Data-based leakage detection and uncertainty quantification in the manufacturing of large-scale CFRP components

Abstract

Autoclave processing stands as a pivotal technique for manufacturing aircraft components from carbon fiber reinforced polymer parts (CFRP).This process involves curing stacked and impregnated materials with the application of heat and pressure within an airtight vacuum bag. The maintenance of vacuum integrity holds paramount importance in ensuring the top-notch quality of the components being produced. Leakages can result in porosity and voids making the final product unfit for use. While detecting the presence of leakage in the vacuum bag through residual mass flow is relatively straightforward, accurately determining the location of the leakage poses a significant challenge. Various methods employed for leakage detection in industrial processes heavily relies on time-consuming manual labor and expert craftsmanship, resulting in considerable production delays. The objective of this project is to develop a machine learning-based methodology that effectively leverages flow rates measured at vacuum ports to precisely and efficiently localize leakages in industrial scale vacuum setups. Initially, we employ a deep neural network to address the scenario of a single leak in the vacuum bag. Subsequently, we investigate various methods like multi-task learning, masking etc. to handle cases involving multiple leaks in vacuum bags. Extensive experiments demonstrate promising results for our multi-leak model in addressing scenarios with multiple leaks.