital twins add data connections to digital representations so that design, operation, diagnosis, maintenance, and optimization can be improved. Their success depends on data quality, model purpose, cybersecurity, governance, human judgment, and continuous monitoring. Emerging tools such as hybrid models, reduced-order models, cloud computing, automation, and AI extend capability but do not remove the need for engineering accountability.

Key Terms

CAD: Computer-based representation and definition of product geometry and design intent.

CAE: Computational tools used to analyse and evaluate engineering systems.

Finite-element analysis: Numerical solution of field problems by subdividing a domain into finite elements.

Computational fluid dynamics: Numerical analysis of fluid flow, heat transfer, and related transport phenomena.

Multibody dynamics: Analysis of motion and force interaction among interconnected bodies.

Model-based systems engineering: Systems engineering practice that uses formal models to represent requirements, architecture, behaviour, and verification.

Product lifecycle management: Governance and control of product information from concept through retirement.

Digital thread: Connected flow of product and engineering information across lifecycle activities.

Digital twin: A digital representation linked to a physical asset or process for a defined purpose.

Verification: Demonstration that a computational model or implementation solves the intended equations correctly.

Validation: Assessment of whether a model represents the real system adequately for its intended use.

Mesh convergence: Reduction of result changes as spatial discretization is refined.

Sensitivity analysis: Study of how output quantities respond to changes in input parameters.

Uncertainty quantification: Representation and propagation of uncertainty in model inputs and outputs.

Reduced-order model: Lower-complexity approximation of a high-fidelity model for rapid computation.

Configuration management: Control of product versions, changes, identities, and approval status.

Model fidelity: Degree to which a model represents relevant physical detail.

Traceability: Ability to connect requirements, models, results, tests, and decisions.

Hybrid model: Model combining first-principles physics with data-driven methods.

Model updating: Revision of model parameters or structure using test or operational evidence.

Review Questions

Short Answer Questions

1. Distinguish a CAD model from an analysis model.

2. What is the difference between verification and validation?

3. Why is mesh convergence performed?

4. Define a digital thread.

5. State two functions of PLM in mechanical-system design.

Long Answer Questions

1. Explain a complete workflow for developing and validating an FEA model of a mechanical component.

2. Discuss how MBSE and PLM improve traceability and configuration control.

3. Explain the architecture and lifecycle applications of a digital twin.

4. Analyse the role of sensitivity and uncertainty analysis in engineering decisions.

5. Discuss the cybersecurity, ethical, and governance responsibilities associated with operational digital twins.

Analytical / Application Questions

1. A bracket simulation predicts a very high stress at a sharp corner, and the value increases with mesh refinement. How should the engineer interpret and investigate this result?

2. A pump digital twin gives frequent false alarms after a maintenance overhaul. Identify likely causes and propose a model-governance and validation plan.

3. A design team must choose between a rapid beam model and a detailed nonlinear solid model for a safety-critical shaft connection. Develop a decision framework for selecting model fidelity and verification evidence.

22. Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes verification from validation in simulation?

A. Verification compares simulation with field measurements, while validation checks mesh quality.

B. Verification checks whether the equations are solved correctly, while validation checks whether the model represents the physical system adequately.

C. Verification concerns only CAD geometry, while validation concerns only manufacturing cost.

D. Verification is performed after production, while validation is performed before requirements are defined.

Correct Answer: B. Verification checks whether the equations are solved correctly, while validation checks whether the model represents the physical system adequately.

Question 2

A stress value continues to increase as the mesh is refined at a mathematically sharp re-entrant corner. What is the most appropriate first interpretation?

A. The solver must have failed because all stresses should converge to one value.

B. The material must have zero strength at that location.

C. The result may represent a stress singularity, so a relevant averaged quantity or realistic local geometry should be assessed.

D. The mesh should be made coarse until the stress stops increasing.

Correct Answer: C. The result may represent a stress singularity, so a relevant averaged quantity or realistic local geometry should be assessed.

Question 3

Which activity is most directly associated with product lifecycle management?

A. Controlling revisions, approvals, bills of material, and product information across the lifecycle.

B. Replacing all physical tests with artificial intelligence.

C. Increasing mesh density automatically in every simulation.

D. Converting every component into a rigid body.

Correct Answer: A. Controlling revisions, approvals, bills of material, and product information across the lifecycle.

Question 4

Why should an analysis model sometimes simplify a detailed CAD model?

A. To remove all features that influence physical behaviour.

B. To reduce computational cost while retaining features relevant to the question being analysed.

C. To ensure that the analysis result is always conservative.

D. To prevent the model from being validated experimentally.

Correct Answer: B. To reduce computational cost while retaining features relevant to the question being analysed.

Question 5

An operational digital twin is most accurately described as:

A. A static rendering of a product used only for marketing.

B. A sensor dashboard with no model or decision purpose.

C. A digital representation connected to a physical asset through data and used for a defined lifecycle purpose.

D. A replacement for all maintenance personnel and inspections.

Correct Answer: C. A digital representation connected to a physical asset through data and used for a defined lifecycle purpose.

Question 6

A simulation converges numerically but disagrees substantially with a well-controlled physical test. What is the most appropriate conclusion?

A. The simulation is valid because numerical convergence proves physical accuracy.

B. The test should be discarded automatically.

C. The model requires investigation of assumptions, boundary conditions, material data, and measurement comparability.

D. The colour scale should be changed to improve agreement.

Correct Answer: C. The model requires investigation of assumptions, boundary conditions, material data, and measurement comparability.

Question 7

Which sequence best represents a systems-engineering traceability chain?

A. Requirements, functions, architecture, design, verification, and validation.

B. Marketing, rendering, advertising, purchasing, and retirement.

C. Mesh, colour map, animation, export, and deletion.

D. Sensor, password, invoice, packaging, and shipment.

Correct Answer: A. Requirements, functions, architecture, design, verification, and validation.

Question 8

Why can a hybrid physics-and-data model be useful in a digital twin?

A. It eliminates the need for data quality controls.

B. It combines physical interpretability with the ability to learn complex patterns from operational data.

C. It guarantees accurate prediction outside every possible operating condition.

D. It prevents models from requiring recalibration after repairs.

Correct Answer: B. It combines physical interpretability with the ability to learn complex patterns from operational data.

Question 9

Which practice most directly improves the credibility of a safety-relevant simulation?

A. Using the most visually detailed contour plot available.

B. Accepting default boundary conditions to reduce analyst bias.

C. Documenting assumptions, performing verification, validating against independent evidence, and reporting limitations.

D. Avoiding physical testing to preserve development time.

Correct Answer: C. Documenting assumptions, performing verification, validating against independent evidence, and reporting limitations.

Question 10

A pump twin produces false diagnoses after a component replacement. Which issue should be examined first?

A. Whether the asset configuration and model revision were updated after the replacement.

B. Whether the display uses enough colours.

C. Whether all sensors can be removed from the system.

D. Whether the finite-element mesh contains only triangular elements.

Correct Answer: A. Whether the asset configuration and model revision were updated after the replacement.