Accurate identification of structural physical parameters is crucial for evaluating the safety and performance of existing multi-story buildings. Conventional identification methods rely heavily on idealized prior models with assumed stiffness or mass matrices; however, given that cumulative damage and construction errors render the acquisition of accurate prior information for long-term service structures nearly impossible, such practices are often impractical and lead to significant error propagation from inaccurate structural assumptions. Moreover, traditional methods frequently exhibit reduced accuracy and stability under the influence of environmental noise and modeling uncertainty. To address these dual challenges, this study develops a comprehensive framework designated as SRCA-SWPINN, which integrates a newly developed stiffness-response collaborative acquisition (SRCA) strategy with a stage-wise physics-informed neural network (SW-PINN). The core innovation of the SRCA lies in its ability to directly reconstruct the actual stiffness matrix through quasi-static loading, thereby reducing reliance on idealized assumptions and ensuring the physical fidelity of identified parameters from the source. Furthermore, the SW-PINN employs a staged optimization strategy to effectively mitigate the non-convexity and local minima issues inherent in conventional PINNs, achieving stable and high-precision identification of structural physical parameters. Numerical validation on a three-degree-of-freedom shear model demonstrates that the proposed framework achieves high identification accuracy, with mass errors remaining below 1.5 % and the relative error of the damping matrix limited to 9.23 %. In contrast, a standard PINN implementation yields significantly larger errors of up to 6.4 % for mass and 14.76 % for damping. Furthermore, a systematic parametric study comprising 35 test cases is conducted to evaluate robustness under mass variation, stiffness degradation, measurement noise, and modeling uncertainty. The results show that mass identification errors are generally maintained within 3 %-4 % for most scenarios and remain below 6 % even under severe modeling uncertainty, while damping errors mostly stay within 10 %-12 % under diverse adverse conditions. Unified statistical analysis further indicates that the Percentage Within Tolerance (PWT) reaches 99 % for mass parameters under a 5 % threshold and 80 % for damping under a 12 % threshold, confirming the effectiveness and reliability of the proposed SRCA-SWPINN framework.
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号