A cascading network of artificial neurons for structural health monitoring

Abstract

In this paper, we demonstrate a cascading artificial neural network architecture for monitoring of structural health. The tool consists of a combination of artificial neural networks of varying architecture fitted with a data compression device. Dynamic response of the damaged structures obtained through experiments and numerical simulations are input. A preprocessor based on the Haar transformation is developed to compress the input data. In the first level, a self-organizing network based on Kohonen architecture is employed to identify the face that is damaged. The approximate location of damage is identified in the second level. In the final stage, the location and the extent of damage are predicted. The tool has been demonstrated using the example of a cantilever beam with transverse cracks. The method has been very successful in detecting damages of varying intensities at various locations.