Combined Model Order Reduction Techniques and Artificial Neural Network for Data Assimilation and Damage Detection in Structures

The last frontier of Structural Health Monitoring is real-time damage localization, which requires reliable and efficient statistical and computational tools. We treat the damage localization problem as a classification problem, considering a finite number of possible damage scenarios in a structure under varying loading conditions. A dataset of possible structural responses, both in the undamaged and damaged conditions, is constructed by using a physics-based numerical model of the structure. This dataset is employed to train a Fully Convolutional Network accomplishing the classification task. Model Order Reduction techniques are exploited to speed up the dataset construction and thus enhance the overall computational efficiency. Once trained, the Fully Convolutional Network is able to classify the actual damage state of the structure on the basis of incoming vibration time series recorded by the monitoring sensor systems. Finally, the proposed procedure is assessed through a numerical benchmark.