What Data Do You Need for an Engineering Simulation?

Before gathering data for a simulation, state what you want to learn: a peak stress, a temperature distribution, a flow rate, a performance trend or a comparison between designs. The question determines the required fidelity, the physics, the geometry detail and the data. Many simulation projects collect too much of the wrong data and too little of what matters.

Geometry may come from CAD files, drawings, scans or measurements. Simplifications such as removing small fillets, threads or holes reduce cost but can change local results. State which features were removed and why. Dimensional tolerances can matter for contact, clearance and flow; consider whether nominal values suffice or whether variations should be studied.

Check units and coordinate systems, and confirm that the geometry represents the as-built condition if comparing with tests.

Mechanical simulations need elastic constants, yield and hardening data, density and possibly fatigue or creep data. Thermal simulations need conductivity, specific heat, density and emissivity. Fluid simulations need density, viscosity and, for compressible or multiphase flows, equations of state and surface tension. Properties depend on temperature, rate, direction and processing history.

Use measured data for the actual material when possible, and cite sources for handbook values; flag them as uncertain.

Boundary conditions define the physical context: supports, contacts, applied forces, pressures, heat fluxes, convection coefficients, inlet velocities and outlet pressures. These are often the largest source of error. Obtain them from measurements, standards or system-level models, and perform sensitivity studies for uncertain ones.

Define the range of operating conditions to be covered, including extremes and transients.

Transient simulations need initial states: temperature, velocity, stress or charge. Time-dependent loads need their histories at suitable resolution. Specify the simulation duration and time step considerations. For periodic problems, check that the solution has reached a periodic state.

Document assumptions about start-up and shutdown.

Mesh quality and resolution affect accuracy. Plan a convergence study with at least three systematically refined meshes where feasible, and choose element types suited to the physics. Solver tolerances, time-step controls and turbulence or contact models must be justified. Record software versions and settings.

Numerical data are part of the data requirements for reproducibility.

Validation requires measurements of quantities that the simulation predicts, with uncertainty: strain gauge readings, temperatures, pressure drops, displacement or velocity profiles. Design validation experiments to reflect the conditions simulated. If experimental data are not available, use benchmark problems and published data, and acknowledge lower confidence.

Separate data used for calibration from data used for validation.

Parameters carry uncertainty from measurement, variability and lack of knowledge. Sensitivity analysis shows which parameters most influence the output and therefore where better data are needed. Propagating uncertainty through Monte Carlo or surrogate methods yields output ranges. Reporting a single deterministic result can mislead.

Prioritise obtaining better data for the parameters that matter most.

Keep a record of all inputs: sources, versions, units, assumptions, simplifications, mesh settings and post-processing steps. Store input files, scripts and results with a clear naming scheme. Provide a short model description with the report. For publication, share files or summarise in the paper's supplement.

Good records make the model reusable and defensible.

Checklist: question defined; geometry and simplifications documented; materials and sources listed; loads and boundary conditions justified; initial conditions stated; mesh convergence planned; validation data identified; uncertainties considered; records kept. If several items are missing, plan data collection before simulation.

Our simulation and computational engineering services support model planning, building, verification, validation and reporting. We do not provide certified analyses.

Suppose a team wants to simulate the thermal performance of a plate heat exchanger. Geometry: plate dimensions, corrugation pattern, gasket and port geometry. Fluids: inlet temperatures, flow rates, density, viscosity, specific heat and thermal conductivity as functions of temperature. Plate material: thermal conductivity and thickness. Boundary conditions: inlet velocity or flow rate, outlet pressure, and thermal insulation to the environment. Validation: measured outlet temperatures and pressure drop at several flow rates, with instrument uncertainty. Numerical: mesh levels, turbulence model, convergence tolerance.

This example is hypothetical and illustrates how a data list is built. It does not describe a real project.

For a bracket under load: CAD geometry with fillets, material grade and heat treatment, elastic modulus, yield strength, and where relevant fatigue data. Loads: magnitudes, directions, spectrum and frequency. Supports: bolted connections, preload and friction coefficients. Mesh: element type and refinement at fillets. Validation: strain gauge measurements or deflection tests. Acceptance criteria: allowable stress or safety factor per the applicable standard.

Again, the list is a template for thinking, not a record of client work.

Missing inputs are normal. Options include measuring the quantity, taking it from the literature with uncertainty, estimating via simple models, bounding it with worst and best cases or treating it as an unknown parameter to be calibrated. Calibrated parameters must not be validated against the same data. Be explicit about each assumption.

Report sensitivity of results to the assumed values.

Agree on formats early: geometry as STEP or similar neutral formats, tabulated properties as CSV with units and sources, measured data with metadata and uncertainty, and results in open formats where possible. Provide a data dictionary. Version everything.

Clear data exchange reduces costly iterations.

Can I simulate without any experimental data? You can run a simulation, but its credibility is limited; use benchmarks and literature validation. How much detail do I need in geometry? Enough to capture the features that affect the quantity of interest. Do I need to supply material curves? If nonlinear behaviour matters, yes. How precise must boundary conditions be? As precise as the results require; sensitivity analysis shows this.

Can you estimate missing data? We can propose estimates with uncertainty, clearly labelled.

The data needed for a simulation follow from the question: geometry, materials, loads and boundary conditions, initial conditions, numerical settings and validation data, each with sources and uncertainty. Collecting and documenting them systematically is the best investment in a credible simulation. Our simulation services help plan and execute this process.

Draft a data list for your problem using the sections above, mark each item as measured, sourced, estimated or missing, and add an uncertainty note. Decide what must be measured before simulation begins. Share the list with colleagues for review; gaps become visible quickly.

Journals and reviewers increasingly ask how a simulation was set up. A well-kept data list becomes the methods section almost directly: geometry, materials, boundary conditions, solver settings, convergence and validation. It also helps you respond to reviewer questions without re-running unknown configurations months later.

Documenting as you go is far easier than reconstructing afterwards, and it supports collaboration when team members change.

Reviewers ask where material properties came from, how boundary conditions were chosen, whether the results are mesh independent, how the model was validated and what the uncertainty is. They also ask whether the geometry matches the tested specimen. Answering these in the paper, with a short table of inputs and sources, avoids a round of revision.

The amount of data you collect should match the consequence of the decision. A quick comparison of two concepts needs less than a study that will underpin a publication or a safety margin. State the purpose and scale the effort accordingly.

Over-collecting delays work; under-collecting undermines it. A short planning discussion at the start helps find the right balance.