Automated fault detection, diagnosis and tolerance in uncertain systems and distributed systems, DPI2006-15476-C02-00, 2006-2009. This project is funded by the Spanish Ministry of Science and Education and ERFD/FEDER. Manager: Rafael M. Gasca
Researchers: Pedro Abad, Rafael Ceballos, María Teresa Gómez, Luis González, Antonio Márquez, Sergio Pozo, Fernando de la Rosa, Antonio J. Suárez, Miguel Toro Bonilla, Carmelo del Valle, Irene Barba, Diana Borrego, Pablo Neira
Project summary: Fault detection, diagnosis and tolerance in systems allow to naibtaib the process in a high level of production and reliability. The present techniques show problems when they are applied to uncertain and distributed systems. For this reason, they are being researched in a very active way to develop new techniques.
The most important objetives of this projects are to decrease de high false positive and false negative rates, to improve the computational efficiency of fault detection and diagnosis and to automate the identification of components that are the cause of the failures.
These problems for the uncertain systems can be treated using the classical diagnosis approaches: the FDI (Fault Detection and Isolation) community, whose foundation are based on engineering disciplines, such as control theory and statistical decision making, and the DX (Diagnosis) community, whose foundation are derived from the fields and of Computer Science and Artificial Intelligence and bridging the methodologies of the FDI and DX Communities (Bridge). Firstly the project proposes this last approach to automate fault diagnosis in these systems. The Research Group in Girona has developed the Squaltrack tool for the fault detection of these systems and this project proposes its extension for diagnosis with techniques devired from model interval analysis. We consided that it is important the coordination with the Quivir (Sevilla and Huelva University) Group that is also studying these same systems using constraint satisfaction techniques, guaranteed methods and data-mining techniques. Specially, the last techniques can work without the knowledge of the model and can deal with the empiric data (data-based diagnosis). These techniques are being validated in real systems (induction motors).
In distributed systems, firstly this projects addresses the general problems associated with automatic determination of the allocation of sensors to improve the diagnosticability of the distributed systems. Secondly, we will develop a framework for self-diagnosis of distributed systems where we take into account the persistency of the different models and different improvements based on clustering and distributed constraint satistaction techniques.
Finally, we consider that the dependability and fault tolerance are important in the systems. So, we propose the modelling of reliability and security properties and the identification of the faults in the monitoring systems. We will develop different algorithms that take this information and automatically oftain the fault diagnosis and choose recovery actions using planning and re-planning. It can hopefully provide a foundation for building realistic automatic dependability and tolerance systems.
| 