基于CNLS-MMD的孪生混合网络四旋翼飞行器故障诊断方法  

Fault Diagnosis Method of Siamese Hybrid Neural Network Quadrotor Aircraft Based on CNLS-MMD

作　　者：张雅婷[1] 杨蒲[1] 孟宪锋 陆宁云[1] 文琛万 Zhang Yating;Yang Pu;Meng Xianfeng;Lu Ningyun;Wen Chenwan(Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China;AVIC Xi’an Flight Automatic Control Research Institute,Xi’an 710065,China)

机构地区：[1]南京航空航天大学,江苏南京210016 [2]航空工业西安飞行自动控制研究所,陕西西安710065

出　　处：《航空科学技术》2023年第12期125-134,共10页Aeronautical Science & Technology

基　　金：航空科学基金(20200007018001);直升机传动技术重点实验室基金(HTL-O-21G11)。

摘　　要：四旋翼飞行器在众多领域中应用广泛,由于工作环境复杂多变,四旋翼飞行器极易出现结构损伤性故障,给飞行器的安全性带来巨大的挑战,因此开展四旋翼飞行器结构损伤性故障的相关研究对提高四旋翼飞行器可靠性具有重要意义。针对四旋翼飞行器在实际应用中结构损伤性小样本故障数据诊断率低的问题,本文提出了一种在小样本条件下基于卷积神经网络和长短时记忆网络的孪生混合网络(CNLS-MMD)四旋翼飞行器故障诊断方法。首先,设计试验获取四旋翼飞行器多工况结构损伤性飞行数据并对数据进行预处理。其次,建立基于孪生混合网络的故障诊断模型,采用卷积神经网络(CNN)和长短时记忆网络(LSTM)构建CNLS混合模型提取数据特征,利用最大均值差异(MMD)衡量样本的相似度,实现对故障标签的预测。最后,选择不同样本数量的训练集训练模型,使用多工况小样本数据集对搭建的模型进行故障测试。结果表明,该故障诊断方法具有较好的诊断性能和泛化能力。Quadrotor aircraft has been widely used in many fields.Due to the complex and changeable working environment,quadrotor aircraft is prone to structural damage faults,which brings great challenges to the safety of aircraft.Therefore,it is of great significance to study the structural damage faults of quadrotor aircraft for improving the reliability of quadrotor aircraft.Aiming at the problem of low diagnosis rate of structural damage small sample fault data of quadrotor aircraft in practical application,this paper proposes a siamese hybrid neural network(CNLS-MMD)quadcopter fault diagnosis method based on convolutional neural networks and long short-term memory network under small sample conditions.Firstly,an experiment is designed to obtain the multi-condition structural damage flight data of the quadrotor aircraft and to preprocess the data.Secondly,a siamese hybrid neural network based fault diagnosis model is established,and a CNLS hybrid model is constructed using Convolutional Neural Networks(CNN)and Long Short-Term Memory(LSTM)network to extract data features,and Maximum Mean Discrepancy(MMD)is used to measure the similarity of samples to achieve the prediction of fault labels.Finally,training sets with different sample sizes are selected to train the model,and the built model is tested for faults using a small sample data set with multiple working conditions.The results show that the fault diagnosis method has good diagnostic performance and generalization ability.

关 键 词：孪生网络 四旋翼飞行器 故障诊断 小样本 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]