Entropy-Based Sensor Selection Algorithms for Damage Detection in SHM Systems

It is often the case that small faults in a structure lead to irreparable damages that deliver a huge financial loss or even pose safety risks. Thus, an early fault detection is necessary, such that these unfortunate events can be avoided. In this thesis, the problem of structural damage detection is considered. In particular there are 3 main contributions: First, a novel sensors selection algorithm based on the concepts of entropy and information gain from information theory is developed, to reduce the number of sensors without affecting, or even improving, model accuracy; Second, a novel technique based on Kalman filtering and on a combination of Regression Trees theory from Machine Learning and Auto Regressive (AR) system identification from control theory is derived, to build models that can be used to detect structural damages. Finally, a new fault detection algorithm based on Poly-Exponential (PE) models and nonlinear Kalman filtering on the residual is introduced, which is able to enhance the sensitivity of the proposed fault detection algorithm and improve the data prediction quality for some accelerometers in a notably margin. The presented techniques are validated on three different experimental datasets, providing evidence that the proposed algorithms outperform some previous approaches, improving the prediction accuracy and the damage detection sensitivity while reducing the number of sensors.