广义回归神经网络在冗余捷联惯导故障诊断中的应用研究

针对冗余捷联惯导的故障诊断问题，研究提出了一种基于广义回归神经网络的故障诊断方法。该方法在传感器输出数学模型未知的情况下，仅通过传感器之间的冗余关系，利用传感器正常工作时的测量值和改进的神经网络估计输出值生成残差进行故障诊断。仿真试验表明，利用神经网络补偿产生的残差可以检测到未补偿时的故障。该方法不仅可以检测到单故障，还对多传感器同时发生故障具有一定的检测能力。     

Robust wavelet neuro control for linear brushless motors

Design, simulation and experimental implementation of a wavelet basis function network learning controller for linear brushless dc motors (LBDCM) are considered. Stability robustness with position tracking is the primary concern. The proposed controller deals mainly with external disturbances, e.g. nonlinear friction force and payload variation in motion control of linear motors. It consists of two parts, one is a state feedback component, and the other one is a learning feedback component. The state feedback controller is designed on the basis of a simple linear model, and the learning feedback component is a wavelet neural controller. The attenuation effect of wavelet neural networks on friction force is first verified by the numerical method. The learning effect of wavelet neural networks on friction force is also shown in the numerical results. Then, a wavelet neural network is applied on a real LBDCM to on-line suppress the friction force, which may be variable due to the different lubrication. The effectiveness of the proposed control schemes is demonstrated by simulated and experimental results.     

基于磨粒磨损机理的机械磨损状态监测

针对机械设备磨损状态监测准确率较低的问题,基于不同磨损机理下磨粒具有不同的形状和纹理特征,提出了一种基于磨粒特征识别的机械磨损状态监测的数学模型。通过形状特征识别球状磨粒和切削磨粒,结合形状、纹理特征识别疲劳磨粒和严重滑动磨粒,基于提取的特征参数建立机械磨损状态监测的特征向量,通过量子粒子群优化(QPSO)的径向基函数神经网络模型,实现对机械磨损状态的监测和判别。实验结果表明:QPSO-RBF神经网络数学模型结构简单,比传统PSO-RBF神经网络模型的识别准确率高5%,可用于常见机械磨损状态的检测。      

RFNN control for PMLSM drive via backstepping technique

A robust fuzzy neural network (RFNN) control system is proposed in this study to control the position of the mover of a permanent magnet linear synchronous motor (PMLSM) drive system to track periodic reference trajectories. First, an ideal feedback linearization control law is designed based on the backstepping technique. Then, a fuzzy neural network (FNN) controller is designed to be the main tracking controller of the proposed RFNN control system to mimic an ideal feedback linearization control law, and a robust controller is proposed to confront the shortcoming of the FNN controller. Moreover, to relax the requirement for the bound of uncertainty term, which comprises a minimum approximation error, optimal parameter vectors and higher order terms in Taylor series, an adaptive bound estimation is investigated where a simple adaptive algorithm is utilized to estimate the bound of uncertainty. Furthermore, the simulated and experimental results due to periodic reference trajectories demonstrate that the dynamic behaviors of the proposed control systems are robust with regard to uncertainties.     

含时滞结构线性时变系统伺服机构设计的PGOPO法

采用线性、连续、有界算子－分段广义正交多项式算子对含时滞结构的线性时变系统的线性伺服机构设计问题进行了研究,得到了一种形式简洁、计算精度较Ｗａｌｓｈ函数、块脉冲函数法更高的逼近方法。文中所列举的两个例子充分说明了分段广义正交多项式算子法在时滞系统伺服机构设计中的适用性和有效性。     

A point cloud deep neural network metamodel method for aerodynamic prediction

Aiming to reduce the high expense of 3-Dimensional (3D) aerodynamics numerical simulations and overcome the limitations of the traditional parametric learning methods, a point cloud deep learning non-parametric metamodel method is proposed in this paper. The 3D geometric data, corresponding to the object boundaries, are chosen as point clouds and a deep learning neural network metamodel fed by the point clouds is further established based on the PointNet architecture. This network can learn an end-to-end mapping between spatial positions of the object surface and CFD numerical quantities. With the proposed aerodynamic metamodel approach, the point clouds are constructed by collecting the coordinates of grid vertices on the object surface in a CFD domain, which can maintain the boundary smoothness and allow the network to detect small changes between geometries. Moreover, the point clouds are easily accessible from 3D sensors. The point cloud deep learning neural network, which employs re-sampling technique, the spatial transformer network and the fully connected layer, is developed to predict the aerodynamic characteristics of 3D geometry. The effectiveness of the proposed metamodel method is further verified by aerodynamic prediction and robust shape optimization of the ONERA M6 wing. The results show that the proposed method can achieve more satisfactory agreement with the experimental measurements compared to the parametric-learning-based deep neural network.     

Stretch: A Time-Transformation Technique

Stretch is a passive, linear, time-variant technique for performing temporal operations on many classes of signals. The technique employs three dispersive networks and a mixer. Signal slowdown, speedup, or time reversal can be attained by choice of network slopes. These temporal operations are performed within a signal "window," and the duration of the window is determined by the network time-bandwidth products. Both heuristic argumentation and rigorous analysis are presented, as are the results of a simple laboratory experiment.     

Constructive neural-networks-based MEMS/GPS integration scheme

This article exploits the idea of developing an alternative data fusion scheme that integrates the outputs of low-cost micro-electro-mechanical systems (MEMS) inertial measurements units (IMUs) and receivers of the global positioning system (GPS). The proposed scheme is implemented using a constructive neural network (cascade-correlation network (CCNs)) to overcome the limitations of conventional techniques that are predominantly based on the Kalman filter (KF). The CNN applied in this research has the advantage of having a flexible topology if compared with the recently utilized multi-layer feed-forward neural networks (MFNNs) for inertial navigation system (INS)/GPS integration. The preliminary results presented in this article illustrate the effectiveness of proposed CCNs over both MFNN-based and Kalman filtering techniques for INS/GPS integration.     

基于对传过程神经网络的航空发动机转子仿真故障诊断研究

为更好地解决动态模式识别及故障诊断问题,将过程神经元引入传统对传神经网络的竞争层,提出一种对传过程神经网络模型。并将所提出的网络模型及其学习算法用于航空发动机转子仿真故障诊断中,并得到了满意的结果。     

优选小波神经网络在周跳探测与修复中的应用

北斗导航定位过程中,传统的周跳探测与修复方法缺乏检验环节,无法保证修复结果的可靠性。为此,提出了一种级联式小波变换结合NARX神经网络多步循环预测修复的方法处理周跳问题。该方法通过构造载波相位双差模型检验量,探测周跳发生历元,采用NARX神经网络预测方法修复周跳,利用优选小波基函数进行周跳修复效果检验。实验证明,相较于经验模态分解和变分模态分解等模态分解法,优选小波神经网络周跳探测与修复方法可用于小周跳探测并判断出周跳正负性;构造的NARX神经网络周跳修复模型,解决了普通神经网络模型和传统多项式拟合法容易造成的二次奇异值问题。相较于长短期记忆(LSTM)和门控循环单元(GRU)深度学习神经网络模型,周跳预测精度分别提高了45.2%和55.9%。     

一类非线性系统的反演变结构控制及应用

针对一类非线性系统 ,将其变换为易于设计变结构控制规律的规范型。利用小脑关节神经网络( CMAC)估计系统中的不确定函数 ,利用 CMAC神经网络和多面滑模技术估计出变换后系统的状态 ,最后利用变结构控制技术设计出控制器 ,特点是无需已知不确定性函数及其各阶导数的上界 ,与经典设计方法相比 ,所提出的方案允许非参数化不确定性。最后将此方法应用于空空弹控制系统的设计中 ,仿真结果表明了该方法的有效性     

卷积和离散过程神经网络及其在航空发动机排气温度预测中的应用

 针对航空发动机排气温度的变化过程受复杂非线性时变因素的影响而难以用精确数学模型描述的问题,提出了卷积和离散过程神经网络(CSDPNN)模型,并将其应用于航空发动机排气温度(EGT)预测。该模型以离散样本作为直接输入,采用卷积和算法实现对时间累积效应的处理。相较于以连续函数作为输入的过程神经网络(PNN),不需要拟合离散样本得到连续函数后进行正交基展开,减少了精度损失,具有更高的预测精度。给出了卷积和离散过程神经网络模型的学习算法,并通过对Mackey-Glass混沌时间序列的预测对提出的方法进行应用说明和验证。通过航空发动机EGT预测实例,并与卷积和离散过程神经网络模型的连续函数输入过程神经网格以及传统人工神经网络(ANN)的预测结果进行了对比。结果表明,相较于连续函数输入过程神经网络以及传统人工神经网络,卷积和离散过程神经网络具有更高的预测精度,且对于EGT的预测具有较好的适应性,因而为航空发动机EGT预测提供了一种有效的方法。     

基于余弦基神经网络的FIR滤波器设计

提出了一种基于余弦基神经网络FIR滤波器的设计方法。根据线性相位FIR滤波器的幅频特性是有限项的傅里叶级数，构造了一个三层余弦基神经网络模型，并给出了最优隐层神经元的个数。经模拟仿真，滤波器性能非常理想，具有同时输出低通、高通、带通、带阻各种功能，通带、阻带无过冲无波动，边界频率可以精确控制。     

动态神经网络在液体火箭发动机故障检测与分离中的应用

应用动态神经网络在线辨识方法，提出了一种液体火箭发动机故障实时检测与分离基本系统。检测逻辑通过度量包含发动机故障信息的辨识残差信号实现火箭发动机故障检测，故障分离通过分析辨识误差相关函数的不同空间特征来实现。仿真研究表明动态神经网络可成功地应用于泵压式液体火箭发动机故障检测与分离。     

前向神经网络快速学习算法在发动机模型辨识中的应用

介绍一种采用前向神经网络辨识发动机非线性模型的快速学习方法,与现有同类方法相比,该方法采用矩阵的QR分解求解线性方程组,解决了维数过高的矩阵直接求逆带来的解失真问题,并把这种方法应用于航空发动机的非线性模型辨识。算例仿真验证了算法的可行性。      

一种计算螺旋桨性能的高效升力线法

本文提供了一种能有效计算螺旋桨性能的方法。方法的特点在于:(1)边界条件中利用了中弧面上的局部斜率以代替平面边界条件,使本方法适用于具有较大弯度翼型的螺旋桨;(2)用近似解析式与数值积分结合起来计算螺旋形尾涡产生的诱导速度,在保证精度的条件下,使计算工作量降低到原有的 10 %;(3)利用翼型的升阻特性考虑翼型阻力及压缩性影响。利用本方法对某型飞机的螺旋桨进行了计算,并与实验数据作了比较,在进距比不太小的情况下,计算结果与实验数据符合很好。      

一种基于神经网络补偿动态逆误差的方法

讨论了一种基于神经网络自适应补偿动态逆误差的方法，并应用于超机动飞机控制器设计中，飞机的基本控制采用非线性动态逆方法进行设计，对于模型不准确导致的逆误差采用神经网络进行在线补偿，仿真结果表明，采用神经网络补偿误差，弥补了非线性动态逆要求精确数学模型的缺点，而且可以简化动态逆控制律的设计，改善整个控制系统的性能。     

Neural network-based fault diagnosis for spacecraft with single-gimbal control moment gyros

This paper proposes a neural network-based fault diagnosis scheme to address the problem of fault isolation and estimation for the Single-Gimbal Control Moment Gyroscopes(SGCMGs) of spacecraft in a periodic orbit. To this end, a disturbance observer based on neural network is developed for active anti-disturbance, so as to improve the accuracy of fault diagnosis.The periodic disturbance on orbit can be decoupled with fault by resorting to the fitting and memory ability of neural network. Subsequ...     

物理启发的深度学习模型及其在热端部件散热中的应用

热端部件散热是众多空天设备的关键技术。表面温度分布是散热设计中用到的重要信息,常规的解析建模手段和机器学习方法均无法有效地表达此类高维信息。近年来兴起的图像深度学习算法是解决表面温度信息预测的有效手段。然而,现有的基于大数据的深度学习方法往往对于物理数据和小样本数据不适用,体现为泛化精度差、数据兼容性差、可解释性差。因此,有必要结合传热的先验知识发展物理启发的新型深度学习算法,以增强高自由度、高复杂度散热对象上的设计能力。本文基于卷积算子和有限差分求解方式的类比关系,提出了一种物理启发式的循环卷积神经网络。以横向出流的冲击冷却为例,开展了变计算域大小、变工况、变尺寸的批量数值模拟,获取了冲击冷却关键特征的小样本图像数据。进一步通过神经网络的训练,构建了多参数、大范围内有较好拟合能力的温度、传热系数、压力代理模型。研究结果表明,本文提出的物理启发神经网络模型,对于计算域大小没有限制,可以统一表达不同空间范围内获取的物理数据的共性规律。模型的各类超参设定均具有明确的物理意义,且与经典的微分方程求解理论有一定的类比关系,增强了神经网络调参的方向性。通过传热物理规律与黑箱模型的融合,本文实现了小样本多参数物理数据的共性建模。该方法可以迅速重构热端部件的高维分布信息,可服务于热端部件的快速分析设计以及优化。     

基于面向对象方法的压气机性能计算

提出一套预测压气机未知特性的方法,并基于面向对象思想采用变比热容计算方法进行压气机性能计算的分析和编程.结合粒子群优化（PSO）的全局寻优能力和反向传播（BP）神经网络的局部寻优能力提出基于PSO的BP神经网络（PSO-BP神经网络）预测压气机特性,分析了其预测误差和拟合误差:拟合误差基本都小于0.5%,预测误差基本都小于0.8%.其拟合精度和预测精度满足要求.采用变比热容计算方法来计算压气机性能,并采用面向对象方法编写了压气机性能计算程序.对几个压气机变工况点进行验证,各输出参数的最大误差为1.12%.因此,特性预测方法和性能计算的数学模型适用于压气机性能计算,这套方法同样适用于燃气轮机性能计算.