Predicting Reduced Beam Section (RBS) Connection Performance in Steel Moment Frames

Reduced Beam Section (RBS) connections are widely adopted within seismic design codes for steel moment resistant frames. Accurate cyclic representations for RBS connections are important for the design and retrofitting of connections. Research has shown that the geometry of RBS connections affects its non-linear cyclic hysteresis. Currently, there is no method available which can accurately represent non-linear strength and stiffness degradation for any size RBS without the need for full finite element or experimental tests for calibration purposes. In this research, a more efficient design methodology for RBS connections is proposed by investigating the geometries which define the RBS connection. Then a detailed and comprehensive database of highly accurate cyclic hysteretic models of 1480 different RBS connections is presented. This database should prove useful in better understanding the seismic performance of RBS connections. Using this database, calibrated models which could accurately predict the non-linear cyclic behaviour including stiffness and strength degradations were developed. Using this calibrated data set, highly accurate and reliable neural networks were trained and developed that are capable of predicting the full cyclic hysteresis of any RBS connection; given the geometries which define the connection as an input. A mode accuracy of 98% for these networks was achieved. Following on from this work, a proof of concept for the application of this work for the design and assessment of steel moment resistant frames was investigated, by utilizing the deep learning neural networks to predict the local RBS connection hysteretic models. Finally, the potential for additive printing (3D printing) in the design of future connections has been suggested through a critical literature review. The results from this research clearly show that the three geometrical parameters which define the RBS connection influence the seismic design parameters of the connection. Design equations to predict the effects the key seismic design parameters have on RBS connections compared with a corresponding full steel connections has been proposed. Interestingly, the sections could be categorised into two sets depending on their buckling characteristics after being subjected up to a performance-based design loading criteria corresponding to collapse prevention. This database was used to calibrate improved models capable of representing RBS non-linear hysteresis. Comparisons between these calibrated parameters and the values predicted from equations available in literature gave significant differences. The differences were a result of 1) the regressional equations available in literature were based on a limited number of data points 2) a number of different types, sizes and shapes of RBS configuration were in the database and 3) the cut which defines the RBS geometry was not taken into account. Consequently, a set of deep learning neural networks enabled the non-linear cyclic hysteresis incorporating strength and stiffness degradation to be predicted for any section given the section properties and RBS geometry of the connection as an input. A proof of concept shows how the models developed in Chapter 4 would prove useful in the design and assessment of steel moment resistant frames. To conclude this research, the realistic implementation of how this work can be applied to the potential use of additive printing through a critical literature review has been presented. Finally, future work and recommendations have been suggested.