Share Email Print

Proceedings Paper

Processing signals for damage detection in structures using neural networks
Author(s): Julian E. Chance; Keith Worden; Geoffrey R. Tomlinson
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

The effective use of neural networks for fault detection, location, and classification requires training data. Distinguishing features in the data can be enhanced by pre-processing. Feature vectors that often prove advantageous in training networks for fault detection are modeshapes and curvatures; however, the procedures and sensors used to determine these can introduce problems. This paper uses numerical and practical experiments to investigate the use of acceleration, displacement, and strain response signals to extract modeshape and curvature functions from a cantilever plate and beam with localized damage. The importance of spatial accuracy, noise, and fault severity for fault detection is studied. It is shown that limiting spatial conditions occur with direct dynamic displacement measurements that have to be differentiated to obtain curvatures that can be overcome by using strain gauges that directly return quantities proportional to the curvature.

Paper Details

Date Published: 1 May 1994
PDF: 12 pages
Proc. SPIE 2191, Smart Structures and Materials 1994: Smart Sensing, Processing, and Instrumentation, (1 May 1994); doi: 10.1117/12.173946
Show Author Affiliations
Julian E. Chance, Univ. of Manchester (United Kingdom)
Keith Worden, Univ. of Manchester (United Kingdom)
Geoffrey R. Tomlinson, Univ. of Manchester (United Kingdom)

Published in SPIE Proceedings Vol. 2191:
Smart Structures and Materials 1994: Smart Sensing, Processing, and Instrumentation
James S. Sirkis, Editor(s)

© SPIE. Terms of Use
Back to Top