Damage identification in beams using acousto-ultrasonics and artifical neural network

Flick, Jason R (2005). Damage identification in beams using acousto-ultrasonics and artifical neural network B.Sc Thesis, School of Engineering, The University of Queensland.

Attached Files (Some files may be inaccessible until you login with your UQ eSpace credentials)
Name Description MIMEType Size Downloads
Flick_Jason_R_THE19135.pdf Full text application/pdf 879.90KB 0
Author Flick, Jason R
Thesis Title Damage identification in beams using acousto-ultrasonics and artifical neural network
School, Centre or Institute School of Engineering
Institution The University of Queensland
Publication date 2005
Thesis type B.Sc Thesis
Supervisor Associate Professor Martin Veidt
Total pages 136
Language eng
Subjects 290501 Mechanical Engineering
Formatted abstract

Non Destructive Testing (NDT) is a commonly used technique for identifying flaws in beams. In particular Acousto-ultrasonics is a method using guided waves to avoid the standard practice of scanning with an ultrasonic probe. Incorporating this technique and intelligent signal processing of a single signal and its reflections within a beam to characterise flaws in a specimen, the flaws will be identified in the beam. This technique, although being time optimal, requires operating technicians to undergo extensive training if a reliable and accurate result is to be attained.

Incorporating Artificial Neural Network (ANN) technology with acousto-ultrasonic NDT techniques is one way of solving this problem. In this paper an ANN approach to acousto-ultrasonic NDT is taken. In particular, the development of the ANN model is studied in depth to identify network structures, topologies and supervised learning algorithms that not only optimise performance of a designed network on a training set, but also minimise error when tested on a test set.

Analysis finds that using a feed-forward back propagation algorithm with 2 layers attains an optimally performing network. Results show that using ANN to identify laminar flaws in beams is a good approach when estimating flaw location and the impedance in the damaged region, but when characterising the size of the damaged region the network struggles to achieve an adequate response. This is especially seen when considering situations where the boundaries of the networks training are extended and extrapolation is required of the network.

Keyword Non destructive testing
Incorporating artificial neural network
Additional Notes * Mechanical engineering undergraduate theses. Sem 2, 2005

Document type: Thesis
Collection: UQ Theses (non-RHD) - UQ staff and students only
Citation counts: Google Scholar Search Google Scholar
Created: Tue, 07 May 2013, 12:30:13 EST by Mr Yun Xiao on behalf of Scholarly Communication and Digitisation Service