A deep learning empowered smart representative volume element method for long fiber woven composites

Abstract

In response to the global trend of carbon reduction over the last few years, various industries, including the aviation and automobile industries, have gradually begun research, design, and production of carbon fiber composite materials. These have excellent mechanical properties, such as being lightweight, high strength, and of high rigidity, which provide weight reduction and energy savings in applications across many fields. When used as a load-beam structure, the weave pattern determines the primary mechanical properties of the composite material. Therefore, the production of diverse products and components can be carried out using different patterns of weaving and manufacturing according to an application’s requirements. The mechanical properties of woven fiber composites can be obtained by using simulation analysis software, which can reduce unnecessary waste during design and manufacturing. However, difficulties arise in the simulation analysis due to the complexity of the weaving method. With the continuous improvement of computer technology in recent years and the enormous amount of training data available, many research teams have begun to implement artificial intelligence (AI) technology, which has been widely used to overcome long-standing obstacles in many different fields. For example, the problems involved in the prediction of protein folding sequences and the prediction of the physics of structural materials have all been resolved by AI. We implement a convolutional neural network (CNN), a deep learning method, to establish a model that utilizes a representative volume element for the prediction of the mechanical properties of a woven fiber composite material. The predictive model significantly streamlines the computational complexity involved in analyzing woven composite materials, resulting in a substantial reduction in processing time compared to conventional methods. Unlike traditional finite element simulations, which necessitate intricate boundary conditions and interactions on a case-by-case basis, our research simplifies these complex procedures and accommodates a wide range of scenarios. This research offers substantial advantages for industrial manufacturing, particularly in the design and mass production of woven fiber composite materials.