K-BP neural network-based strain field inversion and load identification for CFRP

Abstract

The strain information and loads conditions of composite wings are important basis for aircraft health evaluation. In this paper, firstly to demonstrate the accuracy of Fiber Bragg Grating (FBG) sensors and provide guidance for the following study, experiment is carried out to study the effect of the adhesive layer thickness on the strain transfer. Then finiteelementanalysissoftware ABAQUS is applied to analyze the strain distribution, and the simulation data is used to fit the pseudo-Kriging interpolation model to invert the strain information of crucial points. Finally, K-BP model is proposed by combining the interpolation model with back propagation (BP) neural network, which can be used to improve the accuracy on both strain field inversion and load identification. The results show that the proposed model can achieve great load identification results with fewer samples and optimize the interpolation parameters at the same time, which could provide important basis for evaluating the accuracy of strain field.