Structural health monitoring of road bridges - Part 2: Anomaly detection with principal component analysis

Structural health monitoring of bridges has the long-term potential to be established as an essential additional tool for condition assessment. Recent research in this field focuses increasingly on signal features of various sensor types as well as machine learning methods. Following those approaches, part 2 of this article discusses how structural damage can be identified by means of anomaly detection with machine learning models. Part 1 introduces a signal feature that is based on influence lines: the R-signature. Simulations show that the R-signature is significantly more sensitive to structural damage than the considered natural frequencies. Part 2 described an anomaly detection procedure that identifies structural damage as changes in the correlation structure of the R-signature. The underlying data model uses the principal component analysis. The presented approach can be verified with strain measurements of an existing road bridge.