Inverse scattering algorithms for nondestructive testing

Abstract

This thesis describes the implementation of the ultrasonics system hardware for the nondestructive testing problem, inverse scattering algorithms, and a block Schur algorithm for vector computers with hierarchical memory. The problem of the nondestructive testing is, in principle, identical to the problem of identifying the transfer function of the system from the known input and output signals. The system identification approach chosen for this research is also known as the deconvolution to electrical engineers, the layer peeling to geophysicists and the inverse scattering to physicists. The ultrasonics system built for the nondestructive testing application has the A/D converter (ADC 0804) of 8 bits with sampling period of about 170 ㎲, and the transducer with the nominal frequency of 40kHz and 24V peak-to-peak applied voltage. The inverse scattering problem is tackled with the Schur algorithm which is now widely known as a fast method of obtained for the triangular factorization of Toeplitz matrices. In this thesis it is also shown that the block Schur algorithm in which operations are performed on submatrices rather than individual matrix elements, is faster than the (scalar) Schur algorithm for the vector machine with hierarchical memory structure. Detailed test results are presented for a CRAY Y-MP vector computer.