A Supervised Scene Adaptive Model for Identifying Impact Load with Few Samples

Deep learning-based impact load identification technology for the next generation of large aircraft structures has garnered significant attention and has become one of the focal points in aircraft structural health monitoring. However, this technology relies on a large number of training samples and exhibits poor scalability. One of the current challenges in system-level multi-structure monitoring is how to construct deep learning models with a small number or even zero impact training samples, and improve the models' ability to migrate between different structures. To address this challenge, a novel method for impact load identification using only a small number of samples, based on a supervised scene adaptive model, is proposed. The performance of the model is validated on real aircraft structures. For large and complex structures, the model can be applied to other similar structural areas or different structural areas by using samples from the baseline area for training. Then, a very small number of calibration samples from the migrated area can be used for calibration. The results demonstrate that the proposed model, calibrated with just a single sample, achieves 97.22% accuracy in impact location identification and 99.44% accuracy in energy identification under similar regional structural conditions. Under different structural region conditions, the location identification accuracy of the proposed model is 87.65%, while the energy identification accuracy remains at 98.85%. The position identification accuracy of the model is 91.98% under different impact energy level conditions, and the identification accuracy remains at 87.04% even under varying impact energy levels and structural region conditions.