AUTOPROGRESSIVE TRAINING OF NEURAL-NETWORK CONSTITUTIVE MODELS

A new method, termed autoprogressive training, for training neural net works to learn complex stress-strain behaviour of materials using glob al load-deflection response measured in a structural test is described . The richness of the constitutive information that is generally impli citly contained in the results of structural tests may in many cases m ake it possible to train a neural network material model from only a s mall number of such tests, thus overcoming one of the perceived limita tions of a neural network approach to modelling of material behaviour; namely, that a voluminous amount of material test data is required. T he method uses the partially-trained neural network in a central way i n an iterative non-linear finite element analysis of the test specimen in order to extract approximate, but gradually improving, stress-stra in information with which to train the neural network. An example is p resented in which a simple neural network constitutive model of a T300 /976 graphite/epoxy unidirectional lamina is trained, using the load-d eflection response recorded during a destructive compressive test of a [(+/-45)(6)](s) laminated structural plate containing an open hole. T he results of a subsequent forward analysis are also presented, in whi ch the trained material model is used to simulate the response of a co mpressively loaded [(+/-30)(6)](s) structural laminate containing an o pen hole. Avenues for further improvement of the neural network model are also suggested. The proposed autoprogressive algorithm appears to have wide application in the general area of Non-Destructive Evaluatio n (NDE) and damage detection. Most NDE experiments can be viewed as st ructural tests and the proposed methodology can be used to determine c ertain damage indices, similar to the way in which constitutive models are determined. (C) 1998 John Wiley & Sons, Ltd.