Multi-Domain Co-Simulation for Heterogeneous Systems

47 concepts38 papers2 proposals

When engineering complex modern systems—such as autonomous vehicles, hybrid powertrains, or smart cities—designers face a fundamental hurdle: these systems do not exist within a single domain of physics. An autonomous vehicle relies on continuous mechanical tire dynamics, non-linear hydraulic braking, discrete electronic control units, and high-frequency wireless network signals, all while operating inside a macro-scale traffic grid.

To simulate this accurately, you cannot rely on a single, monolithic simulator. If you attempt to model a whole city using millisecond-level differential equations required for tire physics, the computation becomes mathematically intractable. Conversely, if you model the vehicle using the coarse, discrete steps of a traffic simulator, the physics lose all fidelity.

Multi-domain co-simulation solves this dilemma. Instead of forcing diverse physical phenomena into a single solver, it orchestrates a network of specialized, independent simulators, allowing each to run at its own natural scale while precisely exchanging data at synchronized boundaries.


The Two Paths: Unified Modeling vs. Co-Simulation

To understand why co-simulation works, it helps to contrast it with its closest alternative: unified multi-domain modeling.

Unified Modeling (Monolithic)
[ Mechanical + Electrical + Hydraulic ] ---> Driven by a Single Solver

Co-Simulation (Modular)
[ Simulator A: Physics ] <--- Gateway / FMU ---> [ Simulator B: Traffic ]

In unified modeling, every subsystem is reduced to a single, shared mathematical framework. For example, Thanh-Son Dao et al. demonstrated in their Mathematics-Based Modeling of a Series-Hybrid Electric Vehicle that symbolic computing can compress hundreds of equations spanning mechanical, electrical, and chemical domains into a tight, unified system. Similarly, Ziqiang Li et al. leveraged object-oriented languages like Modelica in HIL Simulation of Aircraft Thrust Reverser Hydraulic System in Modelica to define standardized interfaces between mechanical and hydraulic components. This monolithic approach is highly accurate and completely eliminates synchronization errors because a single solver advances the entire system simultaneously.

However, unified modeling breaks down when subsystems are built on fundamentally incompatible software paradigms or require vastly different scales. You cannot easily compile a 3D graphics rendering engine, a deep learning planning model, and a stochastic traffic simulator into a single set of differential equations.

This is where co-simulation becomes necessary. As outlined by Gary Tan et al. in AsiaSim 2013: 13th International Conference on Systems Simulation, Singapore, November 6-8, 2013. Proceedings, co-simulation connects disparate software environments through standardized gateways. Each simulator acts as a black box that computes its own internal states and exposes only its inputs and outputs to the external coordinator.


Bridging Scale and Fidelity

The primary power of co-simulation lies in its ability to bridge the "fidelity gap" between micro-level physics and macro-level environments.

Consider the validation of cooperative driving automation. To test how connected automated vehicles interact, researchers need both realistic vehicle physics and dense traffic environments. To achieve this, Yi Guo et al. developed the OpenCDA: An Open Cooperative Driving Automation Framework Integrated with Co-Simulation framework, which was later expanded into a full ecosystem by Runsheng Xu et al. in The OpenCDA Open-source Ecosystem for Cooperative Driving Automation Research. OpenCDA couples high-fidelity vehicle physics and graphics engines with large-scale traffic flow simulators (such as SUMO).

By separating these layers, the traffic simulator can manage thousands of background vehicles efficiently, while the physics engine focuses its computational power entirely on the ego-vehicle's immediate maneuvers. To push this further, Laura Ferrarotti et al. integrated a physical driving cockpit into this exact pipeline in Autonomous and Human-Driven Vehicles Interacting in a Roundabout: A Quantitative and Qualitative Evaluation, allowing a real human driver to interact directly with simulated autonomous agents in a synchronized virtual environment.

This cross-domain bridging is equally vital for capturing hidden physical dependencies, such as:


The Mechanics of Synchronization

If simulators run independently, how do they stay in sync? If Simulator A steps forward by 10 milliseconds, but Simulator B steps forward by 1 millisecond, they will quickly diverge and pass invalid data.

Co-simulation solves this using a modular testing architecture and standardized data exchange formats, a necessity highlighted by Andrea Piazzoni et al. in Challenges in Virtual Testing of Autonomous Vehicles. The most widely adopted standard for this is the Functional Mock-up Interface (FMI).

Under this standard, domain-specific models are compiled into Functional Mock-up Units (FMUs). An FMU contains its own internal solver and a standardized C-interface for setting inputs, stepping the simulation forward, and retrieving outputs.

[ Master Algorithm / Coordinator ]
       |                  |
       v (Step to t+dt)   v (Step to t+dt)
   [ FMU 1 ]          [ FMU 2 ]
 (Fluid Cooling)    (Power Grid)

A master algorithm coordinates the execution, acting as a conductor:

  1. It pauses all simulators at a specific time $t$.
  2. It collects the output variables from each FMU.
  3. It maps and transfers those outputs to the input variables of dependent FMUs.
  4. It commands each FMU to advance by a discrete communication step ($\Delta t$).

This mechanism was utilized at a massive scale by Wesley Brewer et al. in A Digital Twin Framework for Liquid-cooled Supercomputers as Demonstrated at Exascale. To build a digital twin of an exascale supercomputer, the authors exported transient thermal-fluidic cooling models as FMUs and integrated them directly with discrete resource allocators and power simulators, creating a holistic energy-tracking environment without rewriting the underlying code of either system.


The Frontiers of Co-Simulation

While co-simulation provides immense flexibility, it introduces new research challenges at the boundaries of automation and speed.

One major hurdle is integrating autonomous control logic directly into multi-domain physical simulations. As noted by Martin R. Kuhn et al. in Aspects of Train Systems Simulation, testing automated driving algorithms against heavily coupled mechanical, electrical, and thermal models remains a complex software engineering challenge due to the strict real-time constraints of control software.

Additionally, researchers are turning to deep learning to accelerate individual simulation domains. However, as Feng Qin et al. point out in Deep-learning-based surrogate model for fast and accurate simulation in pipeline transport, coupling independent neural surrogate models into a unified, closed-loop simulation that spans multiple physical domains is incredibly difficult. If one surrogate model introduces a tiny bias, that error can feed back into the next model, causing the entire co-simulation loop to destabilize. Developing generalized frameworks that maintain stability across these integrated, learned subsystems represents the next major frontier in system simulation.

Key Foundational Work

HIL Simulation of Aircraft Thrust Reverser Hydraulic System in Modelica (2009) — contributed 1 concept elements to this direction.
59015 Mathematics-Based Modeling of a Series-Hybrid Electric Vehicle(Vehicle Dynamics &amp; Control including Tire Dynamics) (2010) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_multi_domain_co_simulation_f291c3ec,
  title = {multi-domain co-simulation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/multi-domain-co-simulation/rdr_f291c3ecf87f4445b443e7a4df488b56},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_multi_domain_co_simulation_f291c3ec_v1,
  title = {multi-domain co-simulation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/multi-domain-co-simulation/rdr_f291c3ecf87f4445b443e7a4df488b56/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake