Ensemble of the Distance Correlation-Based and Entropy-Based Sensor Selection for Damage Detection

In this paper, a novel ensemble Principal Component Analysis (PCA) algorithm is proposed to detect the presence of damage by exploiting the structure's historical data. In particular, there are 2 main contributions highlighted in this paper: First, a sensor selection algorithm is derived from the distance correlation coefficient from the correlation analysis, to reduce the number of sensors without affecting the model accuracy and fault detection sensitivity. Next, a novel technique based on the combination of the distance correlation-based and the previously introduced entropy-based PCA, is derived, to generate the ensemble PCA algorithm, which can be used to detect structural damages and improves the robustness of the previous methods. The presented algorithms are validated on three different damage cases, providing evidence that the proposed ensemble PCA algorithm outperforms the previous approaches, in the sense that it improves the fault detection sensitivity and model prediction accuracy, while also offering information on the most sensitive subset of sensors in detecting faults.