A formal approach for correct-by-construction system substitution

Safety-critical systems depend on the fact that their software components provide services that behave correctly (i.e. satisfy their requirements). Additionally, in many cases, these systems have to be adapted or reconfigured in case of failures or when changes in requirements or in quality of service occur. When these changes appear at the software level, they can be handled by the notion of substitution. Indeed, the software component of the source system can be substituted by another software component to build a new target system. In the case of safety-critical systems, it is mandatory that this operation enforces that the new target system behaves correctly by preserving the safety properties of the source system during and after the substitution operation. In this thesis, the studied systems are modeled as state-transition systems. In order to model system substitution, the Event-B method has been selected as it is well suited to model such state-transition systems and it provides the benefits of refinement, proof and the availability of a strong tooling with the Rodin Platform. This thesis provides a generic model for system substitution that entails different situations like cold start and warm start as well as the possibility of system degradation, upgrade or equivalence substitutions. This proposal is first used to formalize substitution in the case of discrete systems applied to web services compensation and allowed modeling correct compensation. Then, it is also used for systems characterized by continuous behaviors like hybrid systems. To model continuous behaviors with Event-B, the Theory plug-in for Rodin is investigated and proved successful for modeling hybrid systems. Afterwards, a correct substitution mechanism for systems with continuous behaviors is proposed. A safety envelope for the output of the system is taken as the safety requirement. Finally, the proposed approach is generalized, enabling the derivation of the previously defined models for web services compensation through refinement, and the reuse of proofs across system models