On the application of machine learning techniques to map porosity across carbon fibre reinforced polymer layers

Abstract

Carbon Fibre Reinforced Polymer (CFRP) composites are extensively used in the Automotive industries due to their excellent structural and mechanical properties. However, the occurrence of porosity within these materials can significantly affect their performance and durability. Porosity, defined as void inclusion, often occurs during the manufacturing process for these materials. Even for small amounts of porosity, this defect can alter the composite’s mechanical properties by reducing its inter-laminar shear strength. It is therefore important to characterise and accurately map this defect, characterising the porosity distribution within CFRP layers. In this work, a Finite Element method that accounts for circular cross-section pores subjected to an ultrasound excitation is developed. This simulated data is then used to apply a Machine Learning (ML) technique such as Convolutional Neural Networks (CNN) to characterise the porosity within the CFRP sample. This technique leverages the capabilities of ML algorithms to analyse and interpret ultrasound data for porosity detection. By training the ML model on a dataset of ultrasound images and corresponding porosity measurements, the model can learn patterns and features indicative of porosity. Results obtained for the simulation data are presented and discussed. The application of CNN in processing ultrasound data has shown exceptional potentials in identifying and quantifying porosity. Results obtained after applying this technique to real ultrasound data measured with an immersion tank are also presented. CNN technique shows interesting capabilities for extracting defects such as porosity from complex ultrasound data. This work contributes to a vast project that aims at underpinning the design of more efficient composite structures.