An introduction to neural networks for automated NDT data analysis

The improvement of automated inspection to the level of the trained human operator is one of the most current research fields in NDT. Recent classification methods, such as neural networks and expert systems, aim to include past measurements and classifications into training data that can encapsulate past experience and so mimic the learning process through which every human operator progresses. An introduction will be given to neural networks. Their biological background is fast becoming forgotten, as is their once perceived ability to perform miracles of pattern recognition. They must now be considered as one among several available adaptive learning methods. The work from the ESPIRIT project ANNIE which lead to this conclusion will be briefly described. The particular benefits of neural networks in NDT are their ability to encapsulate large amounts of directly collected data and to perform rapid classifications based on such data. Their present world-wide position in NDT will be reviewed