Machine Learning Based Predictive Model for AFP-Based Unidirectional Composite Laminates

Manufacturing of composites using automated fiber placement (AFP) is a complex process that involves large number of processing conditions and variables. Improper selection of these parameters adversely affects the quality and integrity of the manufactured laminates. Thus, it is important to develop a predictive model that can assess how changes in critical process conditions alter the outputs of the manufacturing process. The goal of this investigation is to learn the complex behavior of composites by developing an intelligent model, which can subsequently be used for the prediction of various characteristics of the composites. However, manufacturing of AFP composites is both expensive and time-consuming and therefore the available data samples are less, from the prospective of machine learning, which leads to the small data learning problem. This article first solves this problem through virtual sample generation, and then a neural network based predictive model is developed to accurately learn the complex relationships between various processing parameters in AFP.