Creep modelling of Polyolefins using artificial neural networks

Notwithstanding the increasing demand for polymeric materials in an extraordinary variety of applications, the engineers have often only limited tools suitable for the design of parts made of polymers, both in terms of mathematical models and reliable material data, which together constitute the basis for a finite-elements based design. Within this context, creep modelling constitutes a clear example of the needs for a more refined approach. An accurate prediction of the creep behaviour of polymers would definitely lead to a more refined design and thus to a better performance of the polymeric components. However, a limited number of models is available within the f.e. codes, and when the model complexity increases, it becomes sometimes difficult fitting the models parameters to the experimental data. In order to predict the polymer creep behaviour, this paper proposes a solution based on artificial neural networks, where the experimental creep curves are used to determine the parameters of a neural network which is then simply implemented in an Abaqus user subroutine. This allows to avoid the implementation of a complex material law and also the difficulties related to match the experimental data to the model parameters, keeping easily into account the dependence on stress and temperature. After a discussion of the selection of the appropriate network and its parameters, an example of the application of this approach to polyolefins in a simplified test case is presented.