Regularized Tensor Completion for Structural Health Monitoring Data Imputation

Missing sensor data is a ubiquitous issue in structural health monitoring (SHM), often resulting from sensor failures and restricting system effectiveness. SHM measurements collected by spatially distributed sensors within a time interval often exhibit intricate temporal patterns and locally-dependent spatial patterns, bringing significant challenges for conventional imputation methods, particularly in continuous missing data scenarios. To address these challenges, our research proposed a novel missing data imputation method named Spatial Regularized Tensor Completion (SRTC), which exploits the low-rank structure to handle temporal characteristics, and a spatial sparsity to capture spatial characteristics. An optimization algorithm is developed to ensure efficient estimation. The performance of the proposed method is evaluated with a real-world case study, demonstrating superior performance in missing data imputation comparing benchmark methods.