Artificial Intelligence and Machine Learning in Friction Pendulum Seismic Isolation Systems: A Focused Review of Methods, Applications, and Research Gaps

Friction pendulum systems (FPSs) are nonlinear seismic isolation devices whose analysis and design can require costly repeated dynamic simulations. This paper presents a focused review of artificial intelligence (AI), particularly machine learning (ML), applications to FPSs and related frictional-isolation systems. Sixteen core primary studies published from ۲۰۲۲ to ۲۰۲۶ are compared by engineering task, data source, model architecture, and validation strategy, with one additional functionality-oriented study treated as scope-adjacent. The literature covers response prediction, inverse bearing design, risk and operability assessment, hysteresis modeling, and physics-informed, hybrid, probabilistic, and surrogate-learning approaches. Tree-based ensembles remain prominent, while recent studies introduce recurrent neural networks (RNNs), physics-informed neural networks, Gaussian-process models, Kriging and polynomial-chaos surrogates, and support-vector regression. Most studies rely on numerical databases, whereas direct experimental learning is rare. The main gaps are experimental validation, temporal response learning, calibrated uncertainty, cross-domain generalization, and real-time deployment. The review concludes that the next step is to combine physically informed learning with stronger validation and connected intelligent engineering workflows.

کلیدواژه ها:

artificial intelligence ، machine learning ، friction pendulum systems ، seismic isolation ، surrogate modeling ، physics-informed neural networks ، seismic response prediction