Full-Scale Experimental Investigation of Universal Filter for Joint Input and State Estimation

Structural health monitoring depends significantly on the availability of input signals to assess system behavior. However, in most real-world applications, dynamic input loads are unknown or difficult to measure due to practical and cost constraints. Accurate input estimation is therefore essential for reliable state estimation, system identification, damage detection, and developing digital twins of physical structures. While various methods have been proposed for joint input-state estimation, many face limitations in handling systems with direct feedthrough, without feedthrough, or with rank-deficient feedthrough matrices. This study presents the first experimental validation of the universal filter (UF), a recursive Bayesian estimator proposed by the coauthors. The method is validated using a six-story laboratory shear frame and a full-scale seven-story building subjected to shake table testing. Unlike methods such as the augmented Kalman filter (AKF), the UF does not depend on fictitious models or require hyperparameter tuning for inputs. Moreover, it eliminates the need to assume an initial guess for the input and its error covariance. The robustness of the UF is first tested through a numerical case study under varying parameter conditions. Experimentally, it is evaluated using sparse dynamic measurements obtained from base-excited structures. The performance of the UF is compared to state-of-the-art algorithms, including the AKF and Gillijns and de Moor filter, using a system with a direct feedthrough matrix (GDF-WDF). This work establishes a complete framework from data acquisition and model calibration to hyperparameter tuning and inverse analysis for joint input-state estimation.