An improved BP neural network algorithm for accurate prediction of impact defects depth in civil aircraft composite materials

Abstract

To address the problems of low prediction accuracy and poor stability in traditional BP neural networks applied to quantitative defect depth evaluation of composite materials in civil aircraft, this study proposes an improved defect depth prediction model based on a BP neural network optimized using a Tent chaos enhanced Particle Swarm Optimization algorithm (Tent-PSO-BP). This algorithm integrates the PSO algorithm to effectively determine the optimal initial parameters of the network, thereby enhancing prediction accuracy. Furthermore, the Tent chaotic mapping strategy is employed to initialize the population in the PSO algorithm, ensuring a uniform distribution of initial parameters and mitigating the stability issues commonly encountered in traditional BP neural networks during prediction tasks. Results from ablation experiments demonstrates that the improved Tent-PSO-BP model achieves a reduction of 97.10% in Mean Squared Error (MSE) and 93.01% in Mean Absolute Error (MAE), compared to the traditional BP network; Additionally, it improves the coefficient of determination (R²) by 34.56% and the Relative Standard Deviation (RSD) by 29.38%. Comparative experimental results show that, compared with existing algorithms such as Long Short-Term Memory (LSTM), Genetic Algorithm Optimized BP (GA-BP) and Whale Optimization Algorithm Optimized BP (WOA-BP), the Tent-PSO-BP model achieves improvements in R² by 12.01%, 11.00% and 8.80%, respectively, and enhancements in RSD by 24.51%, 32.01% and 35.92%, respectively. These results fully demonstrate the superior accuracy and stability of the Tent-PSO-BP model in predicting defect depth in civil aircraft composite materials. With the best overall performance among the evaluated methods, the model holds significant application potential in the field of infrared nondestructive testing in aviation and offers a novel research perspective for the quantitative analysis of composite material defects.