Fault diagnosis using qualitative and quantitative methods.

Format:

Book

Thesis/Dissertation

Author/Creator:

Vinson, Jonathan Martin.

Contributor:

Ungar, Lyle H., advisor.

University of Pennsylvania.

Language:

English

Subjects (All):

Computer science.

Artificial intelligence.

Chemical engineering.

0542.

0800.

0984.

Penn dissertations--Chemical engineering.

Chemical engineering--Penn dissertations.

Local Subjects:

Penn dissertations--Chemical engineering.

Chemical engineering--Penn dissertations.

0542.

0800.

0984.

Physical Description:

123 pages

Contained In:

Dissertation Abstracts International 56-05B.

System Details:

Mode of access: World Wide Web.

text file

Summary:

The focus of this research has been on developing on-line fault detection and diagnosis systems for the chemical process industries, for the purposes of plant safety and quality control. The result is four prototype systems: QMI, Q scMIMIC, APE and a PLS-based system. Each of these systems was designed with the intention of being able to detect and diagnose faults during both normal and transient process operations, and to continue monitoring the plant after a process upset. QMI and Q scMIMIC are based on qualitative models in the Q scSIM framework, and were designed with the additional goal of pushing the limits of qualitative reasoning technologies.

In QMI (Qualitative Modeling and Interpretation), fault detection and diagnosis are conducted simultaneously by comparison of a qualitative interpretation of the observations to several models of the plant. The comparison is based on fuzzy logic tests, since the predictions are qualitative.

Q scMIMIC uses qualitative models supplemented with quantitative information to provide predictions of the plant. Fault detection is conducted with single-sensor tests based on the Student's t-statistic. Q scMIMIC exploits a hypothesize-and-test algorithm for fault diagnoses.

APE (The Automatic Process Evaluator) is similar to Q scMIMIC, but uses fully quantitative, rather than qualitative, models and multivariable statistical tests for fault detection. For diagnosis, APE tests all user-defined fault models against the plant data, selecting the best-fitting model(s) as the diagnosis.

The PLS-based system uses statistical tests based on historical data for detection, and diagnosis is conducted by comparison of plant data to stored fault data. It is effective for detecting changes from known steady states, but not otherwise.

These systems have been tested on a simulation of a propylene glycol continuous stirred reactor, and on a larger simulation of a chemical plant provided by Tennessee Eastman. QMI and Q scMIMIC operate well on the small reactor, but the larger system is beyond the current capabilities of qualitative simulation. APE works well both on the reactor and the Tennessee Eastman system. The PLS-based system is extremely fast for fault detection in the Tennessee Eastman case study, and the diagnosis algorithm has accuracy comparable to other research on the same case study.

Notes:

Thesis (Ph.D. in Chemical Engineering) -- University of Pennsylvania, 1995.

Source: Dissertation Abstracts International, Volume: 56-05, Section: B, page: 2756.

Supervisor: Lyle H. Ungar.

Local Notes:

School code: 0175.

Access Restriction:

Restricted for use by site license.