Paper Link: https://doi.org/10.1016/j.compstruc.2021.106507

Number of sensors deployed on a skyscraper for structural health monitoring is usually very limited due to cost. We propose an optimization technique for sensor location selection based on attention scores in a deep neural network that learns the relations between ground motion and displacement responses of the building.

Architecture of the proposed multi-scale attention network.

Convergence of attention scores for selected sensor locations. Neurons connected to indices of unselected sensors are then removed, and the network is thereafter retrained. This process is known as pruning in deep learning.