A Supervised Deep Learning Method to Classify Structural Damage of a Bridge Deck Mock-Up

Structural damage detection and prediction under various type of demands has been a significant area of research over the past few decades. Applications of machine (ML) and deep learning (DL) to this topic have provided essential insight into damage detection and prediction in many engineering disciplines, including structural health monitoring (SHM). This study mainly focuses on implementation of DL-based algorithms to help identify and classify imposed structural damage to a full-scale bridge deck mock-up whose structural response was monitored under varying loading conditions and damage levels. Strain time-history data, which represents a healthy, undamaged, state and three incremental damage states, was collected from the field experiments. Supervised ML was used to construct a dedicated two-dimensional (2D) convolutional neural network (CNN), which can extract and classify features, using sensor readings as input “images.” The proposed 2D-CNN model utilized four fully connected, dense layers with pooling operations integrated after each layer. Rectified linear unit (ReLU) and SoftMax activation functions were used in the hidden and last output layers, respectively. The experimental data was split into training, testing, and validation sets. Damage labels and corresponding images were initially known for the training dataset. The constructed CNN model was trained, and damage location and a high prediction accuracy were obtained for training, validation, and testing datasets.