APPLICATION OF NEURAL-NETWORK COMPUTING TO THERMAL NONDESTRUCTIVE EVALUATION

A methodological study on the use of neutral networks for defect chara cterisation by means of a thermal method is presented. Neural networks are used here as defect classifiers, based oil the infrared emission of the target object after heating. In this kind of application, there is a high degree of uncertainty in defect class boundaries due to sev eral factors, such as the noise in the measurement, the uneven heating of the target object and the anisotropies in its thermal conductivity . For this reason, the classical 'l of N' coding scheme during trainin g did not provide satisfactory results. Much better results have inste ad been obtained ruing a smoother activation function for the output u nits during training. The non-destructive evaluation of material using neural networks proved extremely satisfactory, especially when compar ed to the classical procedures of thermographic analysis.