Landing gear condition monitoring based on back propagation neural network-based on multi-strategy cooperative optimization

基于多策略协同优化神经网络的起落架状态监测

¹ School of Aeronautic, Northwestern Polytechnical University, Xi'an 710072, China
² Beijing Institute of Aerospace Systems Engineering of China Aerospace, Beijing 100076, China

Received: 2 July 2022

Abstract

To effectively monitor the operation state of landing gear, a back propagation neural network-based on multi-strategy cooperative optimization(MSCO-BPNN) is proposed. The multi-strategy optimization algorithm is composed of chaotic mapping strategy, adaptive spiral capture strategy, crossover mutation strategy and whale optimization algorithm(WOA). WOA is applied to find the optimal hyperparameters of back propagation neural network(BPNN). The search efficiency, multi-local search ability and global search performance of model can be improved by using chaotic mapping strategy, adaptive spiral capture strategy and crossover mutation strategy. The BPNN with optimal hyperparameters is introduced to establish the implicit model of input parameters and output responses. Based on quick access recorder(QAR) data, landing gear left side brake temperature is act as the monitoring objective of this paper. The validity and applicability of MSCO-BPNN are verified by compared with WOA-BPNN, particle swarm optimization BPNN and traditional BPNN. The results show that MSCO-BPNN can monitor the operation status of landing gear with high efficiency and accuracy. The efforts of this paper provide a promising insight for the precise condition monitoring of complex structures.

摘要

为了有效监测飞机着陆阶段起落架运行状态, 提出一种基于多策略协同优化的反向传播神经网络算法(back propagation neural network-based on multi-strategy cooperative optimization, MSCO-BPNN)。多策略优化算法由混沌映射策略、自适应螺旋捕获策略、交叉变异策略及鲸鱼优化算法(whale optimization algorithm, WOA)组成, 其中WOA旨在捕获神经网络的最优超参数值, 混沌映射策略、自适应螺旋捕获策略及交叉变异策略分别用于提升鲸鱼优化算法在寻优过程中的搜索效率、多局部搜索能力及全局最优搜索性能。具备最优超参数的BP神经网络用于建立输入参数与输出响应间的隐式模型。以机载快速存储记录(quick access recorder, QAR)数据中起落架左侧刹车温度运行状态为对象进行健康监测, 通过对比WOA-BPNN、粒子群优化BPNN、传统BPNN与所提MSCO-BPNN算法, 分别验证所提算法在起落架运行状态监测建模方面的有效性和适用性。结果表明MSCO-BPNN能够以高效率和高精度进行起落架运行状态监测, 可推广至复杂系统的运行健康监控领域。