Data-Efficient Machine Learning for in-situ Curing-aided Additive Manufacturing

Abstract

In-situ curing assisted additive manufacturing (AM) of thermoset composites received numerous attentions and accurate prediction of the degree of cure (DOC) in thermosets is essential to online tailor process parameters for controlling the resultant quality and performance. Traditional methods for measuring DOC, such as Differential Scanning Calorimetry (DSC), Dynamic Mechanical Analysis (DMA), and Dielectric Analysis (DEA), are destructive or invasive, and difficult for a online test. In this paper, we present a data-driven approach to predicting DOC using machine learning. Our approach overcomes the limitations of traditional methods by considering localized variations, adapting to complex curing kinetics, and directly predicting DOC. We compare two machine learning algorithms: support vector classifier (SVC) and random forest, using a small dataset. SVC achieved an accuracy of 81%, a precision of 0.86, and a recall of 0.81, while random forest achieved an accuracy of 78%, a precision of 0.83, and a recall of 0.78. This demonstrates
the feasibility of data-driven DOC prediction in AM, paving the way for enhanced manufacturing processes.