Capability · Simulation
Discrete-event, process, and digital-twin simulation across the cell, the line, the plant, and the fleet. The same simulator that runs scenario studies for industrial engineers runs as the guard for autonomous dispatch in production — the agent is constrained to actions that simulate cleanly first.
01
SimPy / OpenDESS-class engines wrapped in a common interface. Line balancing, station-bottleneck identification, AGV/AMR fleet sizing, dock scheduling, shift-pattern optimisation. The simulator runs an ensemble — typically n=32 — across stochastic demand and reliability realisations, returning quantile bands rather than single trajectories.
02
Aspen / gPROMS / Cadet-class steady-state and dynamic process simulators for chemical, refining, pharma, food. Coupled with the RL controller training loop so SAC / PPO / CIRL controllers learn against the simulator before touching the real loop. Validated against published catalytic-reforming, gasoline-blending, fermentation, and paper-machine baselines.
03
Plant-scale photorealistic twins for visual replay of dispatch decisions, robot policy training, and ergonomic / safety review. Siemens Digital Twin Composer launched at CES 2026 on the Xcelerator marketplace; NVIDIA Omniverse provides the rendering and physics substrate. Reference deployments include the Siemens-NVIDIA AI-driven Erlangen factory.
04
OpenVLA and Octo fine-tunes happen in simulation first — the platform spins up vendor-specific cobot kinematics + dynamics, generates demonstrations, validates the policy against safety envelopes, and only then deploys to the real cobot. This compresses the validation cycle from weeks to hours and keeps the cell available for production.
05
What-if studies on intervention strategies: which assets benefit most from condition-based monitoring? What's the ROI of installing AE sensors on this gearbox population? Which spare-parts inventory levels minimise downtime risk under projected demand growth? The platform models each scenario against the asset-level reliability surface.
06
Before the agent commits any dispatch action — a cobot trajectory, an AGV reroute, a batch-recipe parameter change, an autonomous-truck handoff — the simulator validates the action against current state and policy. If it doesn't simulate cleanly, the agent doesn't act. This is the safety mechanism that lets autonomy ship inside IEC 62443 / NIS2 / ISO 10218 regimes.
Models in play
ProcessSim DES
discrete-event
Line, fleet, dock, scheduling — wrapped behind a common API.
Process simulator (Aspen / gPROMS-class)
continuous process
Steady-state + dynamic for chemical and refining loops.
NVIDIA Omniverse + Siemens DTC
digital twin
Plant-scale photorealistic replay; CES 2026 launch onward.
ROS 2 + Isaac Lab
robot training
Cobot kinematics + dynamics in sim before real-cell deployment.
Related capabilities