Vacuum bag leak detection using machine learning with simulated data from an analogous electric circuit

Vacuum-assisted composite manufacturing techniques, including resin transfer molding (VARTM) and prepreg methods, rely on atmospheric or autoclave pressure to consolidate fabric components. However, potential leakages on vacuum bags can cause air bubbles, voids, resin traps, uneven surface finishes, and weakened mechanical properties. Detecting and repairing leaks prior to the curing stage is essential to ensure high-quality manufacturing outcomes.

A common method for leak detection involves estimating their locations by monitoring the volumetric flow rates at multiple vacuum ports on the layup. However, traditional numerical approaches often fall short due to the extensive size of vacuum bags, complex geometries, and various vacuum port configurations. These challenges lead to inaccurate or unreliable predictions of leakage sites.

Machine learning (ML) offers a promising alternative, capable of predicting leakage locations using flow rate data. However, ML models require extensive datasets to accurately represent the flow characteristics of diverse layup configurations and leakage scenarios. Generating these datasets experimentally is resource-intensive, both in terms of time and labor. Conventional numerical solutions are also very time consuming.

To address this challenge, we propose a novel analogy between vacuum bag assemblies and electrical circuits to simulate flow rate data efficiently. This approach has been experimentally validated across various geometries and leakage patterns, demonstrating high accuracy in mimicking real-world setups. Furthermore, the method accommodates a wide range of leakage sizes within sensor limits.

Using this framework, we generated synthetic datasets to train regression-based ML models. These models achieved high validation accuracy and were further tested on unseen experimental data, where they demonstrated strong predictive performance. This integrated approach holds significant promise for improving leakage detection, enhancing manufacturing efficiency, and ensuring superior composite material quality.

Keywords: Vacuum bagging, composite manufacturing, machine learning, leak detection, leak localization