Application of Machine Learning in Structural Health Monitoring of Bridges

Abstract

The necessity of structural health monitoring (SHM) for the safety, reliability and service life of bridge structures is well recognized. Traditional inspection techniques that are commonly manual and tedious and susceptible to human errors, now can be complemented or replaced by more advanced data driven methods. Machine Learning (ML) has powerful tools for processing massive sensor data, anomaly detection, predicting structural decay and promoting proactive maintenance decision making. In this paper, such ML techniques are reviewed in the context of applying them to bridge SHM including supervised and unsupervised learning intact damage detection/condition assessment algorithms along with those that estimate time till failure (or remaining service life).The review was conducted using secondary data derived from published literature, case studies and experimental reports and assessed the potential of each ML algorithm such as ANN, SVM, decision trees and CNN. The analysis shows accuracy in detection, early warning and action precision increase as benefit while challenges on data quality, sensor location, the interpretability of models and computation time are also considered. Results show that ML-based SHM could help improve the safety and reliability of bridges, reduce maintenance cost, and facilitate the transformation to smart infrastructure.