How Artificial Intelligence supports us to develop and qualify faster then certify

Abstract

Nowadays, Artificial Intelligence is increasingly used to develop and support progress in many fields and industries, such as finance, medical, transportation …, especially for complex problem resolution. The paper presents how Airbus Helicopters introduces Artificial Intelligence in Material & Process activities, aiming, amongst other things, to reduce the time to market and optimize qualification then certification costs/risks. The paper integrates the results of a proof of concept, achieved on flame resistance behavior of composite materials, related to interior compartment / cargo self-extinguishing requirements (CS27/29 §853 and 855) and demonstrates how Artificial Intelligence supports Engineering activities. The significant novelty introduced in this work is the use of advanced data-analysis software to support engineers and experts throughout development and qualification steps. Within this study, various AI models have been trained using available experimental datasets from Airbus Helicopters and suppliers as described in Figure 1. Following that, the trained AI model has permitted to identify the most influencing parameters and allowed to focus interest on both critical and optimal setups to help materials experts to reach targets in terms of material performance. In addition, AI model also allows to predict the fire behavior of the material, for resin/fiber reinforcement/fire agent combinations that have not been tested experimentally. This point could be particularly useful for material development purpose. The main concern when initiating this study was the very small amount of data available for material science compared to usual “big data” applications. In material science, the number of influencing parameters (eg input parameters of the AI model) describing the variability of the problem is rather large: detailed description of the material itself, parameters influencing the production process, material testing conditions, etc… Considering this amount of parameters and the small quantity of data available, one could expect that the capability of AI models trained on such limited data sets would be quite poor to predict accurately the actual behavior of materials. Nevertheless, as material behavior is enforced by physics and chemistry laws, it has been shown that a few hundred of experimental data are enough to get reasonably good predictions with the models. This work demonstrates that, thanks to Artificial Intelligence support, Airbus Helicopters has improved its understanding of complex phenomena like flame resistance behavior. Main influencing parameters have been identified for the different tests configurations. And for each parameter, strong/weak ranges have been established. Doing tests in such critical conditions during materials screening phase should help to avoid failing tests in representative helicopter configurations and permit to speed up helicopter development and certification. The presented study also paves the way for material and processes optimizations for helicopter designs.