Using Artificial Neural Network in Reverse Design of Fiber Reinforced Plastic Composite Materials

Abstract

Composite materials are composed of two or more constituents, wherein the favorable properties of each material are contributed to improving the overall properties of the final composite material. Performing homogenized properties of composites is often challenging because it has a strong association with basic constituents at the micro-structure level. Therefore, the inverse approach for designing composite materials is a modern technique that can provide sophisticated theoretical support for composite materials.

In this study, an Artificial Neural Network (ANN) was employed to predict the parameters of the basic constituents on a micro-scale based on the final homogenized properties demanded by the designer.  the necessary data was derived from Finite Element Method (FEM) model.

A micro-level structure was used to conduct the homogenization analysis, which consisted of the reinforcing phase (fiber) and supporting phase (matrix). While the required data for building the ANN model was obtained using the FEM model of the composite unit cell in conjunction with the Monte Carlo Simulation. Then, The input features were mapped to the output features by utilizing Backpropagation (BP) algorithm in the neural network.

The input variables were the homogenized properties of demanded composite material. While the output was the properties of the constituent’s materials (i.e. fiber, fiber diameter, and matrix).

The outstanding performance of the reverse ANN model was revealed through a low value of mean square error (MSE) with a value of 0.00033, and also the coefficient of determination (R2) value which approached one. The contribution of this study is to produce an Artificial Neural Network (ANN) model, which offers a faster and highly accurate approach for obtaining the properties of composite constituents. This, in turn, provides significant practical engineering value in designing composite materials.