Innovative approaches for modelling of inelastic material behaviours (applications of neural networks and evolutionary algorithms)

摘要

This paper presents two approaches for the modelling of inelastic constitutive properties, each using a neural network or an evolutionary algorithm. In the first approach, two techniques are proposed to identify the parameter set of an existing constitutive model. One is to use evolutionary algorithms as an optimization method to minimize errors between the measured data and the corresponding data computed. In the other technique, an neural network is used as a parameter estimator given measured data as input. In the second approach, two neural networks are used as a mapping for the inelastic behavior off materials. These approaches were tested with the actual experimental data under uniaxial loading and stationary temperature and the results of the test show the effectiveness of the approaches.