Data-Based Leakage Detection in the Manufacturing of Large-Scale CFRP Components

Abstract

The manufacturing of components from carbon fiber reinforced polymers (CFRP) requires that composite preforms are cured under heat and pressure. In corresponding processes, vacuum bags are used to apply even pressure to a component surface. In practice, vaccuum bags often leak -- Haschenburger et al. (2019) estimate one in ten cases. This is a serious problem as untight vacuum bags can crucially impair the quality of the final product. Hence, leakages need to be identified, located and repaired or, if this is not possible, the vacuum bag must be replaced. The replacement of entire vacuum bags is time-consuming and costly, especially in case of large-scale components. Consequently, there is a demand for fast, accurate and robust leakage detection in CFRP production to save process time and cost. As a result of the market analysis and comparison of available technologies in Haschenburger et al. (2019), and in close exchange with the manufacturing industry, a two-stage process for leakage detection has been proposed that combines the advantages of sensor-supported leakage detection (speed) and infrared thermography (accuracy). This two-stage process provides that a sensor-based prediction or narrowing of the leakage position is made first. In a subsequent step, a thermografic camera is used to refine and visualize the exact position. Regarding the first step, the authors of Haschenburger et al. (2021) proposed to leverage volumetric flow rates at multiple vaccuum ports to further improve and speed up the process. In this contribution, we discuss machine learning based approaches that use volumetric flow rates as input to predict positions of one or multiple leakages in the vacuum bag. More specifically, the presented methods constitute examples of informed machine learning as they are not purely data-driven, but also take prior knowledge of sensor positions and component geometry into account.