5. On Predicting Uncertainties in the Dynamic Response of a Welded Structure

Present-day engineering projects are highly dependent on numerical models; thanks to the improvements in computing capabilities that have contributed significantly in this area. Although models these days are far better representations of engineering structures than before, every model is limited by its mathematical representation and the knowledge about the underlying physics. Validating numerical models involves obtaining test or performance data, which may not be practical in the case of many engineering structures. In such cases where data for the full structural model are not available, the subsystems or components can be tested and the associated models calibrated and validated separately. Inferring response at system level from these subsystem validation results is not straightforward and needs a proper uncertainty propagation strategy. Furthermore, the response of a structure depends not only on the components that it is made of, but also equally on the joints. Joints determine how the components interact within a structure, making validating the models for joints as important as validating the subcomponents. A joint does not physically exist on its own but co-exists with the components it connects, which makes it difficult to define a joint model. Isolated models, where the same type of joints is employed to connect ‘simple’ components with well-established numerical models can serve such a joint model for the purpose of validation. This paper describes a probabilistic approach to dealing with joints via such isolated models, where the uncertainties related to a welded joint are quantified and propagated into a target model of a welded structure to predict the dynamic response.