Accession Number:

ADA341901

Title:

Damage Detection Using Pattern Classifiers

Descriptive Note:

Master's thesis

Corporate Author:

AIR FORCE INST OF TECH WRIGHT-PATTERSONAFB OH SCHOOL OF ENGINEERING

Personal Author(s):

Report Date:

1998-03-01

Pagination or Media Count:

96.0

Abstract:

The research focused on developing and tuning finite element models to train pattern classifiers to detect and locate damage in a real structure. The research was broken into three distinct phases finite element FE model development, FE model tuning, and pattern classifier training and testing. In the finite element development phase, a low order FE model called the baseline model and a high order model called the stiff model were created. In the FE model tuning phase, these FE models were tuned using measured Frequency Response Functions FRFs, and the results were compared with previous research in which tuning was accomplished using using modal data. In the pattern classifier training and testing phase, the tuned models were used to generate FRFs to train various pattern classifiers. Features or properties of the FRFs were extracted through an adapted feature extraction process commonly used in speech processing. This new feature set was developed to reduced the amount of data by a factor of 40 while retaining the salient properties that made the changes in the FRFs unique to each damage state. The method was tested on the Flexible Truss Experiment FTE at the Air Force Institute of Technology AFIT. The FE models were developed and tuned inthe Structural Dynamics Toolbox trade mark for MATLABtrade mark. To prove that the different features extracted from 32 damage states were unique, some initial tests were performed in which five classifiers were trained and tested using measured data. These tests resulted in no classification errors. Since the different damage states produced unique feature vectors, the majority of the research effort was spent developing and tuning different FE models that are then used to train five pattern classifiers to detect damage.

Subject Categories:

  • Infrared Detection and Detectors

Distribution Statement:

APPROVED FOR PUBLIC RELEASE