Learning factory

Simulation study based on VDI 3633

Simulation becomes reliable when objectives, data, model boundaries and experiments are defined before the first result.

In brief

A simulation study prepares, conducts and evaluates targeted experiments on a dynamic model. It does not replace planning or data validation. The model should only be as detailed as the decision requires.

What is a simulation study?

A simulation study examines the time-dependent behaviour of a real or planned system in an executable model. Repeated runs vary parameters or structures so that throughput, inventory, waiting, utilisation and service levels can be compared under defined scenarios.

VDI 3633 Part 1:2014-12 describes the fundamentals. The series has since been extended, including simulation and optimisation in Part 12 (2020), 3D movement and process simulation in Part 8 (2023), and tool selection in Part 4 (2025).

When is a question suitable for simulation?

Simulation is particularly useful when time-dependent or random influences interact and a static calculation cannot reproduce system behaviour. Typical cases include queues, variable arrivals, failures, complex priorities, shared resources and feedback loops.

An analytical estimate should first determine whether the target is theoretically achievable. A clear capacity deficit is not solved by a detailed model. Simulation produces insight into dynamics; it does not replace system design.

Eight elements of a reliable study

  1. Decision question and objectives: define targets, conflicts and stop criteria.
  2. System boundary: specify inputs, outputs, resources, rules and exceptions.
  3. Data basis: check workload, technical and organisational data for completeness and consistency.
  4. Analytical estimate: determine bottlenecks, minimum capacities and plausible result ranges.
  5. Conceptual and computer model: represent only decision-relevant properties.
  6. Verification and validation: test whether the model is implemented correctly and is sufficiently realistic for the question.
  7. Experiment plan: document scenarios, warm-up, run length, replications and random-number handling.
  8. Evaluation: report uncertainty, sensitivity and transferability transparently.

Which data is needed?

Data classExamplesQuality question
System loadOrders, product mix, arrivals, batch sizesIs the period representative and are peaks included?
TechnologyCapacity, processing time, failure, repairAre values nominal or observed in operation?
OrganisationShifts, priorities, releases, staffingAre special rules and breaks represented?
TopologySources, destinations, buffers, routesDoes the model structure match the current layout?

As detailed as necessary, not as detailed as possible

More detail does not automatically improve validity. Every additional parameter needs data, validation and maintenance. Detail follows the target measures: routes and traffic rules may matter for vehicle-fleet sizing; vehicle colour does not.

What belongs in the decision report?

Besides averages, the report should include variability, confidence intervals or suitable quantiles, bottleneck states and sensitivities. Assumptions and omitted effects must be visible. Animation supports communication and debugging, but it is not evidence of performance.

Technical basis

  • VDI 3633 Part 1:2014-12, fundamentals of simulation studies.
  • VDI 3633 Part 12:2020-07, combining simulation and optimisation.
  • VDI 3633 Part 4:2025-09, selection of simulation tools.

See also material flow simulation in projects and our material flow simulation glossary page.

FAQ

Frequently asked questions

When is a static calculation sufficient?

When load, capacity and relationships can be assessed without relevant time-dependent or random interactions. Every simulation should begin with an analytical estimate.

How many simulation runs are required?

That depends on variability, warm-up, run length and the desired statistical confidence. A universal fixed number would be methodologically unsound.

Is a 3D animation a quality criterion?

It helps communication and plausibility checks. Objectives, correct logic, validated data and a documented experiment plan are decisive.

What is the difference between verification and validation?

Verification asks whether the model was implemented as intended. Validation asks whether it represents reality sufficiently for the specific purpose.

Contact

Turn your challenge into a sound plan

Talk to us about your data basis, alternatives and the next robust decision.