This thesis takes place in the general context of the calibration for industrial application. It aims at evaluating the quality of a data base by checking that the data, with respect to our objectives, "best fill" the space. This work provides a synthesis of algorithmic and mathematic tools to achieve such a purpose. Extraction and importation techniques to improve the global quality of the data are proposed. These methods allow identifying some defaults of the data structure. An illustration of its application is exposed in the context of functional estimation with orthogonal functions.
