Asee peer logo

Using A Neural Network To Identify Flaws During Ultrasonic Testing

Download Paper |

Conference

2002 Annual Conference

Location

Montreal, Canada

Publication Date

June 16, 2002

Start Date

June 16, 2002

End Date

June 19, 2002

ISSN

2153-5965

Conference Session

Lab Experiments in Materials Science

Page Count

15

Page Numbers

7.1246.1 - 7.1246.15

Permanent URL

https://peer.asee.org/10697

Download Count

118

Request a correction

Paper Authors

author page

Glenn Kohne

Download Paper |

Abstract
NOTE: The first page of text has been automatically extracted and included below in lieu of an abstract

Main Menu Session 3264

Case Study: Using a Neural Network to Identify Flaws during Ultrasonic Testing

A. Kayabasi, G. S. Kohne and P. J. Coyne, Jr.

Loyola College in Maryland Department of Electrical Engineering and Engineering Science Baltimore MD 21210-2699

Abstract: A feed forward neural network with a single hidden layer was used to identify a series of cylindrical samples based on the first ultrasonic echo. The simulated flaws were placed at varying distances directly in front of a 1 MHz broadband transducer with water acting as the medium. This case study will explore the decision process required for selecting the form of the input data presented to the neural network, the topology of the network, and the encoding of the desired output of the network, which must be identified prior the training phase of the network. Results will be reported for three variations in training and testing using a data set consisting of fifty-seven return echoes from five different diameter samples. Additionally, the role of the number of nodes in the hidden layer on the learning rate will be explored. A windows based implementation of the neural network, detailed in the paper, will be available for conference attendees to take home and apply to their problems.

Introduction: A neural network or more formally an artificial neuronal network involves the implementation, either in hardware or software, of a parallel distributed processing computation structure or neurocomputers that is capable of tasks such as pattern recognition in nondestructive testing, which is the example chosen for this case study. These neurocomputers can function similarly to the human brain, although they are not limited to this function. It is estimated that the human brain is composed of 10 11 neurons; each connected to 10 4 other neurons [1], which implement a massively parallel computation structure capable of very complex tasks, some still far beyond the capabilities of the today’s computers.

The neuron is the fundamental processing unit in the brain and in the artificial neuronal network. Each neuron, Figure 1, receives inputs from many other neurons or from sensory processing elements. Each input is weighted, positively or negatively, by a connection strength and then combined or summed. When the weighted-combined inputs exceed a threshold, the neuron fires or activates its output line, which is then used as the input to another neuron. The key to this parallel processing unit is the connecting weights, which have been arrived at as a

Proceedings of the 2002 American Society for Engineering Education Annual Conference & Exposition Copyright Ó 2002, American Society for Engineering Education

Main Menu

Kohne, G. (2002, June), Using A Neural Network To Identify Flaws During Ultrasonic Testing Paper presented at 2002 Annual Conference, Montreal, Canada. https://peer.asee.org/10697

ASEE holds the copyright on this document. It may be read by the public free of charge. Authors may archive their work on personal websites or in institutional repositories with the following citation: © 2002 American Society for Engineering Education. Other scholars may excerpt or quote from these materials with the same citation. When excerpting or quoting from Conference Proceedings, authors should, in addition to noting the ASEE copyright, list all the original authors and their institutions and name the host city of the conference. - Last updated April 1, 2015