The main goal of the works developed by the research group dedicated to "Dependability and system diagnosis" (french acronym SURFDIAG), relates to the elaboration of a global approach for the design and the operation (supervision, maintenance, reconfiguration) of complex automated industrial systems. The dependability, also known as RAMS (for Reliability, Availability, Maintainability and Safety) in the concerned community, can be split into several subproblems.

Firstly, at the design stage, it is necessary to guarantee the system safety, i.e. to guarantee that the system will operate according to the given specifications (in normal operation mode) and forecast alternate modes allowing the system to continue to operate even if some part of it are out of order.

Secondly, at the operating stage, the supervision system must:

- detect any fault, i.e. decide that the system does not operate normally, using the overall available information on the actual behavior (obtained through the measurements) and on the expected behavior (often forecast by a system model or a model of the data);

- isolate the faults, i.e. decide which function (or, at least, which component) is faulty based on data redundancy;

- compensate for the faults, i.e. conceive and implement a fault tolerant control (that leads eventually to a degraded system performance) or reconfigure either the control architecture or, if it is possible, the process architecture itself.

Research activities are structured into three projects which interact each others.

Embedded diagnosis et reconfiguration of distributed control systems through communication network. (DESYR) The main goal of the project DESYR are twofold. The first is a methodological problem and deals with the collaboration and co-design of FDI/FTC algorithms. The second, related to component-oriented technology, is intended to implement on-board computers and/or nomadic the developed algorithms. On the scientific level, this issue includes important bolts of a fundamental nature in the field of automation of information theory, or networks. In particular, the control of networked systems (Networked Control Systems) still raises many questions: taking into account the delays, congestion and packet loss, implementation of distributed systems under communication constraints, implementation of diagnosis/decentralized modules possibly accommodation with a capacity for autonomy, non-synchronous sampling, introduction of service quality in the feedback loop, etc..

Systems diagnosis/reconfiguration: analysis and safe design (DIRAC) The research conducted within the project DIRAC focus on the recent issue of tolerance to faults/failures in dynamic systems. The key issue is to contribute to the development of a methodology of diagnosis/reconfiguration of complex systems and/or uncertain starting from the establishment of a "diagnosis model" until the reconfiguration phase of the system, passing through a stage of diagnosis and/or observation. We focus on specific problems dealing with different residual generation methods with the overall aim of improving their performance. Still a problem little discussed is the analysis of structural properties related to "diagnosability", observability and "reconfigurability" of systems. This analysis is essential to conduct during the system design since a compromise is often sought between cost minimization and a slight over-instrumentation useful in fault diagnosis and their accommodation. Another way of research is to optimize the interactions that link the fault diagnosis to the modeling of systems. Indeed, a reliable diagnosis (in terms of isolation or estimation the fault magnitude) of a system is inconceivable without an effective combination of all the knowledge needed for its establishment. In this context, our research focuses both on the implementation of methods of residual generation and on the establishment of a modeling methodology dedicated to the system diagnosis.

Safety and diagnosis interactions (ISD) Today, problems of control, safety and security of dynamical systems are often treated separately. It should however take into account all interactions between supervision, safety and security in an integrated model in order to propose a comprehensive approach necessary to solve such problems. The safe design and quantitative evaluation of dependability and the diagnosis and recognition of modes of operation are two elements contributing to solving the previous problem. The project "Safety and Diagnosis Interactions" revolves around three actions. The first relates to automated systems constrained by the dependability and the security. It is necessary, in particular, to assess the dependability parameters of a complex system on a probabilistic point of view. These issues are placed firmly within the framework of the dynamic reliability. The second action concerns the diagnosis and recognition of modes of operation. Research are conducted for (i) the design of sensitive inputs (with regard to the recognition of modes), (ii) the separation of modes in the presence of measurement noise, (iii) the robustness of the detection of the changing of mode with regard faulty measurements or (iv) the research and the identification of modes without a priori model. The third will explore complementarities and interactions between the state estimation established during the monitoring and the evaluation of dependability. Indeed, in the context of dynamic reliability, one must take into account the diagnosis process in the predictive evaluation models.