A deep learning-based method for calculating aircraft wing loads

The purpose of this paper is to propose a novel aircraft wing loads calculation model, called long short-term memory residual network (LSTM-ResNet), which can evaluate the loads based on the strain distribution. To achieve this goal, firstly, the data acquisition experiment is designed and performed with a real aircraft wing. In this experiment, we used the Fiber Bragg Grating (FBG) technology as the measurement method to collect strain-load data from the aircraft wing. Then, we propose the LSTM-ResNet model with the one-dimensional convolutional(1D-CNN) architecture. This model is capable of extracting the temporal and spatial representational information from the strain-load data of the aircraft wing. Experimental results demonstrate that the proposed method effectively evaluate the loads of the aircraft wing. To prove the superiority of LSTM-ResNet model, we compared the proposed model with existing loads calculation methods on our experimental dataset. The results show it has a competitive average relative error (0.08%). Moreover, those promising results may pave the way to use the deep learning algorithm in aircraft wing loads calculation.