A data fusion based approach for damage detection in linear systems,

One of the main goal of dynamic identification is to derive information about the health of systems by monitoring changes affecting their dynamic properties due to the presence of damage. Indeed, considering damage indicators just based on modal parameters of systems before and after the damage, three possible levels of information can be derived: presence of damage (level 1), position of damage (level 2), severity of damage (level 3). In this context modal strain energy (MSE) is often used as basis for defining a damage detection index for systems based on the change of its dynamic properties and elemental stiffness of systems. The studies available in literature, among which [1-4], show the ability of this indicator to cover all the levels of information concerning the status of damaging of systems. Nevertheless, the same studies have also underlined some drawbacks of this approach, which generally arise when multiple damages occur or significant levels of noise/error affect the identified dynamic properties of systems. In these cases the major difficulties depend on identifying only a limited number of modes and not always those more sensible to the damage scenarios. It is, thus, important to improve the performance of the traditional MSE based damage indicator in the case of multi damage locations, noise-polluted data and reduced number of modes. Recently, in order to improve the ability of classical indicators to detect damage of systems, some literature studies [5-9] propose to extent the classical data information fusion techniques [10] to structural damage identification with the intent of combining information from different sources and improving the final result. In this context, this paper presents an approach for damage identification of systems, which combines the use of damage indicators derived through the MSE with a multi stage data-fusion procedure. Specifically, considering different sets of the identified modes of vibration as information sources, modal strain energy change ratios are evaluated and converted in local decisions. The single decisions provided by each source constitute the data sent to the fusion center, where they are combined on the basis of a fusion approach that provides the global decision. More specifically, the methodology followed in the present paper is based on the classical Dempster-Shafer (DS) theory of evidence. The approach is applied to some numerical examples with different damage scenarios, set of identified modes of vibrations and, also, noise levels. The obtained results clearly show that the proposed approach can improve the performances of the classical MSE based damage indicator in the case of single damage scenarios and, mainly, in the case of multiple damage scenarios where generally the classical indicators fail. The results show also a significant robustness of the approach in presence of noises. In all these cases the proposed approach provides efficient information in terms of location and also extent of damage.