How AI Predicts Wing Failures: A Plain Look at Safer Composite Aircraft

What is it about?

This study introduces a new method for predicting failures in a three-dimensional woven composite wing box by combining a deep neural network (DNN) with an explainable AI approach called DeepSHAP. By cutting the prediction time from about 3 hours—typical for traditional experimental and numerical methods—to just 0.13 seconds, the DNN model accurately forecasts failure indices based on variations in fill yarn layers. In addition, the use of DeepSHAP allows for a clear breakdown of how different components, such as the skin, spar, and rib, contribute to the failure prediction. The analysis revealed that the skin plays the most critical role, which in turn enhances the model's reliability and transparency. This work not only tackles the challenges posed by the complex yarn patterns in 3D woven composites but also addresses the often-cited "black box" problem of AI models.

Featured Image

Photo by Sofiane Gargouri on Unsplash

Photo by Sofiane Gargouri on Unsplash

Why is it important?

The importance of this research lies in its ability to drastically reduce prediction time while simultaneously providing a trustworthy explanation of the AI’s decisions. Such an approach is invaluable for safety-critical applications like aerospace structures, as it lays the groundwork for rapid yet reliable failure predictions. The integration of DNN and XAI represents a significant step forward in developing predictive models that effectively consider the intricate characteristics of composite materials.

Perspectives