Deep Neural Network and YUKI Algorithm for Inner Damage Characterization Based on Elastic Boundary Displacement

Code

• github.com/Brahim-Benaissa/YUKI_ANN
  none★ 2

Abstract

The efficiency of deep neural networks has been proven in several research fields. In this study, we suggest using this method of inverse crack identification based on the structural response of boundary displacement. This structural response is particularly challenging for surrogate modelling due to the overall similarity in the effect of different cracks. From the inverse problem perspective, this corresponds to a problem of many local minima. To solve this problem we use the newly suggested search technique of the dynamic search space reduction by the YUKI algorithm, build to solve this type of problem. We compare the performance of the suggested approach of the RBF modelling technique in terms of direct problem prediction and inverse problem identification accuracy. Deep Neural Networks are found to have better performance in both problems, although the computation time is significantly higher than RBF.

Reproductions