Monitoring bridge performance is crucial to ensure safety and allocate resources in a cost-effective manner. This paper aims to reduce the gap between researchers and practitioners by showing how predictive analytics can be employed in the process of distilling operational information out of bridge monitoring data. Furthermore, it has the goal to aid infrastructure owners and managers in evaluating bridge performance over time and making data-driven decisions to prolong the life of the structure. To achieve this goal, the paper presents a comparative study of three predictive analysis models to estimate bridge response to heavy trucks: multilinear regression, artificial neural network, and regression tree. Following this comparison, an alternative strategy, based on the analysis of influential observations, is proposed. This approach brings together predictive power with other important capabilities such as explanatory capabilities and interpretability. The test bed structure is a short-span highway bridge which was monitored for 3 years using weigh-in-motion (traffic data) and structural health monitoring (bridge data) systems.
Read full abstract