Structural seismic resilience evaluation through real-time hybrid simulation with online learning neural networks

Abstract

Seismic resilience provides a comprehensive assessment of the ability of a community to withstand and recover from earthquake disturbances. To support the design of seismic resilient structures, quantitative assessment of seismic resilience is needed and requires numerical simulations to be performed under a risk-based context. The associated large uncertainties can lead to large computational costs and limited accuracy in the numerical simulation, especially for structural systems with critical components having complex nonlinearity and rate-dependent behaviour. To cope with such uncertainties and address simulation accuracy, a framework integrating real-time hybrid simulation is proposed to ensure the assessment accuracy of the seismic resilience of structures. With real-time hybrid simulation, modelling accuracy under wide range of design scenarios can be improved. To more efficiently develop fragility curves using the results of real-time hybrid simulation, experimental substructure component metamodelling is included through an online learning approach using long-short-term memory neural networks. The proposed integration of real-time hybrid simulation and metamodelling in the fragility analysis to support resilience assessment is demonstrated through a proof-of-concept case study on the seismic retrofit of a six-storey building using inter-storey isolation.