On model- and data-based approaches to structural health monitoring

• Main Author: Barthorpe, Robert James
• Other Authors: Worden, Keith ; Manson, Graeme
• Published: University of Sheffield 2010
• Subjects: 624.17
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.531188

Structural Heath Monitoring (SHM) is the term applied to the process of periodically monitoring the state of a structural system with the aim of diagnosing damage in the structure. Over the course of the past several decades there has been ongoing interest in approaches to the problem of SHM. This attention has been sustained by the belief that SHM will allow substantial economic and life-safety benefits to be realised across a wide range of applications. Several numerical and laboratory implementations have been successfully demonstrated. However, despite this research effort, real-world applications of SHM as originally envisaged are somewhat rare. Numerous technical barriers to the broader application of SHM methods have been identified, namely: severe restrictions on the availability of damaged-state data in real-world scenarios; difficulties associated with the numerical modelling of physical systems; and limited understanding of the physical effect of system inputs (including environmental and operational loads). This thesis focuses on the roles of law-based and data-based modelling in current applications of. First, established approaches to model-based SHM are introduced, with the aid of an exemplar ‘wingbox' structure. The study highlights the degree of difficulty associated with applying model-updating-based methods and with producing numerical models capable of accurately predicting changes in structural response due to damage. These difficulties motivate the investigation of non-deterministic, predictive modelling of structural responses taking into account both experimental and modelling uncertainties. Secondly, a data-based approach to multiple-site damage location is introduced, which may allow the quantity of experimental data required for classifier training to be drastically reduced. A conclusion of the above research is the identification of hybrid approaches, in which a forward-mode law-based model informs a data-based damage identification scheme, as an area for future work