Damage detection and severity estimation in plates using neural networks and supervised learning

Plates are thin structures extensively utilized for various engineering applications. They frequently experience significant deformations and extreme weather conditions that result in structural damage. Early detection of damages, their localization, and estimation of severity are essential for the operation and maintenance of the constructions. The study presents a supervised machine learning approach for damage detection and severity estimation in plates, taking inspiration from vibration-based methods. The input data of the neural network represents the deformed shape of the plate at some instant time from its forced vibration response. The deformed shape is expressed as a vector from the transverse displacements recorded at various locations on the plate’s surface. The neural network training data is generated by simulating the time response of damaged plates. The research shows that the neural network is capable of accurately determining the location and magnitude of the damage. The model exhibits high accuracy even with the introduction of random noise in the time response. The reduction in the number of locations for measuring and storing transverse displacements for training the neural network is investigated and presented. Additionally, several strategies for sensor distribution, encompassing asymmetric locations on the plate's surface, are examined.