IoT Thrust Seminar | AI4UQ: Artificial Intelligence for Uncertainty Quantification in Structural Dynamics
Supporting the below United Nations Sustainable Development Goals:支持以下聯合國可持續發展目標：支持以下联合国可持续发展目标：
Structural dynamics faces complex uncertainties, making uncertainty quantification and robust prediction crucial for safety and reliability. This talk will introduce uncertainty-aware and error-resilient artificial intelligence solutions for forward and inverse learning of complex dynamical systems. For forward problems, a multi-output, multi-physics Bayesian deep learning method with parallel subnetworks is developed to solve dimension-reduced probability density evolution equations in noisy data regimes, enabling efficient spatiotemporal uncertainty propagation for dynamic responses in high-dimensional nonlinear stochastic dynamical systems. For inverse problems, we propose a brand new and ultrafast equivalently explainable neural-networked variational inference framework that overcomes mean-field limitations, computational challenges, the curse of dimensionality, and scalability issues of conventional Bayesian model updating when dealing with an implicit solver. Additionally, a variational hierarchical Bayesian modeling framework, replacing the Gaussian white-noise prediction error model with a Gaussian process, is developed to capture non-stationary prediction discrepancies and account for inherent structural variability. These innovations enhance computational efficiency, error resilience, and uncertainty-awareness, offering significant potential for structural stochastic dynamics and structural health monitoring.
Prof. Wang-Ji Yan earned his Ph.D. from the Hong Kong University of Science and Technology in 2013 and has held positions as a full Professor at Hefei University of Technology (2014) and as an Marie Skłodowska-Curie Fellow at the University of Nottingham (2018). He is currently an Associate Professor at State Key Laboratory of Internet of Things for Smart City and Department of Civil and Environmental Engineering, University of Macau. Specializing in structural vibration and structural health monitoring, he has published over 100 top-tier journal papers, including 35 in Mechanical Systems and Signal Processing as a leading journal in his area. His research, supported by nearly 20 grants (totaling over 10 million MOP) from sources such as NSFC, Ministry of Science and Technology, EU Horizon 2020, and Macau FDCT, has advanced transmissibility theory and Bayesian learning for dynamics, earning over 2,700 citations and inspiring extensive follow-up studies. Prof. Yan serves as an editor for five international journals and on technical committees of more than ten academic bodies. He is the first Chinese scholar to receive the prestigious European Association for Structural Dynamics Junior Research Award and has won over ten other honors, including Natural Science Awards from the Chinese Ministry of Education, the Macau SAR government, and the Chinese Association of Vibration. He was invited to deliver over 20 keynote/invited speeches, and has mentored over 30 postgraduate students and postdoctoral researchers.