In order to reduce errors due to a manual analysis, we have automated the procedure through the digitisation of the films, and added to our analysis of the measurement chain the parameters linked to this operation, essentially the relation between the measured optical density and the illumination.

We have also developed a model which compute the densitometric profiles obtained during the control of cylindrical structures, with input data such as known geometrical data (thickness, radius), estimated or measured parameters describing the measurement chain and arbitrary values given to the unknown geometrical parameters (here the gaps between each cylinder).

The actual values for the unknown parameters are obtained after several interactions by fitting the actual densitometric profiles with computed profiles base on a set of arbitrary values. The conventional optimisation algorithms reduce the "target function" at each iteration, but are only able to converge towards the closest local minimum from the chosen starting point. the "simulated annealing" algorithm makes use of aleatory pullings for the variables in the model, which allows to explore unsuitable set of parameters, but can eventually help to cross "passes" in order to reach a better minimum for the "target function". The application which is presented here deals with the evaluation of small gaps in stacks of 3 cylindrical structures made with absorbent materials.

The application of this method makes possible a clear improvement of the quality of the measurements for the accuracy, the detection threshold and the repeatability. At this occasion, a software has been written which uses all the steps of this procedure and allows a serially processing of digitised pictures.