Artificial neural network for predicting the mechanical performance of additive manufacturing thermoset carbon fiber composite materials

Abstract

Abstract Composites have been evolved rapidly due to their unique performance in comparison with other conventional materials, such as metals. Although additive manufacturing (AM) has attracted considerable attention in recent years to produce reinforced complex composite structures as in reinforced carbon fiber composites, it is difficult to control the fiber content concentration within the composites to obtain tailored materials properties, especially at high loads of fibers. In fact, high load of fibers usually leads to technical issues, such as nozzle clogging and fiber agglomeration that hinder the 3D printing process. Therefore, a customized artificial neural network (ANN) system was developed in this work to predict the mechanical characteristics of 3D printing thermoset carbon fiber composites at any carbon fiber concentration. The developed ANN system was consisting of three model techniques for predicting the bending stress as well as the flexural modulus of the thermoset carbon fiber composites, even when handling small experimental datasets. The system architecture contained connected artificial neurons governed by non-linear activation functions to enhance precise predictions. Various schemes of ANN models were utilized namely: 1-4-1, 1-4-8-1, and 1-4-8-12-1 models. The developed models have revealed various accuracy levels. However, the 1-4-8-12-1 model has demonstrated a very high level of predictions for the mechanical performance of the AM epoxy/carbon fiber composites. This would enhance predicting the performance of such composites in 3D printing with very minimal experimental work to optimize the fiber content for the desired overall mechanical performance.