08.11.2014 Views

Probabilistic Performance Analysis of Fault Diagnosis Schemes

Probabilistic Performance Analysis of Fault Diagnosis Schemes

Probabilistic Performance Analysis of Fault Diagnosis Schemes

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Abstract<br />

<strong>Probabilistic</strong> <strong>Performance</strong> <strong>Analysis</strong> <strong>of</strong> <strong>Fault</strong> <strong>Diagnosis</strong> <strong>Schemes</strong><br />

by<br />

Timothy Josh Wheeler<br />

Doctor <strong>of</strong> Philosophy in Engineering–Mechanical Engineering<br />

University <strong>of</strong> California, Berkeley<br />

Pr<strong>of</strong>essor Andrew K. Packard, Co-chair<br />

Pr<strong>of</strong>essor Peter J. Seiler, Co-chair<br />

The dissertation explores the problem <strong>of</strong> rigorously quantifying the performance <strong>of</strong> a fault<br />

diagnosis scheme in terms <strong>of</strong> probabilistic performance metrics. Typically, when the performance<br />

<strong>of</strong> a fault diagnosis scheme is <strong>of</strong> utmost importance, physical redundancy is used<br />

to create a highly reliable system that is easy to analyze. However, in this dissertation, we<br />

provide a general framework that applies to more complex analytically redundant or modelbased<br />

fault diagnosis schemes. For each fault diagnosis problem in this framework, our<br />

performance metrics can be computed accurately in polynomial-time.<br />

First, we cast the fault diagnosis problem as a sequence <strong>of</strong> hypothesis tests. At each<br />

time, the performance <strong>of</strong> a fault diagnosis scheme is quantified by the probability that<br />

the scheme has chosen the correct hypothesis. The resulting performance metrics are<br />

joint probabilities. Using Bayes rule, we decompose these performance metrics into two<br />

parts: marginal probabilities that quantify the reliability <strong>of</strong> the system and conditional<br />

probabilities that quantify the performance <strong>of</strong> the fault diagnosis scheme. These conditional<br />

probabilities are used to draw connections between the fault diagnosis and the fields <strong>of</strong><br />

medical diagnostic testing, signal detection, and general statistical decision theory.<br />

Second, we examine the problem <strong>of</strong> computing the performance metrics efficiently<br />

and accurately. To solve this problem, we examine each portion <strong>of</strong> the fault diagnosis<br />

problem and specify a set <strong>of</strong> sufficient assumptions that guarantee efficient computation. In<br />

particular, we provide a detailed characterization <strong>of</strong> the class <strong>of</strong> finite-state Markov chains<br />

that lead to tractable fault parameter models. To demonstrate that these assumptions enable<br />

efficient computation, we provide pseudocode algorithms and prove that their running time<br />

is indeed polynomial.<br />

Third, we consider fault diagnosis problems involving uncertain systems. The inclusion<br />

<strong>of</strong> uncertainty enlarges the class <strong>of</strong> systems that may be analyzed with our framework. It<br />

also addresses the issue <strong>of</strong> model mismatch between the actual system and the system used<br />

1

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!