Artificial Intelligence and Digital Twin for Railway Bridge Structural Health Monitoring

Abstract

The purpose of this paper is to present some preliminary results obtained from the application of a dedicated Artificial Intelligence-based monitoring software to an OpenSees numerical model of a railway bridge to assess its health conditions in near real-time. The proposed approach is based on the construction of an unsupervised Machine Learning algorithm in a full data-driven scenario, with the aim of establishing a reliable method for anomaly detection even in the absence of a numerical model. A reference pattern is obtained by collecting data at high sampling frequency on several fixed nodes as trains of various masses cross the undamaged bridge at different velocities; after pre-processing, the data are fed into various types of autoencoders which are trained to produce outputs as close as possible to the inputs. It is then shown that the algorithm actually flags the data produced when damage scenarios are activated in the OpenSees model as coming from a damaged structure. Our solution can be also adopted as a binary classifier once a threshold for reconstruction errors has been fixed.