Data-driven damage identification technique for steel truss railroad bridges utilizing principal component analysis of strain response

Abstract

Railway bridges are a critical part of the railway infrastructure system and the majority of these bridges are approaching their expected design lifespan. These bridges need to be maintained effectively. This article proposes a damage detection framework for truss railway bridges based on operational strain responses. The method relies on principal component analysis (PCA) of strain responses of the truss bridge. Strain time-history responses under baseline and damaged bridge conditions are used to compute the principal components which are then ranked based on their corresponding eigenvectors. The results are demonstrated in terms of damage indicators which is obtained by comparing the geometric distance of coordinates of the principal component space between the baseline and damaged bridge condition. The method is numerically verified through a finite element model of a truss railroad bridge with added artificial noise. It is shown that the proposed method could identify, locate and relatively assess the severity of the damage induced by stiffness change (at least 20% to as low as 10%, depending on operational variability and measurement noise level) in instrumented truss elements. This method can be useful in assisting the existing bridge maintenance techniques and formulating an effective structural health monitoring framework.