Testing Pyramid on a Diet: Accelerating Material Testing Through Machine Learning Models

Kurzfassung

Propellers manufactured with fibre-reinforced polymers (FRP) are increasingly used in Advanced Air Mobility (AAM), specifically for Normal-Category Aeroplanes (CS-23). The design and certification of these composite structures rely on the testing pyramid, also known as the building block approach. Traditionally, this process relies on extensive coupon-level testing to establish design allowables. Currently, the production and testing of CS-23 composite propellers are constrained by manual processes. There is also a heavy reliance on physical experiments. High cost and time are required to acquire sufficient material data through experiments. To address these challenges, the aerospace industry is increasingly integrating machine learning (ML) models to shorten development cycles and accelerate the certification of FRP composite structures. As a first step, this study focuses on low-velocity impact (LVI) testing as a use case. Understanding impact damage, such as barely visible impact damage (BVID), is critical for demonstrating the performance and safety of composite structures. Firstly, a finite element (FE) simulation model for LVI testing is established and rigorously validated against real experimental data. Secondly, the calibrated FE model is used to generate a comprehensive dataset that broadly represents the good design space. Finally, surrogate models based on Gaussian Processes and Artificial Neural Networks are developed. These ML-based surrogate models with an inverse inference model can replace computationally expensive FE simulations. The surrogate model enables virtual exploration of various impact conditions. Consequently, the surrogate model reduces the number of physical experiments required, as only a limited set of design aspects needs to be considered in the test pyramid. This enables a certified composite propeller to be realised much faster. Future collaborations with regional industry partners will support practical application and continuous refinement of these predictive tools. Integrating data-driven models helps propeller manufacturers streamline the various stages of testing and accelerate the certification process for composite structures.