Data-driven assessment and modelling of extended endplate connections

Extended endplate connections (EEPCs) are widely used in construction practice to resist gravity, wind, earthquakes, progressive collapse, as well as other hazards. EEPCs can be designed as either fully- or semi-rigid achieving wide ranges of rotational stiffness, moment resistance, and ductility. As demonstrated in past research, accurately predicting EEPCs’ behaviour can be challenging, particularly for semi-rigid ones where plastic deformations may occur in any of the connection components, individually or simultaneously. This has been the case because of simplified assumptions and limited datasets. In this thesis, data-driven approaches are used to develop robust empirical methodologies for assessing and modelling semi-rigid EEPCs. This is achieved by mobilizing the extensive body of experimental research available in the literature. A major effort is made towards collating, digitising, and curating all available data. The data are used to assess all types of existing predictive models and confirming their limitations. The data is further supplemented by parametric finite element simulations, with emphasis on ductility quantification, i.e., bolt tensile rupture. Traditional regression and machine learning algorithms are then used to develop robust predictive models to capture semi-rigid EEPCs’ moment-rotation response and governing deformation mode. The new models’ novelty lies in their generality (larger applicability range), consideration of uncertainty, characterization of ductility, post-failure behaviour, and hysteretic response. The proposed models are detrimental in probabilistic and performance-based approaches to design, assess, and retrofit steel frame structures. They can effectively help the industry deliver economical designs with less material consumption, which is critical to achieve the goal of the net-zero strategy by 2050.