A method and system of monitoring a structure, the method including: a) synchronously acquiring data comprising a plurality of operational parameters and at least one strain data, b) building a significant points dataset from the data acquired in step a), and c) modelling a relationship between the operational parameters and the strain data using the built significant points dataset to train a non-adaptive prediction functional supervised approximation method, wherein the step of building a significant points dataset comprises deletion of redundant information from the acquired data. The resultant models may be used to process structure real operation data in order to estimate the eventual crack initiation and crack growth on a set of predefined locations of the structure.