PROCESS CONTROL SYSTEMS : DEPENDABILITY AND PERFORMANCE MODELING

Ricardo Messias Fricks

Today, a wide variety of computer systems are required to react automatically to various events generated by external processes and/or human operators. An important class of these process control systems, called hard real-time systems, need to respond to events under severe timing constraints expressed in terms of deadlines and earliest starting times. In general, missing deadlines is often as dangerous as producing incorrect results. Most of such real-time process control systems are also required to deliver correct service even in the presence of hardware faults and software errors in their components. The fault-tolerant attribute is essential when system failures may cause economic disasters or loss of human lives. The main purpose of this dissertation is to report research results related to the mathematical state space modeling of the three major characteristics of a process control system: timeliness, dependability, and external environment dependencies. Our results are presented in the context of Supervisory Control and Data Acquisition (SCADA) of power systems and Energy Management Systems (EMS). Particularly, SCADA systems are fault-tolerant, real-time process control systems hosted in distributed architectures composed of heterogenous machines, with a technology of widespread utilization that pervades several industrial control applications. We show how Markov reward models can be applied in the availability analysis of EMS computer architectures before proposing and purchasing any equipment. We also characterize a distributed architecture with associated fault-tolerant strategies, representative of the latest generation of master stations deployed in utility industry applications. We study the effect of failure and repair dependencies in a data acquisition computer model developed using stochastic Petri nets and continuous-time Markov chains. Special attention is given to the representation of external influences and human errors. Additionally, we introduce novel techniques for determining importance measures using Markov reward models. The advantage of bringing these measures to the context of Markov modeling is that the mapping extends the applicability of these substantial results used in criticality analysis of engineering systems. We propose an analytic approach based on Markov renewal theory to evaluate two real-time database architectures in a distributed SCADA system. We first construct exact performance models using Markov regenerative processes. Then, due to limitations on the exact solution, we propose an approximate solution technique (named delayed shadow server) which is then used to study the response time distribution of large SCADA systems under normal and emergency situations. Besides the modeling approach, we also develop the idea of incorporating human operator behavior in the performance models to add more flexibility and realism. An important side result of this research was the development of a new approach of graphically representing Markov chains. By adding new dimensions to the state transition diagram of a Markov chain we provide better visualization of model symmetries not apparent in a conventional planar representation. The dissertation concludes with the description of an integrated modeling environment we helped design. This environment aims to deliver a new generation of modeling tools with improved user interfaces.

PROCESS CONTROL SYSTEMS : DEPENDABILITY AND PERFORMANCE MODELING

AUTOR(ES)

FONTE

DATA DE PUBLICAÇÃO

RESUMO

ASSUNTO(S)

ACESSO AO ARTIGO

Documentos Relacionados