Hierarchical ensemble-based data fusion for structural health monitoring

In structural health monitoring, damage detection results always have uncertainty because of three factors: measurement noise, modeling error and environment changes. Data fusion can lead to the improved accuracy of a classification decision as compared to a decision based on any individual data source alone. Ensemble approaches constitute a relatively new breed of algorithms used for data fusion. In this paper, we introduced a hierarchical ensemble scheme to the data fusion field. The hierarchical ensemble scheme was based on the Dempster–Shafer (DS) theory and the Rotation Forest (RF) method, it was called a hierarchical ensemble because the RF method itself was an ensemble method. The DS theory was used to combine the output of RF based on different data sources. The validation accuracy of the RF model was considered in the improvement of the performance of the hierarchical ensemble. Health monitoring of a small-scale two-story frame structure with different damages subject to shaking table tests was used as an example to validate the efficiency of the proposed scheme. The experimental results indicated that the proposed scheme will improve the identification accuracy and increase the reliability of identification.