Proceedings of SPIE - the International Society for Optical Engineering
A novel automated inspection technique to recognize, locate, and quantify damage is developed. This technique is based on two already existing technologies: video moire metrology and artificial neural networks. Contour maps generated by video moire techniques provide an accurate description of surface structure that can then be automated by means of neutral networks. Artificial neural networks offer an attractive solution to the automated interpretation problem because they can generalize from the learned samples and provide an intelligent response for similar patterns having missing or noisy data. Two dimensional video moire images of pipes with dents of different depths, at several rotations, were used to train a multilayer feedforward neural network by the backpropagation algorithm. The backpropagation network is trained to recognize and classify the video moire images according to the dent's depth. Once trained, the network outputs give an indication of the probability that a dent has been found, a depth estimate, and the axial location of the center of the dent. This inspection technique has been demonstrated to be a powerful tool for the automatic location and quantification of structural damage, as illustrated using dented pipes.
Grossman, B. G., Gonzalez, F. S., Blatt, J. H., & Cahall, S. C. (1993). Detection and location of pipe damage by artificial-neural-net-processed moire error maps. Paper presented at the Proceedings of SPIE - the International Society for Optical Engineering, , 1821 415-427.