This work presents a surrogate model-based Bayesian model updating (BMU) approach for automated damage identification of large-scale structures, which outperforms methods currently available in the literature by effectively solving the real-time damage identification challenge. The computational difficulties involved in Bayesian inference using intensive numerical models are circumvented by implementing a high-fidelity surrogate model and an adaptive Markov Chain Monte Carlo (MCMC) algorithm. The developed surrogate model combines adaptive sparse polynomial chaos expansion (PCE) and Kriging meta-modelling. The optimal order of the polynomials in the PCE is automatically identified by a model selection technique for sparse linear models, the least-angle regression (LAR) algorithm. Then, the optimal PCE is inserted into a Kriging predictor as the trend term, while the stochastic term is fitted through a global optimization algorithm. Afterwards, the surrogate model bypassing the original numerical model is used for BMU exploiting monitoring data extracted from continuous ambient vibration measurements. The computational demands of the MCMC algorithm are kept minimal by implementing an adaptive Metropolis sampling with delayed rejection (DRAM). The effectiveness of the proposed methodology is demonstrated through three case studies: an analytical benchmark; a planar truss structure; and a real case study of an instrumented historical tower, the Sciri Tower in Italy. The presented results demonstrate that the proposed BMU approach is compatible with real-time Structural Health Monitoring (SHM), providing promising evidence for the development of digital twins with superior probabilistic damage identification capabilities.

这项工作提出了一个基于代理模型的贝叶斯模型用于大型结构的自动损伤识别的更新（BMU）方法，通过有效解决实时损伤识别挑战，优于文献中目前可用的方法。通过实施高保真代理模型和自适应马尔可夫链蒙特卡罗 (MCMC) 算法，避免了使用密集数值模型进行贝叶斯推理所涉及的计算困难。开发的代理模型结合了自适应稀疏多项式混沌扩展 (PCE) 和克里金元建模。PCE 中多项式的最佳阶数通过稀疏线性模型的模型选择技术、最小角度回归 (LAR) 算法自动识别。然后，将最优 PCE 作为趋势项插入到 Kriging 预测器中，而随机项是通过全局优化算法拟合的。之后，绕过原始数值模型的代理模型用于 BMU，利用从连续环境中提取的监测数据振动测量。通过实施具有延迟拒绝 (DRAM) 的自适应 Metropolis 采样，将 MCMC 算法的计算需求保持在最低限度。通过三个案例研究证明了所提议方法的有效性：分析基准；平面桁架结构；以及意大利的Sciri Tower 历史塔楼的真实案例研究。所呈现的结果表明，所提出的 BMU 方法与实时结构健康监测 (SHM) 兼容，为开发具有卓越概率损伤识别能力的数字双胞胎提供了有希望的证据。