Use this URL to cite or link to this record in EThOS:
Title: Investigation and application of anomaly detection methods for industrial gas turbines
Author: Daithi, Norman Lee
Awarding Body: University of Manchester
Current Institution: University of Manchester
Date of Award: 2002
Availability of Full Text:
Access from EThOS:
This thesis initially considers the commercial and practical aspects of the problem. being addressed. Several different modelling methods were then assessed including linear, nonlinear, static and dynamic methods in order to accurately capture a description of the normal operation of a family of industrial gas turbines. Most of the methods that were assessed were found to be unsuitable, in large part due to the nature of the data available. The method that was found to perform best, because it could meet all of the required objectives within a single methodology, was the Self-Organizing Map (SOM). A comprehensive evaluation of the SOM was carried out using data from several commercial gas turbines. Some difficulties in the application of the 80M were encountered. These were largely overcome using empirical studies and a novelty threshold. A computer package was developed using the SOM to detect novelty and provide the user with the tools necessary to proceed from novelty to anomaly detection with the minimum of effort. An alternative method to create SOM-like maps was then developed using existing optimisation techniques rather than the heuristic methods associated with the conventional SOM algorithm.. The resulting maps were found to have several properties that were superior to the conventional SOM. These included being more effective in both novelty detection and in constructing meaningful topology preserving maps of the process.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (D.Eng.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available