• Saifullah, Abu Naser Muhammad
• Giasin, Khaled
• Lupton, Colin John
• Barouni, Antigoni
• Zhang, Zhongyi
• Dhakal, Hom
• Papananias, Moschos

Abstract

The design and behaviour of advanced composite material systems have been investigated and studied for several decades now. A huge amount of time-consuming experimental tests supported by analytical and numerical models have been used extensively to gain a better understanding of the material’s behaviour and, ideally, predict the performance of a composite structure under specific loading conditions. Composite materials, being an inherently complex structure with more than one constituent, require extremely intensive computational effort to maintain sufficient accuracy of the numerical models in the behavioural prediction, with a highly time-consuming solution process. For the above reasons, this paper uses a set of statistical and machine learning modelling methodologies to optimise the design and manufacture of sustainable composite laminates made of flax and basalt fibres. A preliminary Design-of-Experiments (DoE) was constructed which included manufacturing parameters, such as temperature, pressure, and time of the curing cycle as well as variations in the material layers of the laminate. A series of laminates were manufactured using a hot press compression moulding process and experimental tests were performed to characterise the behaviour of each laminate. Machine Learning (ML) models, including Gaussian Process Regression (GPR) and Bayesian Regularized Artificial Neural Network (BRANN) models, were then developed, capable of predicting the mechanical properties of the laminate so that extensive experimental testing can be minimised.

Topics

• experiment
• composite
• curing
• machine learning