A neural network enhanced system for learning nonlinear constitutive relation and failure initiation criterion of fiber reinforced composites

摘要

An artificial neural network (ANN) enhanced system containing a subsystem with one or multiple ANN is proposed. Instead of directly defining the loss function as the output of an ANN model, the proposed method uses the system output to define the loss function, which is contributed by the outputs from one or multiple ANN models through a subsystem. As a result, a set of new back-propagation equations have been derived for this system. Two examples are given using the proposed system: learning the nonlinear in-plane shear constitutive law and learning the failure initiation criterion of fiber reinforced composites (FRC). The ANN models in both examples are trained at the lamina level using the measurable experimental responses of laminates as the input and output. The results obtained from the learned ANN models agree well with the corresponding analytical solutions. Moreover, the proposed approach is further extended to couple ANN with finite element method (FEM) which could be used in more complex systems. The above proposed methods can be used to train ANN models in a subsystem using the measurable input and output of the entire system.