Abstract
In the recent years, machine learning algorithms have been widely employed for structural health monitoring applications. As an example, Artificial Neural Networks (ANN) could be useful in giving a precise and complete mapping of damage distribution in a structure, including low-intensity or localized defects, which could be difficult to detect via traditional testing techniques. In this domain, Convolutional Neural Network (CNN) are employed in this work along with one-dimensional refined models based on the Carrera Unified formulation (CUF) for surface strain/displacement based damage detection in composite laminates. A layer-wise kinematic is adopted, while an isotropic damage formulation is implemented. In detail, CUF-based finite element models have been exploited in combination with Monte Carlo simulations for the creation of a dataset of damage scenarios used for the training of the CNN. Therefore, the latter is fed with images of the strain or displacement field in a region of particular interest for each sample, which are subjected to the same boundary conditions. The trained CNN, given the strain/displacement mapping of an unknown structure, is therefore able to detect and classify all the damages within the structure, solving the so-called inverse problem.