ABSTRACT
Lead-core rubber bearing is a widely used device for seismic isolation. It remains a preferred option for important structures such as bridges, nuclear power plants, hospitals, etc., mainly due to its considerable energy dissipation capability. Accurate predictive models are a prerequisite for a safe and efficient design and thus prolong the life cycle of buildings and structures. A detailed finite element model generally provides insight into the complex behavior of the lead-core rubber bearing. However, being demanding computational resources, its implementation in the base-isolated structure analysis might not be appropriate. The contribution focuses on a more efficient model from a computational point of view – a neural network. Although the neural network doesn’t use the physical model, the latter is employed to produce the target data. Some technological aspects of the multilayer perceptron are discussed, such as architecture, choice of the activation function, and training strategies.
