| Title | : | System Health Management |
| Speaker | : | Sriram Narasimhan (IBM Almaden) |
| Details | : | Thu, 21 Aug, 2014 11:00 AM @ BSB 361 |
| Abstract: | : | System health management deals with supporting the operation of a system in the presence of faults in one or more components (hardware and software) of the system being monitored. Tasks involved include diagnosis (fault detection, isolation and identification), prognosis (fault impact analysis, end of life prediction), decision making (fault mitigation and recovery). The end goal is to maximize the achievement of original system objectives while maintaining the requirements of safety and cost. Model-based diagnosis techniques use structural, behavioral and/or functional models of the system to detect and isolate faults. Given the wide variety of faults (component/sensor, discrete/continuous, persistent/intermittent etc.) that occur in real systems and the presence of uncertainties (sensor noise, unknown environment etc.) no single technique can handle all situations effectively and efficiently. In this talk I will present 2 model-based diagnosis techniques (FACT and HyDE) that deal with entirely different situations. FACT uses a formal modeling framework and diagnoses single parametric faults (abrupt change in parameter value) while HyDE uses a more general modeling framework to diagnosis multiple discrete faults (abrupt change in state of component). These 2 approaches are just a couple of drops in the large ocean of techniques proposed by researchers in the model-based and other (data-driven for example) communities. I will also present work I was involved in that attempted to develop a standard set of metrics to evaluate and compare various diagnosis methodologies. This work also resulted in the development of a framework to execute diagnostic applications on a common platform which led to a series of international diagnostic competitions. Finally I will present some applications at NASA that are using/have used model-based reasoning applications. |
