Assessment of damage indicators for structural health monitoring using a probabilistic framework

Civil and mechanical structures are exposed to natural ageing and accidental overloading. To increase safety and reduce costs simultaneously, current classical maintenance strategies can be supported by advanced fully automatic vibration-based structural health monitoring (SHM) systems. Such vibration-based SHM systems consist of a data acquisition system with sensors applied on a structure. From recorded time histories, damage indicators are generated and can be monitored in control charts. Sensitivity to damage and insensitivity to confounding effects are essential properties for a well-designed indicator. To design such damage indicators, experimental tests are inevitable. As such tests can be expensive with respect to time and labor costs, researchers are interested in replacing these tests by numerical simulations. Recently, a semi-analytical probabilistic framework has been proposed by the authors to determine the statistical properties of damage indicators based on the known statistical properties of a random excitation in the time domain and a finite element model. Therefore, a time consuming numerical time integration scheme, as typically applied for each sample of a sample-based stochastic structural analysis, can be avoided. In the presented paper this probabilistic framework is applied to compare efficiently two different damage indicators based on modal filters using a numerical cracked concrete beam model. For this numerical example, the statistical properties of the damage indicators will be investigated with progressing damage. From these statistical properties, it can be determined under which conditions the distribution of the damage indicators can be assumed to be normal and, therefore, if the application of certain control charts is justified.