Neural network approach to holographic nondestructive testing

A neural network approach for the automatic detection of defects by evaluation of holographic interference patterns of the loaded technical components is described. Translation- as well as rotation-invariant features are defined based on the maximal local slope of the intensity and a partition of the interference pattern into nonoverlapping areas. The training sample set is generated by computer simulation of interferograms directed by a few typical experimentally measured samples. Practical results show the feasibility of the method. A strategy for application of neural networks to any holographic nondestructive testing task is outlined.