Development of Constitutive Laws for Whisker Reinforced Ceramics: A Neural Network Approach

Abstract

The formulation of constitutive laws for ceramic-matrix-composites (CMCs) using the mathematical treatment involving arbitrary scalar constants is a difficult task due to a large number of parameters, their complex interaction, and involvement of a weak bi-material interface in the mechanics of failure. A weak bi-material interface is necessary in the case of CMCs to avoid catastrophic failure. Because of the presence of such weak interface, slip and debonding occurs at the interface making the mechanics complex. As a result, analytical models to simulate the material behaviour become rather complicated, involving a number of equations for modelling the material behaviour. Further, a number of scalar constants are used in these equations to include the material non-linearity. Calibration of these constants warrants a comprehensive experimentation, which is very difficult in case of ceramic composites due to a variety of reasons. In the present paper, a novel approach to modelling the behaviour of whisker reinforced CMCs, using the artificial neural networks approach, was presented. An artificial neural network was used to postulate the constitutive law for A1 2 0 3 (matrix)/SiC (whisker) composite. Finite element analysis was carried out to generate the training examples for the network. The singularities on the output surface were successfully taught to the network using a hybrid training strategy involving manual as well as automatic learning. The network training is demonstrated and the validation of the network constitutive model for new examples is presented.

Neural Networks in Materials Science.

“…For an Al203 matrix SiC whisker composite a constitutive law has been derived using an artificial neural network, using inputs generated by finite element analysis. 70) Hybrid models can be created by training neural networks on data generated by physical models.…”