Simulation is one of the strongest arguments for MBSE: instead of hoping a model describes the behaviour you intended, you can execute it and watch. With CATIA Magic 2026 Refresh 1 there is now a simulation plugin for SysML v2. We unboxed it - and here is what already holds up, where the limits are, and which pitfalls are better known up front.

Why Simulation Is the Most Valuable Feedback Channel in Design

Simulation moves feedback forward. Failures that would otherwise surface only when the real product is tested show up on the model instead - where corrections are cheap and easy. MBSE delivers plenty without simulation, but the view of dynamic behaviour is missing.

And there is a new reason to look closer: AI needs feedback on its work. An executable model is exactly that - objective evidence of whether a generated design behaves as intended.

What the Plugin Covers Today

The focus is clearly on state-based behaviour. What gets simulated is exhibit states with their states, transitions and triggers - including payload, change and time triggers, plus guards. Entry, do and exit sections can modify attribute values through assign actions. Also on board: an expression evaluator that continuously evaluates expressions, and a requirement checker that shows during the simulation whether a requirement is satisfied.

The simulation window itself is pleasantly plain: a console with the full log, current variable values, single-step execution, manual firing of transitions, a speed control, time series charts for selected variables, and CSV export of the results.

The bundled water tank example shows how far this reaches: a controller regulates the fill level through a valve, and the time series chart shows the characteristic rise and fall of a control loop. This is recognisably real system behaviour.

What's Still Missing

The plugin is a technology preview - and you notice that at the edges. Parametrics and the simulation of action behaviour are not included. Definitions are barely supported; you essentially work with usages. The simulation does not consider units and quantities, which calls for care with physical quantities. And the direction of directed features on ports is not checked: the payload is transferred whether the direction is right or not.

Five Pitfalls From Practice

First: model behaviour through parts with an exhibit state. States used directly behave unexpectedly - the vendor's own samples use exhibit states too, and semantically it is the better fit.

Second: set the initial node explicitly. Without it, the simulation starts every state that no transition points to - an unnecessary ambiguity in a team setting.

Third: only values declared with := are editable at runtime. If you want to change values during a simulation, model them that way.

Fourth: for payload transfer, sender and receiver must sit in a common context - an enclosing part that contains both.

Fifth: use the console. In fast sequences you won't see the intermediate steps in the diagram, but the log captures every value change.

Outlook: Simulation Meets AI

Parametrics as well as units and quantities are announced. The more interesting prospect is the combination taking shape: simulation as automated feedback for AI-assisted modelling. Once generated models can not only be checked syntactically but actually executed, you get a correction loop that works without manual recalculation.

Getting started takes little effort: one part with an exhibit state, two states, one transition without a trigger. Then add triggers, track an attribute, check a requirement. That is all it takes to be up to date.

Do you have a topic that you would like us to write about? Let us know.