Skip to main content

NEURAL NETWORK INVOLVING NEW SIGNAL PROCESSING PROCEDURES FOR FAULT ASSESSMENT IN CIVIL ENGINEERING STRUCTURES

Objective

This project is directed toward the development and application of neural networks new signal processing procedures in a novel approach to detecting and characterizing faults and other nonlinear mechanisms related to defects or structural damage.
This project is concerned with the development and application of neural networks new signal processing procedures in a novel approach to detecting and characterizing faults and other nonlinear mechanisms related to defects or structural damage.

The major difference in this work with respect to existing methodologies is the preprocessing of the raw data to enhance training procedures. Complex behaviour as in neurological systems has been successfully treated by using the digital equivalent of the human brain (ie neural networks). These can be trained to recognize highly complex patterns and therefore should have the capability of recognizing behaviour such as stress wave emissions from crack propagation or similar phenomena, hence they have the potential for use in identifying and distinguishing fault conditions. The establishment of the principles enables a significant advance in the technology for detecting and characterizing faults.

A new methodology for fault assessment in civil engineering structures has been developed. The results demonstrate the importance of separating the discriminatory features in the raw transducer time signals from the pattern classification in the form of a knowledge hierarchy. By doing this one can exploit the best aspects of temporal data preprocessing with the best neural classification procedures. This adds engineering insight and maximizes the probability of detecting faults. Feature extraction based on narmax data processing, curvature data and response only data appears to provide excellent signatures which can be uded to detect and discriminate nonlinear effects and faults.

The fault assessment procedures offer considerable potential and warrant further development and refinement to produce commercial products.
Although the study is directed toward components used in civil engineering structures in order to take advantage of specific competence in these areas, it is fundamental in its nature and therefore is directed at establishing principles that will result in applications to other key sectors of industry. The major difference in this work with respect to existing methodologies is the pre-processing of the raw data to enhance "training" procedures. Complex behaviour as in neurological systems has been succsfully treated by using the digital equivalent of the human brain, i.e. neural networks. These can be "trained" to recognise highly complex patterns and therefore should have the capability of recognising behaviour such as stress wave emissions from crack propagation or similar phenomena, hence they have the potential for use in identifying and distinguishing fault conditions. The establishment of the principles will enable a significant advance in the technology for detecting and characterizing faults.SP Such technologies could gainfully be developed for use in key sectors of the Aerospace industry, Civil Engineering Infrastructural systems (bridges/monumental buildings), and Manufacturing and Automobile industries, by initiating a programme of research involving of research involving both theoretical and experimental studies based upon new processing and neural network methodologies.

Coordinator

University of Manchester
Address
Simon Building Oxford Road
M13 9PL Manchester
United Kingdom

Participants (2)

University of Sheffield
United Kingdom
Address
Mappin Street
S1 4DU Sheffield
Università degli Abruzzi 'Gabriele d'Annunzio'
Italy
Address
Viale Pindaro 42
65127 Pescara