Retour à la liste des thèses
Ajouté le: 14/05/2014
Directeur : OLARU Sorin -
Thèmes : Automatique, Signal, Télécoms, Systèmes embarqués
Laboratoires : E3S Supélec Sciences des Systèmes EA 4454
Description :


Sorin Olaru (, Vasso Reppa (


Cyber-physical systems (CPS) are comprised of
(i) physical and engineered systems, which are usually large-scale and complex,
(ii) a cyber-core, which consists of communication and computational means that monitor, coordinate and control the physical part.
The key motivation for migrating from “conventional” systems to CPS is the need for enhancing the “intelligence” of the physical systems in many application domains in order to be able to plan and modify their actions based on self-awareness and the evolving environment, and for handling a huge amount of data of different time and space characteristics.
In various CPS, a large number of sensors and sensor networks is used for
(a) monitoring and control of smart grid, smart buildings, intelligent transportation systems, mobile robotics,etc;
(b) providing rich and redundant information for executing safety-critical tasks in e.g. aerospace systems, nuclear power generation;
(c) offering information to the citizens and the government agencies for prompt decision making, especially in emergency situations.
Sensor faults can provoke severe consequences, leading to system instability, loss of information fidelity, wrong decisions and disorientation of remedial actions, which can even jeopardize human life.
Among the key priorities in designing smart cities is the reliability of sensor-based information, necessary for the safety and fault tolerance of various CPS. The primary contribution of this thesis will be the design of non-centralized model-based sensor fault diagnosis (SFD) architectures for detecting and isolating multiple sensor faults in a network of CPS. SFD methods that rely on mathematical models derived using first prinicples can be very efficient, especially for fault isolation compared to data-driven (model free) techniques that require huge amount of historical faulty data.
Due to the large-scale and interconnected nature of CPS, distributed and decentralized sensor fault diagnosis architectures can significantly contribute to the reduction of computational complexity and communication capacity, as well as scalability, since they are deployed such that local monitoring agents perform diagnosis in smaller parts of large-scale CPS. The secondary contribution of this thesis will be the design of fault tolerance agents for compensating sensor fault effects in CPS by reconstructing or correcting the output of isolated faulty sensors.

Brief scientific description
The thesis will address two methodological problems :

  1. Sensor fault diagnosis, by designing model based techniques for multi-sensor dynamical CPS. Among the important design aspects will be: (i) the distributed/decentralized deployment of local monitoring agents, (ii) the formulation of decision logic based on residuals (the difference of signals between the process/system and the model) and adaptive thresholds for robustness with respect to modelling errors, disturbances and noise methods, (iii) the management of interconnections, including mechanisms for handling time delays due to the communication,
  2. Sensor fault accommodation, by designing appropriate actions for reducing the effects of sensor faults multi-sensor dynamical CPS. Among the important design aspects will be: (i) the development of reconfiguration strategies (sensor fusion, switching) based on the fault isolation outcome for supporting the decision makers, (ii) the design of virtual sensor agents by aggregating the data of healthy sensors and using analytical redundancy relations.

From the theoretical point of view, the thesis will benefit from the advancements in set theoretic methods for fault detection and isolation and will contribute to this line of research with an emphasis on interconnected systems with multiple sensors.

Candidate profile

The ideal candidate will have (i) a background in control engineering (system modeling, control design), (ii) solid programming skills (e.g Matlab/Simulink), (iii) communication skills and a flexible approach in tackling a wide variety of tasks.

Skills aquired during the Phd work and collaborations

The PhD student will gain a solid expertise in designing fault diagnosis and fault tolerance techniques and a strong experience in a real-world application. He/she will learn to conduct innovative research, work methodologically and communicate research results.
The research work will be enhanced by the involvement of the supervisors in the EU project FUTuRISM.

1. R. Iserman, Fault-Diagnosis Systems: An Introduction from Fault Detection to Fault Tolerance. Springer, 2006
2. F. Stoican and S. Olaru, Set Theoretic Fault Tolerant Control in Multisensor Systems . Wiley & Sons, Inc., 2013
3. V. Reppa, M. Polycarpou, and C. G. Panayiotou, “Adaptive approximation for multiple sensor fault detection and isolation of nonlinear uncertain systems," IEEE Trans. on Neural Networks and Learning Systems, pp. 137-153, 2013.
4. V. Reppa, M. M. Polycarpou, and C. G. Panayiotou, “A distributed detection and isolation scheme for multiple sensor faults in interconnected nonlinear systems,” in 52nd Conference on Decision and Control (CDC2013), Florence, Italy, 2013, pp. 4991–4996.