Neural Networks for Inverse Problems in Damage Identification and Optical Imaging

Abstract

Artie cial neural networks (ANNs) are employed in solving inverse problems in damage detection in structures, as well as detection, using optical imaging, of an anomaly in a light-diffusive media, such as a human tissue. Both of these problems, namely, identifying the damage parameters in a damaged structure and identifying the representative properties in a tissue, require solving highly complex inverse problems. The neural networks (NNs) for both problems are similar, and a method found suitable for solving one type of problem can be applied for solving the other type of problem. In the damage identie cation problem, the natural frequencies of a damaged beam model obtained from analytical and numerical methods were used to identify damage parameters by employing feedforward backpropagation, and also radial basis NNs. In the optical imaging problem, the tissue under investigation was illuminated by a number of near-infrared light sources placed around the circumference of the tissue. Both the location and the size of the anomaly were identie ed by studying the ine uence of the anomaly on the light intensity received at the boundary of the tissue. The near-infrared light measurements are assumed to be available at a number of light detector positions, also along the circumference of the tissue. NNs were used to determine the location and the size of the anomaly in a tissue. The direct problem for the case of optical imaging was solved using the e nite element method to generate the training and testing sets for NNs.