Use this URL to cite or link to this record in EThOS:
Title: The characterisation of multiple defects in components using artificial neural networks
Author: Farley, Simon John
ISNI:       0000 0004 2708 5737
Awarding Body: Oxford Brookes University
Current Institution: Oxford Brookes University
Date of Award: 2011
Availability of Full Text:
Access from EThOS:
Access from Institution:
This thesis investigates the use of artificial neural networks (ANNs) as a means of processing signals from non-destructive tests, to characterise defects and provide more information regarding the condition of the component than would otherwise be possible for an operator to obtain from the test data. ANNs are used both as pattern classifiers and as function approximators. In the first part of the thesis, finite element analysis was carried out on a simple component containing a single defect modelled as a void, simulating three kinds of non-destructive test: an impact method that sent a stress wave through the component, an analysis of natural frequencies, and an ultrasonic pulse-echo method. The inputs to the ANNs were data from the numerical model, and the outputs were the x and y co-ordinates of the defect in the case of the impact and frequency methods, and the size and distance to the defect in the case of the ultrasonic method. Very good accuracy was observed in all three methods. Experimental validation of the ultrasonic method was carried out, and the ANNs returned accurate outputs for the position and size of a circular hole in a steel plate when presented with experimental data. When the ANNs were presented with noisy input data, their reduction in accuracy was small in comparison with published data from similar studies. In the second part of the thesis, the case of two defects lying within one wavelength of each other was considered, where the reflected ultrasonic waves from each defect overlapped, partially cancelling each other out and reducing the overall amplitude. A novel ANN-based approach was developed to decouple the overlapping signals, characterising each defect in terms of its position and size. Optimisation of the ANN architecture was carried out to maximise the ability of the ANN to generalise when presented with previously unseen data. Finally, an ANN-based general defect characterisation ‘expert system’ is presented, using data from an ultrasonic test as its input, and classifying cases according to the number of defects present. The system then characterised the defects present in the component in terms of their location and size, providing more information regarding the component’s condition than would be possible by existing techniques.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available