Damage detection of plates based on vibration‐based methods and supervised learning approach

Abstract A supervised learning method that uses neural networks is developed and trained, and its potential to successfully localize damage on plates is presented. The training data, which represents an essential part of the overall process, is generated by the transverse displacements of the plate at multiple locations and times. Damages are modeled by reducing the thickness of the plate in the damaged area. The transverse displacements are obtained by numerically solving the equation of motion for various locations of damage and excitation frequencies. It is shown that the neural network accurately predicts damages that were included in the training data, but it also predicts very well new damages, even smaller ones, and with added noise.