TY - JOUR
T1 - Three-dimensional inversion of volumetric defects profiles from electromagnetic nondestructive testing signals by means of stochastic methods with the aid of parallel computation
AU - Yusa, Noritaka
AU - Rebican, Mihai
AU - Chen, Zhenmao
AU - Miya, Kenzo
AU - Uchimoto, Tetsuya
AU - Takagi, Toshiyuki
PY - 2005/2
Y1 - 2005/2
N2 - This study applies five stochastic inversion methods to the reconstruction of three-dimensional volumetric defect regions from eddy current testing signals, and evaluates their efficiencies in relation to this problem. The five stochastic methods considered here are iterative local search, tabu search, simple genetic algorithm, parameter-free genetic algorithms, and simulated annealing. Since stochastic methods require evaluation of many solution candidates, simulations here are performed on a supercomputer aided by parallel computation to avoid too lengthy computational times. Three-dimensional volumetric defects in a thin flat conductive plate are considered. The results of this study demonstrate that the stochastic methods applied to eddy current inversion problems are highly compatible with parallel computation, and that computational time can be significantly shortened with the use of parallel computation. Among the five algorithms adopted in this study, tabu search and simulated annealing provided good results whereas genetic algorithms had very poor performances.
AB - This study applies five stochastic inversion methods to the reconstruction of three-dimensional volumetric defect regions from eddy current testing signals, and evaluates their efficiencies in relation to this problem. The five stochastic methods considered here are iterative local search, tabu search, simple genetic algorithm, parameter-free genetic algorithms, and simulated annealing. Since stochastic methods require evaluation of many solution candidates, simulations here are performed on a supercomputer aided by parallel computation to avoid too lengthy computational times. Three-dimensional volumetric defects in a thin flat conductive plate are considered. The results of this study demonstrate that the stochastic methods applied to eddy current inversion problems are highly compatible with parallel computation, and that computational time can be significantly shortened with the use of parallel computation. Among the five algorithms adopted in this study, tabu search and simulated annealing provided good results whereas genetic algorithms had very poor performances.
KW - Eddy current testing
KW - Inversion
KW - Non-destructive testing
KW - Parallel computing
KW - Volumetric defects
UR - http://www.scopus.com/inward/record.url?scp=12344283232&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=12344283232&partnerID=8YFLogxK
U2 - 10.1080/10682760410001710150
DO - 10.1080/10682760410001710150
M3 - Article
AN - SCOPUS:12344283232
VL - 13
SP - 47
EP - 63
JO - Inverse Problems in Engineering
JF - Inverse Problems in Engineering
SN - 1741-5977
IS - 1
ER -