Nondestructive evaluation of steel structures using a superconducting quantum interference device magnetometer and a neural network system

This work combines two state-of-the-art techniques in the area of magnetic nondestructive evaluation: the application of the superconducting quantum i nterference device (SQUID) as the magnetic field sensor; and the use of art ificial neural networks as analysis tools applied to the detected magnetic signals. Pioneering measurements using the SQUID sensor have been made in s teel samples containing various types of flaws, and a neural network system , based on the time-delay neural network and radial basis function algorith ms, has been implemented to characterize the flaws. The neural network syst em aims to, based on the measured magnetic field, provide information about defect geometry, thus allowing the assessment of defect severity, as a bas is for maintenance and repair procedures. (C) 2000 American Institute of Ph ysics. [S0034-6748(00)01611-7].