Application of machine learning for composite moulding process modelling

Abstract

Fibre-reinforced composites are commonly manufactured through moulding processes such as Resin Transfer Moulding (RTM) due to their great reliability and scalability. State-of-the-art RTM process modelling and simulation primarily rely on computationally expensive physics-based modelling methods. Given the great volume of possible input-output combinations, process optimisation via exhaustive physics-based simulations or experiments becomes unfeasible. Hence, this paper proposes the integration of machine learning to facilitate composite moulding process modelling. The well-established PixelRNN, an image-based machine learning model, is employed to model and simulate the complex composite mould filling phenomenon. Upon training and validation, the developed PixelRNN metamodels demonstrated impressive prediction accuracies of up to 97.35 % at 50 % training data proportions, at roughly half the cost of exhaustive simulations. The results of this early study convincingly underscore the promising potential of machine learning in composite moulding applications, particularly when utilising graphical input data rather than the traditional numeric data.