| In the context of complex plants, a fault diagnosis system for on-line applications must provide guaranteed response times completing the diagnosis in a deterministic amount of time, it must be able to reason about time because much information can be deducted from time events, and it must use all the information available for improving and speeding the diagnosis. The fundamental issues of how to model the system and develop a diagnostic procedure arise in this context. In the case of complex system, the application of model-based FDI algorithms becomes computationally very difficult or even intractable.; In this dissertation, we present a new framework for hierarchical model-based fault diagnosis in complex systems that combines qualitative and quantitative methodologies and allows great flexibility, modularity, and fault isolation with a reduced computational effort.; The presented framework is structurally divided into a passive and an active components. The model-based passive component represents knowledge about the fault behavior of the system under analysis, and consists of a Hierarchical Fault Propagation Digraph and a Hierarchical Model-Based FDI scheme. The active component is constituted by a Fault Location and Isolation Process and a Supervisory Process that use the model-based knowledge for diagnosis. The fundamental novelties in the presented method with respect to the past are related to the use of fault propagation digraphs having weighted arcs with fault propagation probabilities and upper/lower bounds on fault propagation times, and nodes corresponding to active/activatable fault detection units. The introduction of these properties and components in the digraph, not only makes the proposed scheme more valuable than approaches as FMEA, event trees and FTA, but also allows the use of model-based FDI to compensate for the absence of system dynamics modeling, making the scheme more simple, attractive and powerful than signed digraphs.; The combination of a quantitative methodology with a qualitative methodology constitutes the innovative idea in this dissertation and contributes to fill one of the biggest gaps in the current literature on fault diagnosis.; The methodology is applied to a vehicle brake-by-wire and steer-by-wire systems with simulations and experimental results that shown the validity of the presented approach. |
