In structural engineering, the application of the finite element (FE) method to model a spatially-distributed system exploits physical insight into the plant dynamics and often results in a lumped model of large order, which is undesirable or even impossible for model-based controller design. This paper bridges the gap between FE modelling and distributed identification by proposing a novel identification technique to extract a distributed model from an FE model for both parameter-invariant and parameter-varying systems. The obtained models define the system dynamics on subsystems and capture the interacted behaviour among neighbouring subsystems. The proposed approach is tested on an experimental beam structure equipped with an array of piezo actuators and sensors, and the model accuracy is compared to that of blackbox identification.