Study on fault diagnosis algorithms of EHA based on CNN-SVM

基于CNN-SVM的飞机EHA故障诊断算法研究

College of Civil Aviation, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China

Received: 18 May 2022

Abstract

Contrapose the highly integrated, complex working conditions and many kinds of faults of aircraft electro hydrostatic actuator(EHA), to diagnose the typical fault of EHA effectively, a fault diagnosis algorithm based on convolutional neural networks (CNN) and support vector machine(SVM) was proposed. Firstly, the fault date sets are entered on CNN for adaptive feature extraction, then the output of the fully connected layer of CNN are classified by using SVM. To improve the performance of SVM, dynamic inertia weight adaptive particle swarm optimization (IWAPSO) was used to optimize the SVM parameters. Finally, the sensitivity of SVM to noise was reduced by introducing ramp loss function. The results show that the accuracy of SVM after parameter optimization is 12.6% higher than that of standard SVM and 17.3% higher than CNN. The SVM based on the ramp loss function showed better robustness when using noisy test sets.

摘要

针对飞机电动静液作动器(electro-hydrostatic actuator, EHA)系统集成度高、工况复杂、故障种类多的特点, 为了对其典型故障进行有效诊断, 提出一种基于卷积神经网络(convolutional neural networks, CNN)和支持向量机(support vector machine, SVM)的故障诊断算法。使用CNN对故障数据进行自适应特征提取, 再利用SVM对CNN全连接层输出进行分类。为提高SVM分类性能, 使用带动态惯性权重的自适应粒子群优化算法(dynamic inertia weight adaptive particle swarm optimization, IWAPSO)实现对SVM参数的优化选择。引入Ramp损失函数降低SVM的噪声敏感性。结果表明: 经过参数优化后的SVM准确率比标准SVM提升了12.6%, 比传统CNN方法提升了17.3%;当使用含噪声信号的测试集时, 基于Ramp损失函数的SVM表现出了更好的鲁棒性。