Discrete event systems (DESs) represent many important engineering systems such as communication networks, computer networks, manufacturing systems, and transportation systems. State variables in such systems change their values only at discrete instants of time, and more importantly such changes are driven by occurrence of discrete-events. DESs are recognized as a new breed of dynamic systems that are structurally different from the conventional models characterized by ordinary differential equations. The control of DESs called supervisory control is recognized as one of effective tools for the analysis and control of large scale dynamic systems which are one of the biggest challenges facing control engineers today. The goal of supervisory control is to achieve desirable logical performances
for a given system through feedback control. Of particular interests are model uncertainty and robust control problems which provide a key to solve practical problems. However, it is difficult to approach them using tools from the existing literature taking no account of model uncertainty.
In this dissertation we study various schemes on model uncertainty and robust control problems of DESs. First, as a framework on model uncertainty, we introduce multiple model representation where a DES is modeled as a set of some possible models. In the scheme we address a fault-tolerant control problem of DESs under partial observation. As a result, we present the conditions for the existence of a robust supervisor which guarantees a fault-tolerant behavior for any model in the set.
Second we introduce nondeterministic models describing nondeterminism in the transition which implies the unobservable aspect or lack of the information of system behavior. In the scheme we handle nondeterministic models using multiple deterministic model approach and language models approach subject to a nonblocking control problem. In the multiple deterministic model approach, it is shown that a supervisor to satisfy certain properties for multiple deterministic models also guarantees the nonblockingness for an original nondeterministic model. Moreover, in the language models approach, we present the existence conditions of a nonblocking supervisor for DESs modeled as nondeterministic automata with θ-moves. Furthermore, for the uncertain DES modeled as a set of some possible nondeterministic automata, based on language models and prioritized synchronous compositions, we present the existence conditions of a robust nonblocking supervisor to guarantee the nonblockingness of any nondeterministic automata of the set.
Finally we develop a systematic method to reflect the existence of the internal and unobservable events in modeling. By such events, systems are supposed to experience at most some positive value step state transition. In the scheme we also address a nonblocking control problem. As a result, we present the existence conditions of a robust supervisor to guarantee the nonblockingness of a system despite of the internal and unobservable events. Moreover, under the assumption that for some states, the internal and unobservable events may occur but the assigned positive integer values for the states are not known exactly, we present the existence conditions of a robust supervisor with the nonblockingness.
Also, in this dissertation we deal with decentralized and hierarchical supervisory control problems of uncertain DESs which are represented as a set of some possible models. For a given global specification, we provide the existence conditions of local supervisors to achieve the specification under model uncertainty. Moreover, to achieve the hierarchical consistency under model uncertainty, we derive the conditions on the properties of the information channel between a low level and a high level and the structure of the low level system.
Lastly, we formulate an analytical framework for supervisory control of real-time DESs under bounded time constraints. In order to address the bounded time constraints of the systems, we introduce the timed languages based on timed transition models. Based on the notion, we present the existence conditions of nonblocking supervisors for real-time deterministic DESs and real-time nondeterministic DESs to achieve given timed language specifications.
이산 사건 시스템의 관리 제어 이론은 생산 시스템, 통신망 등의 현대의 복잡한 첨단 시스템의 체계적 분석과 제어를 위한 방법론으로 시스템 동작에 대한 상위레벨 관점에서의 추상화 모델에 기반한다. 본 논문에서는 특히 이산 사건 시스템에서 나타나는 다양한 형태의 모델 불확정성을 소개하였고, 불확정성 하에서의 강인 관리 제어 문제들을 다루었다. 먼저 기존의 불확정성 모델링 기법들을 소개하고 이들이 다루지 않은 내고장성 제어 기법과 기존의 기법들이 지닌 해석의 불완정성과 복잡성을 감소시키는 방법들을 제안하였다. 특히 제어 시스템의 비막힘성을 보장하는 관리제어기의 존재 조건들을 제안하였으며 주어진 사양의 부분 사양들 중 존재 조건을 만족하는 최대 사양을 찾는 방법들을 제안하였다. 제안된 이론적 결과들의 실제 시스템에의 적용 가능성을 보이기 위해 유연 생산 시스템 및 통신 프로토콜의 응용 예를 보였다. 또한 기존의 모델 불확정성 기법으로 모델링 할 수 없는 불확정성에 대처하기 위하여 본 논문에서는 새로운 형태의 모델 불확정성 틀을 제시하였는데, 이 기법은 시스템의 알려지지 않은 내부적인 상태 천이 양상을 효과적으로 모델링할 수 있다. 또한 제안된 새로운 모델 불확정성 하에서 비막힘성 관리 제어 문제를 정형화하고 관리제어기가 존재하기 위한 조건들을 제시하였다. 마지막으로 분산 제어, 계층 제어, 및 실시간 제어 문제 등의 시스템 제어의 몇가지 주요 분야들에서 모델 불확정성 문제를 소개하고 강인 관리제어기 존재 문제를 다루었다. 향후 연구 방향의 초점은 개발된 이론적 결과들을 실제 시스템에 보다 심도있게 응용하여 그 실용성을 검증하는데 모아질 것이다.
