BROWSE

ITEM VIEW & DOWNLOAD

A Study on Data Filtering Techniques for Event-Driven Failure Analysis

Cited 0 times inthomson ciCited 0 times inthomson ci
Title
A Study on Data Filtering Techniques for Event-Driven Failure Analysis
Author
Lee, Yang-Ji
Advisor
Kim, Duck Young
Issue Date
2013-02
Publisher
Graduate School of UNIST
Abstract
High performance sensors and modern data logging technology with real-time telemetry facilitate system failure analysis in a very precise manner. Fault detection, isolation and identification in failure analysis are typical steps to analyze the root causes of failures. This systematic failure analysis provides not only useful clues to rectify the abnormal behaviors of a system, but also key information to redesign the current system for retrofit. The main barriers to effective failure analysis are: (i) the gathered sensor data logs, usually in the form of event logs containing massive datasets, are too large, and further (ii) noise and redundant information in the gathered sensor data that make precise analysis difficult. Therefore, the objective of this thesis is to develop an event-driven failure analysis method in order to take into account both functional interactions between subsystems and diverse user’s behaviors. To do this, we first apply various data filtering techniques to data cleaning and reduction, and then convert the filtered data into a new format of event sequence information (called ‘eventization’). Four eventization strategies: equal-width binning, entropy, domain knowledge expert, and probability distribution estimation, are examined for data filtering, in order to extract only important information from the raw sensor data while minimizing information loss. By numerical simulation, we identify the optimal values of eventization parameters. Finally, the event sequence information containing the time gap between event occurrences is decoded to investigate the correlation between specific event sequence patterns and various system failures. These extracted patterns are stored in a failure pattern library, and then this pattern library is used as the main reference source to predict failures in real-time during the failure prognosis phase. The efficiency of the developed procedure is examined with a terminal box data log of marine diesel engines.
Description
Engineering & Systems Design
URI
Go to Link
Appears in Collections:
DHE_Theses_Master
Files in This Item:
A Study on Data Filtering Techniques.pdfDownload

find_unist can give you direct access to the published full text of this article. (UNISTARs only)

Show full item record

qr_code

  • mendeley

    citeulike

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

MENU