This work introduces a new strategy for real-time parameter identification or updating in structural mechanics models, defined as dynamical systems. The main idea is to use the Unscented Kalman filtering, which is a Bayesian data assimilation method, in conjunction with an energetic method for inverse problems, the modified Constitutive Relation Error. A PGD model reduction is also introduced to perform real-time computations. The proposed approach is illustrated, and compared to classical approaches, on several applications.