• Visrolia, Amit
• Druiff, Phil P. J.
• Dellanno, Giuseppe
• Arruda, Mauro
• Ward, Carwyn
• Ma, King
• Bolduc, Sean
• Palardy-Sim, Marc

Abstract

The quality and consistency of the components manufactured by Automated Fibre Placement are dependent on multiple process parameters and their interactions. In order to capture the required data, multiple data collection systems are used which output data in different formats, frequencies and reference systems. This results in a long process with manual steps, which can be error-prone and time consuming, taking as much as 60-70% of the total engineering effort. This study implements a data collection and visualisation platform designed to unify and automate data capture from multiple sources and facilitate the deployment of ML models. <br/>As a demonstration of the platform, multiple components of complex and varying geometry are manufactured at the National Composites Centre, UK. A number of process parameters, including compaction force, surface temperature and lay-up speed, are measured continuously and are used to train and deploy machine learning models with the aid of the smart platform. Data is visualised and a ply-wise neural networks-based model is developed, achieving accuracies of up to R2 = 0.80; demonstrating that the data-driven approach is scalable to complex geometries.<br/>

Topics

• impedance spectroscopy
• surface
• composite
• machine learning