Vacuum bag leak detection with geometry-informed machine learning

Abstract

Carbon fiber reinforced polymers (CFRP) play an important role in various industries, including aerospace. Due to their light weight and high strength, these materials are increasingly used by manufacturers for various parts. In a typical manufacturing process, CFRP material is robotically applied to a mold and then cured in an autoclave using heat and pressure. A vacuum is used to apply uniform pressure to the part surface. The tightness of the vacuum is critical to ensure high quality. In practice, however, leaks in the vacuum bag are common. Therefore, the vacuum setup must be tested and any leaks must be located and repaired prior to curing. The detection of such virtually invisible leaks is very challenging and time consuming. In this work, we present a geometry-informed machine learning approach for leak localization based on volumetric flow rates at different vacuum ports. We evaluate the proposed method through empirical experiments, including an industrial-scale wing mold, and compare it to standard neural networks.