Neural networks for ultrasonic grain size discrimination

Author

Unluturk, M. ; Saniie, J.

Author_Institution

Dept. of Electr. & Comput. Eng., Illinois Inst. of Technol., Chicago, IL, USA

Volume

1

fYear

1996

fDate

3-6 Nov 1996

Firstpage

669

Abstract

In this study, we have developed the grain power spectrum neural network (GPSNN) to classify the ultrasonic backscattered grain signals for material characterization. The GPSNN has 32 input nodes, 13 hidden neurons determined adaptively, and one summing output node. A set of 4490 training sequences is utilized to train the neural network. A new set of 12572 testing sequences is used to test GPSNN performance. The samples tested for grain size discrimination are steel with grain sizes of 14 and 50 microns. GPSNN achieves an average recognition performance of over 98%. This high level of recognition suggests that the GPSNN is a promising method for ultrasonic nondestructive testing

Keywords

acoustic signal processing; adaptive signal processing; backpropagation; backscatter; grain size; neural nets; physics computing; ultrasonic materials testing; ultrasonic scattering; backpropagation learning algorithm; grain power spectrum neural network; hidden neurons; material characterization; recognition performance; signal classification; steel; summing output node; testing sequences; training sequences; ultrasonic backscattered grain signals; ultrasonic grain size discrimination; ultrasonic nondestructive testing; Acoustic scattering; Frequency estimation; Grain size; Microstructure; Neural networks; Neurons; Rayleigh scattering; Signal design; Testing; Ultrasonic variables measurement;

fLanguage

English

Publisher

ieee

Conference_Titel

Ultrasonics Symposium, 1996. Proceedings., 1996 IEEE

Conference_Location

San Antonio, TX

ISSN

1051-0117

Print_ISBN

0-7803-3615-1

Type

conf

DOI

10.1109/ULTSYM.1996.584063

Filename

584063