Today, embedded systems are increasingly used to control complex systems. In this thesis, we are interested in critical embedded systems used for the control of transport systems such as railway systems. The aim of this work is to enable the design of fault-tolerant systems for the control of transport systems. We propose a new timed embedded systems modeling approach to diagnose their faults. It is based on decomposition of the system and structural extension of diagnosability context of modular timed systems. In DES, there are two basic approaches for diagnosis: diagnoser based approach and chronicles (Causal Temporal Signature (CTS)) based approach. The major limitation of diagnoser approaches rely in the management of the combinatorial explosion related to the formalism of automata. In this work, our main lock is to combat this limit. We propose new engineering models based method for the diagnosis of critical embedded systems. On the other hand, the major limitation of chronicles approach is first to be able to guaranty the consistency of a database. A second level of difficulty is in interpreting some sequences of events at the input of the diagnostic system under the hypothesis of multiple failures. In this work, we propose two different methods to verify the consistency of a set of CTS and we propose an interpretation algorithm based on a concept of worlds which guarantees the correct diagnosis
