Data interpretation and forecasting of SHM heteroscedastic measurements under typhoon conditions enabled by an enhanced Hierarchical sparse Bayesian Learning model with high robustness

Rapid advances in Structural Health Monitoring (SHM) have led to a new era of Big Data in the past decade. However, collected data may include a huge amount of heteroscedastic data, particularly under extreme events (e.g., typhoons), which makes robust decisions remain a challenging problem. To this end, this article proposed an enhanced Hierarchical Sparse Bayesian Learning (eHSBL) model for SHM field data interpretation and forecasting, uncertainty analysis, as well as correlation analysis. The proposed eHSBL model incorporates the Gaussian kernel function and the Hierarchical Bayesian model for an operational large-scale suspension bridge, i.e., Tsing Ma Bridge (TMB) in Hong Kong, under extreme events exhibiting heterogeneities, strong volatilities, and high uncertainties. The Gaussian kernel function in the eHSBL model is capable of mapping SHM heteroscedastic data into high-dimensional space, and the Hierarchical Bayesian model can be iteratively updated by introducing hyperparameters to achieve the sparse expression of the model, so the model has strong generalization ability and high robustness. Through the incorporation of the extrapolation forecasting algorithm, the eHSBL can carry out the forecasting analysis for the future SHM data. Results show that the eHSBL regression and forecasting results under extreme events can well reflect the strain responses of the structure. The uncertainty and correlation analysis reveal that the eHSBL model error variation is influenced by temperature, traffic loads, and wind speeds, and the corresponding error variation curves and change rules are also provided and summarized, which will shed light on the performance evaluation and early warning for in-service long-span bridges.