Hybrid estimation of complex systems. (Q2571137)

Modern technology is increasingly leading to complex artifacts with high demands on performance and availability. As a consequence, advanced automation and control methods play an important role in achieving these requirements. This monograph deals almost exclusively with monitoring and estimation within the context of automation for complex systems. The author describes a novel model-based approach for on-line estimation that can deal with the complexity of modern automation systems, whilst being able to deal with (partially) unspecified operational situations. More specifically, the author presents: \textit{Component-based modeling formalism:} In order to deal with complex multicomponent systems that exhibit hybrid discrete/continuous behaviors of their physical entities, the author chooses hybrid automata as general modeling paradigm. These automata describe physical systems in terms of a set of operational and fault modes, mode transitions, and a dynamic model for the physical entities for each specified mode. The novelty of this approach is that the author combines a probabilistic mode transition model with stochastic discrete-time difference equations to capture real-world effects such as disturbances, sensor noise, and other non-deterministic effects. This leads to what the author calls a probabilistic hybrid automaton that serves as the basic building block for the modeling paradigm. In terms of the overall model for a complex physical artifact, the author takes a component-based approach that models individual components of the physical artifact in terms of probabilistic hybrid automata and defines the interplay among them in terms of their concurrent composition. This leads to what the author call a concurrent probabilistic hybrid automaton. \textit{Focused hybrid estimation:} Estimation for hybrid systems is generally difficult. This is due to the fact that an estimator has to consider all possible mode sequences with their associated continuous evolutions. This fact has particularly severe implications whenever the number of operational modes of the underlying system is very large so that standard methods from the field of multiple-model estimation cannot be applied anymore. The proposed hybrid estimation method deals with this class of complex systems. It carefully explores possible estimation hypotheses and focuses onto the likely estimates. For this purpose the author carefully reformulates the hybrid estimation problem as a best-first search problem and utilizes advanced search techniques from the toolkit of Artificial Intelligence to solve this problem. This leads to an any-time/any-space hybrid estimation algorithm that is suitable for on-line execution within the context of an automation system. \textit{Automated on-line filter design:} The hybrid estimation system utilizes a set of (extended) Kalman filters, one for each mode under consideration, to anticipate the continuous evolution of the system. Instead of using a precompiled set of (hand-crafted) filters, the author provides a sophisticated model-based mechanism that deduces these filters on-line. This capability has two important implications. Firstly, there is no need to pre-compile a possibly prohibitively large number of filters. The limited computational resources of real-time systems, that execute the automation system, won't allow us to store and utilize them anyhow. Secondly, since the deduction process grounds upon the hybrid model for the physical artifact, we can easily update or modify the estimator to incorporate any modification of the physical artifact or components of thereof. The author simply updates the underlying component model. Hybrid estimation with the underlying automated filter design capability incorporates this change automatically. \textit{Robustness:} It is difficult, if not impossible, to anticipate all possible faults that can occur within a complex physical artifact. Un-anticipated situations do occur, and a wrong classification of thereof can cause severe implications such as the loss of control over a potentially dangerous automated system (chemical plant, power plant, airplane, automobile\(\dots\)). The proposed hybrid estimation scheme takes this into account and provides a generic unknown mode that captures all un-anticipated modes of behavior. In this way, the author obtains an estimation capability that can detect un-anticipated modes of operation, identify the impaired components or subsystem, and continues hybrid estimation in a degraded, but fail-safe manner.