Artificial Neural Networks to Model the Non Linear Behaviour of Hierarchical Composites

In this paper we show some different concepts for the use of Artificial Neural Networks [1-4] in modelling of composites and hierarchical structures. First, the possibility to identify the effective material properties from real experiments or numerical simulations or a combination of the two is investigated. Starting from a relatively small set of suitable numerical experiments performed on a unit cell, a proper set of corresponding input-output data is created to train an artificial neural network to identify the effective properties. Furthermore, ANN based procedures can be exploited in a multiscale analysis as a tool for the stress-strain recovery at lower levels of a hierarchical structure and/or to estimate the state of yielding of the materials. This kind of applications is of great computational importance, since with material non linearity it allows for a significantly improved computational efficiency. Finally ANNs may be used to define the homogenised properties for a class of cells depending on few geometrical parameters. This scheme can be applied in many engineering situation: from analysis of Functionally Graded Materials to composite with random voids (e.g. composites with cavities of not reacted alloys). This approach is applicable also when the materials depend on parameters like temperature, damage, or state of yielding. In such a case the method allows to save a huge amount of computational time, replacing the solution of several BVPs by a simple run of ANN in recall mode. The problem of the best ANN (or sufficiently good ANN) for each type of application is discussed by means of the examples presented.