The strain data acquired from structural health monitoring (SHM) systems of large-span bridges are often contaminated by a mixture of temperature-induced and vehicle-induced strain components, thereby complicating the assessment of bridge health. Existing approaches for isolating temperature-induced strains predominantly rely on statistical temperature–strain models, which can be significantly influenced by arbitrarily chosen parameters, thereby undermining the accuracy of the results. Additionally, signal processing techniques, including empirical mode decomposition (EMD) and others, frequently yield unstable outcomes when confronted with nonlinear strain signals. In response to these challenges, this study proposes a novel temperature-induced strain separation technique based on improved blind source separation (BSS), termed the Temperature-Separate Second-Order Blind Identification (TS-SOBI) method. Numerical verification using a finite element (FE) bridge model that considers both temperature loads and vehicle loads confirms the effectiveness of TS-SOBI in accurately separating temperature-induced strain components. Furthermore, real strain data from the SHM system of a long-span bridge are utilized to validate the application of TS-SOBI in practical engineering scenarios. By evaluating the remaining strain components after applying the TS-SOBI method, a clearer understanding of changes in the bridge’s loading conditions is achieved. The investigation of TS-SOBI introduces a novel perspective for mitigating temperature effects in SHM applications for bridges.

Structural health monitoring (SHM) of composite wind turbine blades (WTBs) is crucial for improving power efficiency, reducing maintenance costs, and ensuring long-term structural reliability. Traditional frequency-based damage detection, often derived from simplified isotropic beam principles, can be challenged by the anisotropy, heterogeneity, and geometric complexity of composite WTBs. Moreover, as global indicators, natural frequencies are sensitive to environmental variations but are also limited in localizing damage. To overcome these challenges, this research introduces a combined approach of relative natural frequency change (RNFC) and Bayesian probability, referred to as the B-RNFC method. The framework includes four stages: (i) analyzing the correlation between natural frequencies and damage conditions (location and severity) in composite cantilever beams and WTBs; (ii) developing normalized RNFC curves from various damage sizes to establish a spatial damage reference dataset, which is then used for the next steps; (iii) integrating the resulting frequency-related data with Bayesian probability to identify damage locations and map them onto the structures; and (iv) evaluating the performance of the B-RNFC in multiple-damage localization. Simulation results demonstrate the effective damage localization range of the B-RNFC method. For a simple cantilever beam, this range is 20–80% of the distance from the fixed end. When applied to the composite WTB, this effective range corresponds to 40–80% of the blade length from the root. In addition, the proposed method can localize the dual damages when the damages are symmetrically located or when one damage is at the mid-span.