Abstract

It is generally known that the accuracy of the finite element analysis is mainly dependent on the choice of an appropriate constitutive model that represents the material behaviour. In the past two decades some research has been done on the development of appropriate constitutive models using data mining techniques such as artificial neural net work ANN or evolutionary polynomial regression EPR and their implementation in the finite element method [e.g., 1].

The conventional approach to constitutive modelling using data mining techniques requires a significant amount of data which could be costly and not available at all cases. Furthermore, obtaining a homogenous stress strain state in experiments could be very challenging especially for complex loading conditions. This paper presents an EPR- based self-learning finite element method to address these problems. A different approach is illustrated for training of EPR-based constitutive model. The proposed method takes advantage of the rich stress-strain data embedded in non-homogeneous structural tests. The proposed approach does not require extensive experimental testes to be carried out. Therefore, the number of experimental tests could be reduced. Load-displacement data collected from experiments are used to iteratively train EPR- based constitutive model using finite element simulation.

The application of the self-learning finite element model is presented in simulation a geotechnical engineering problem. It is shown that the proposed method is very effective in capturing and representing the constitutive behaviour of geomaterials with high accuracy.

References

[1] Faramarzi, A. (2011). Intelligent computational solutions for constitutive modelling of materials in finite element analysis. PhD thesis: University of Exeter.