In many applications, such as control, monitoring, assessment, and failure detection in mechanical and structural systems, the state of the system is sought. However, in almost all cases, the full state vector is not directly measured, and it must be reconstructed from sparse measurements contaminated by noise. If in addition, the excitations are not (or cannot be) measured, such as the case with wind loads, wave loading, and others, the problem becomes more challenging and standard deterministic methods are not applicable. This paper presents a model-based state estimation algorithm for online reconstruction of the complete dynamic response of a partially instrumented structure subject to realizations of random loads. The proposed algorithm operates on noise-contaminated measurements of dynamic response, a finite element model of the system, and a spectral density description of the random excitation and noise. The main contribution of the paper is the development of a second-order estimator that can be directly implemented as a modified version of the finite element model of the system while minimizing the state error estimate covariance. The proposed observer results in a suboptimal Kalman filter that can be implemented as a modified version of the original finite element model of the system with added dampers and collocated forces that are linear combinations of the measurements. The proposed method is successfully illustrated and compared with the standard Kalman filter in a mass-spring-damper system under various ideal and nonideal conditions. trajectory of a system subject to unmeasured disturbances based on a state-space model of the system and noise-contaminated measurements. The algorithm was initially proposed in its present form by R. E. Kalman in the 1960s, and it has since found widespread applicability in many engineering fields [10][11][12][13][14].The KF operates by using a feedback gain matrix that is optimal in the sense of balancing the effect of noise in the measurements used as feedback and the effect of unmeasured disturbances to minimize any linear function of the trace of the state error covariance matrix. In the case of structural dynamics and vibration monitoring, the disturbances correspond to unmeasured excitations, and the measurements typically consist of accelerations and/or strain time histories at discrete locations within the structure.The original derivation is given for linear systems in which measurement noise and disturbances are independent realizations Gaussian random processes and for which the model is known with accuracy. In real world application, these assumptions are seldom fully satisfied; however, experience in many applications has shown that the KF can also operate robustly in nonideal conditions. This will be illustrated by means of simulations in the example at the end of the paper. For the implementation of the KF in nonlinear applications, various modifications of the original algorithm known as extended KFs (EKFs) have been proposed [15,16]. Other methods fo...