Section: New Results

Modeling cyber-physical systems: from Signal to Signal+

Participants : Thierry Gautier, Albert Benveniste.

Based, initially, on two small case studies, we started a reflection on the modeling and analysis of cyber-physical systems by extending the model of synchronous languages, and in particular that of the Signal language [15]. The principle considered here is to remain within the traditional framework, in discrete time, of synchronous languages, but to completely generalize the equational style of specifications. In the usual Signal language, clocks are defined in a relational way by constraint systems. In contrast, data are defined using functionally inspired dataflow expressions. In this new Signal+, we propose to generalize the equational style to the data: numerical quantities also become subject to constraints. This typically corresponds to the way of modeling a system that includes physical components: for example, equations respecting balance laws have to be written. A major interest is that this programming style is much more compositional than more traditional Simulink or Lustre, or even Signal-based programming. The question then arises as to whether such a program should be analyzed. There is no notion of syntactic dependence, but the incidence graph can be used to define a matching that uniquely associates an equation with each variable. A scheduling can then be synthesized, provided that the reasoning is valid, which is the case for some classes of numerical algebraic equations, for which a solver can be used. However, we can observe on our case studies that the transition to Signal+ class is a real step forward in terms of difficulty. In discrete time, at each reaction, the free variables of the system must be evaluated using what is known about states and inputs. But in some cases, there will be more variables than usable equations. By reasoning on the fact that we have systems that are invariant in time, it may then be necessary to “shift” equations, just as if we were in continuous time, we could differentiate a constraint (a program is a discrete approximation of a continuous time system). This amounts, in a way, to proposing some modifications, which are considered as legitimate, to the source program.