基于改进ABC-RBF神经网络的飞机全电刹车系统故障自动诊断方法

An Automatic Fault Diagnosis Method for Aircraft All Electric Braking System Based on Improved ABC-RBF Neural Network Optimization

摘要:

飞机全电刹车系统在飞机的着陆与起飞阶段起到关键作用,随着该系统自动化程度逐渐升高,如何确保该系统运行的安全性、稳定性、可靠性成为亟待解决的问题。现有的故障诊断方法存在存在诊断平均误差值较高、耗时较长的问题,设计基于改进ABC-RBF神经网络的飞机全电刹车系统故障自动诊断方法。设计采用“USB接口+ARM+FPGA”的硬件架构方式,由上位机、信号衰减电路等构成的故障信号采集器,实施飞机全电刹车系统故障信号采集。对于采集信号,设计基于互信息与变分模态分解(Variational Mode Decomposition, VMD)的信号降噪算法对其实施降噪处理。采用改进后的ABC算法对RBF神经网络参数进行寻优,确保寻优参数的有效性。并引入模糊集合的概念来提高网络的性能,利用梯度下降法进行网络训练更新,降低诊断结果误差。最后,输入降噪信号,利用优化训练后的RBF神经网络实现飞机全电刹车系统的故障自动诊断。实验测试结果表明,该方法的偏离因子值最低达到0.08×10-3,三种故障的平均诊断迭代时间均较短,其中主起落架“走步”故障的平均诊断迭代时间最短。

Abstract:

The all electric braking system of an aircraft plays a crucial role in the landing and takeoff stages. As the automation level of the system gradually increases, how to ensure the safety, stability, and reliability of its operation has become an urgent problem to be solved. The existing fault diagnosis methods have the problems of high average diagnostic error and long time consumption. A fault automatic diagnosis method for aircraft electric braking system is designed based on the improved Artificial Bee Colony Algorithm (ABC) optimized radial basis function network (RBF). The design adopts a hardware architecture of "USB interface+ARM+FPGA", consisting of a fault signal collector composed of an upper computer, signal attenuation circuit, etc., to implement fault signal acquisition for the aircraft"s all electric braking system. For collecting signals, design a signal denoising algorithm based on mutual information and Variational Mode Decomposition (VMD) to implement denoising processing. Using the improved ABC algorithm to optimize the parameters of the RBF neural network, ensuring the effectiveness of the optimization parameters. And introduce the concept of fuzzy sets to improve the performance of the network, use gradient descent method for network training updates, and reduce diagnostic result errors. Finally, input the denoised signal and use the optimized trained RBF neural network to achieve automatic fault diagnosis of the aircraft"s all electric braking system. The experimental test results show that the minimum deviation factor value of this method is 0.08 × 10-3. The average diagnostic iteration time for the three types of faults is relatively short, among which the average diagnostic iteration time for the "walking" fault of the main landing gear is the shortest.