Optimizing input data for training an artificial neural network used for evaluating defect depth in infrared thermographic nondestructive testing

Ten different sets of input data have been used for training and verification of the neural network intended for determining defect depth in infrared thermographic nondestructive testing. The input data sets included raw temperature data, polynomial fitting, principle component analysis, Fourier transform and others. A minimum error (up 0.02 mm for defects in CFRP at depths from 0.5 to 2.5 nun) has been achieved by using polynomial fitting in logarithmic coordinates with further computation of the first temperature derivatives (the TSR technique), and close results have been obtained by processing raw data with the PCA technique. Both techniques require no use of reference points.