Artificial Neural Network for On-Line Eddy Current Testing

Eddy current test (ECT) is affected by a large number of influencing parameters such as lift-off, variations in geometry, electrical conductivity, magnetic permeability, surface condition etc. [1]. To carry out meaningful ECT and evaluation, it is essential to eliminate or reduce the influence of unwanted parameters. When the number of unwanted parameter is one, its affect can be eliminated using single frequency eddy currents, for example, by rotating the phase of the signal along one of the impedance axes, abscissa in general and taking measurement along the other axis, i.e. the ordinate. However, in actual practice, the influencing parameters are more than one and defect detection and characterisation in their presence becomes rather difficult using single frequency. For example, the popular application of ECT of heat exchanger tubes, detection of defects under support plates and in the presence of probe wobble, is rather difficult using single frequency method