A behavioral framework for compositionality: linear systems, discrete event systems and hybrid systems

A. Agung Julius, Arjan van der SchaftUniversity 2004

Abstract

In this paper, we formulate a general framework based on the behavioral approach to dynamical systems, in which various issues regarding the interconnection of systems can be addressed. The main part of the framework is that interconnections or compositions of systems can be modeled with the interconnection of behaviors and generalized projection operations. Control problems such as supervisory control or feedback control can be expressed in terms of behavioral interconnection and thus integrated into this framework. We discuss several variants of control problems and provide solutions to them.

Introduction

Modern engineering frequently involves the integration of heterogeneous systems. For instance, a physical motor operating in continuous time might be managed by a digital computer operating as a discrete event system. Traditionally, these domains are treated with distinct mathematical toolsets: differential equations for continuous systems and automata or formal languages for discrete ones. The primary challenge is compositionality—creating a consistent mathematical language to describe how these diverse systems interact. This paper addresses this by proposing a "behavioral" framework that treats systems not as input-output blocks, but as sets of possible trajectories.

The Behavioral Approach to Systems

The core of this method is the definition of a system by its behavior ($B$), which is the set of all signals or trajectories the system can produce. Unlike traditional transfer functions, this approach does not pre-suppose an input-output directionality. Interconnection ($||$) between two systems is defined simply as the intersection of their behaviors ($B_1 \cap B_2$), meaning the combined system can only perform actions allowed by both components.

General control configuration where a plant P and controller C are interconnected through control variables c to regulate variables w.

To handle complex systems where only some variables are shared or observed, the authors introduce projections ($\pi$). Projections map high-dimensional behaviors to lower-dimensional ones, effectively "hiding" internal variables or "sampling" continuous signals into discrete points. This allows for partial interconnection, where a controller only "sees" a subset of the plant's variables.

Key Insights: Projections and Hybrid Systems

A significant contribution of this framework is its ability to unify hybrid systems. By treating sampling and zero-order hold operations as specific types of projections, the authors place discrete controllers and continuous plants on the same mathematical footing.

A hybrid control loop illustrating how projections (sampling and hold) bridge the gap between continuous plant behavior and discrete controller behavior.

The paper also redefines observability as a partial order between projections. A projection is considered "more observable" if it retains more information about the underlying behavior. Furthermore, the homogeneity property is identified as a condition where internal and external behaviors are synchronized, greatly simplifying the synthesis of controllers through the construction of "canonical controllers."

Results and Canonical Controllers

The authors derive two "canonical controllers" that serve as universal solutions. The First Canonical Controller ($C'_{can}$) is the maximal controller that ensures a system adheres to a given specification. They prove that if a control problem is solvable, this canonical controller will achieve the desired outcome. The framework also successfully maps classical discrete event system concepts, such as "normality" of a language, into this generalized setting, demonstrating that it is a true extension of existing theories.

Conclusion

While the behavioral framework provides a powerful "lingua franca" for system composition, challenges remain regarding causality and physical compatibility (e.g., avoiding deadlocks in discrete systems). Future work involves extending these tools to multi-objective control problems, where a single controller must satisfy multiple, potentially conflicting, performance and safety specifications simultaneously.

Summary updated

Abstract

In this paper we formulate a general framework based on the behavioral approach to dynamical systems, in which various issues regarding interconnection of systems can be addressed.The main part of the framework is that interconnections or compositions of systems can be modelled with interconnection of behaviors and generalized projection operations.Control problems such as supervisory control problem or feedback control problem can be expressed in terms of behavioral interconnection, and therefore can be put into this framework.In the paper we discuss some variants of control problems and provide solutions to them.
BibTeX
@inproceedings{julius2004behavioral,
  author = {A. Agung Julius and Arjan van der Schaft},
  title = {{A behavioral framework for compositionality: linear systems, discrete event systems and hybrid systems}},
  booktitle = {University},
  year = {2004},
}
Figure