Frequently asked questions

A CoSimulator is an environment that coordinates the execution of several computational models so that they can exchange information and take part in one coupled experiment. The participating models may represent physical processes, communication networks, control systems, or other interacting domains.

Its purpose is to make the behavior arising from the interaction of those models available for study. Coordination methods differ between environments: models may advance in lockstep, on a negotiated exchange interval, or under discrete-event scheduling, and each arrangement carries its own assumptions about what happens between exchanges.

CoSimulator describes the broader technology category. CoSimu is the name of this research initiative.

CoSimu aims to develop reusable twin kernels of physical processes and reusable security-analysis libraries within an orchestrated experimental environment, so that cyber-physical security questions can be investigated with components that outlast a single study. Power systems provide an initial application context, while the architecture is being developed for broader application across compatible cyber-physical processes.

Cosimulation provides the setting. The architecture proposed within it has four parts: reusable process representations, reusable analytical methods, an orchestration layer that composes them, and experimental evidence recorded together with the conditions under which it was obtained.

A twin kernel is a reusable computational representation of a particular physical process or cyber-physical system, such as an electrical power system, an industrial process, water infrastructure, or another process for which a defensible model can be constructed.

A kernel exposes the dynamics it integrates, the state variables it holds, the parameters that configure it, the interfaces through which it can be read and written, and the operational constraints that bound valid behavior. Its role is to provide a representation of a process that can participate in suitable experimental configurations without being rebuilt for each study.

Kernels differ in modeling fidelity and in internal implementation. An electromagnetic-transient representation and a phasor-domain representation of the same network answer different questions, so the choice of kernel forms part of what a result depends on. Kernels are process-specific: the power-system kernel under development is the first, and each further class of process calls for a kernel of its own.

A standalone simulator is generally used to obtain a result for one system, within its own environment. A twin kernel is built to participate: its interfaces are declared, so that other models and security-analysis methods can interact with it while an experiment runs.

The difference concerns composition. A kernel may well be implemented using established simulation software; what defines it is the declared interface, and the reuse across experimental configurations that the declaration allows.

The libraries are intended to hold reusable experimental and analytical methods, expressed against declared interfaces so that a method is independent of any one simulator. Their intended scope covers:

• Attack scenarios: manipulation of measurements, commands or configuration at defined points
• Operational disturbances and fault conditions: non-adversarial events that affect the same behavior
• Physics-informed anomaly detection: analysis evaluated against process dynamics and operating conditions
• Moving target defense: perturbation of quantities an adversary must estimate
• Security evaluation and resilience analysis: procedures for comparing configurations under the same conditions

A library binds to a kernel through that kernel's interface, addressing the measurements, setpoints, state variables and constraints the kernel declares. The libraries are being developed alongside the first kernel, and a method written against one physical process is adapted before it addresses another.

Orchestration is the layer that assembles the selected models into one experiment and coordinates them while it runs. It binds the declared interfaces, transfers state between models, coordinates how each advances in time, applies interventions at defined points, and records the configuration alongside the run.

The coordination method belongs to the experimental design. Continuous solvers, discrete-event models and scan-based control logic represent time differently, and the approach chosen for a configuration determines how behavior occurring between exchange points is treated.

In the proposed architecture, a coupled experiment is composed in a defined progression: suitable process representations are selected; the relevant security-analysis methodology is chosen; the experimental configuration is established, including interfaces, parameters and operating conditions; compatible components are coupled; their execution is coordinated; a controlled intervention is applied; the resulting cyber-physical behavior is observed; and the evidence is analyzed in the context of the modeled system.

Each step is an experimental decision recorded with the run, because the conclusion a result can support depends on all of them.

Portability in CoSimu refers to reusing suitable computational representations and experimental methodologies across compatible research configurations. The required interfaces, parameters and validation remain specific to the physical process being studied.

A security methodology developed for a power-system process may provide a reusable experimental framework: a way of posing the question, applying the intervention and evaluating the response. Applying that framework to a different process calls for its own interfaces, assumptions, parameters and validation, since the physical dynamics differ.

The methodology is what travels. The configuration, the tuning and the evidence remain specific to the system studied, and that distinction is one of the central ideas of the initiative.

Separating the representation of a physical process from the methodology used to investigate it is intended to:

• reduce the repeated construction of experimental apparatus for each study
• allow suitable methodologies to be compared against the same process representation
• allow compatible processes to be studied through reusable experimental components
• improve the organization of research configurations and the record that accompanies them

Modularity supports reproducibility: an experiment is reproducible when its configuration, assumptions and interventions are described precisely enough for another group to repeat them.

A coupled experiment allows a cyber intervention and its physical consequences to be examined within one system context. What can be investigated includes the intervention itself, the control-system response, the physical behavior that follows, propagation across coupled domains, what a defender at a given monitoring position is able to observe, and how that picture changes under alternative defensive configurations.

The value lies in contextualized evidence: one event as observed from the communication path, from the control system and from the physical process, along a single experimental timeline.

Experimental conclusions are interpreted in relation to the models, the fidelity, the operating conditions and the coupling used in the study. Comparison across configurations then supports the examination of propagation, observability and the effect of defensive changes. Carrying a conclusion to an installed system calls for validation against that system.

The first development direction is electrical power systems, where the dynamics are well characterized and the operational context is documented. Industrial processes, water infrastructure and other processes with compatible dynamics are candidates for later kernels.

Representing a given process depends on the availability of a defensible model, the interfaces it is able to declare, and whether its timescales support coupling with the other models in the configuration. Extension beyond power systems is an active line of work.

CoSimu is being developed around three initial components: the twin-kernel specification, the interface model used by security-analysis methods, and the orchestration layer that composes participating models. The first twin kernel is being developed for an electrical power-system context.

The broader direction adds further process-specific kernels, reusable security-analysis libraries, and larger orchestrated experiments; the Platform page sets out the architecture.