Neural network control of a gas turbine

In this paper we explore the practical use of neural networks for controlling complex non-linear systems. The system used to demonstrate this approach is a simulation of a gas turbine engine typical of those used to power commercial aircraft. The novelty of the work lies in the requirement for multiple controllers which are used to maintain system variables in safe operating regions as well as governing the engine thrust.     

Aeromagnetic compensation using neural networks

Airborne magnetic surveys in geophysical exploration can be subject to interference effects from the aircraft. Principal sources are the permanent magnetism of various parts of the aircraft, induction effects created by the earth's magnetic field and eddy-current fields produced by the aircraft's manoeuvres. Neural networks can model these effects as functions of roll, pitch, heading and their time derivatives, together with vertical acceleration, charging currents to the generator, etc., without assuming an explicit physical model. Separation of interference effects from background regional and diurnal fields can also be achieved in a satisfactory way.     

遥感图像飞机目标分类的卷积神经网络方法

目的 遥感图像飞机目标分类,利用可见光遥感图像对飞机类型进行有效区分,对提供军事作战信息有重要意义。针对该问题,目前存在一些传统机器学习方法,但这些方法需人工提取特征,且难以适应真实遥感图像的复杂背景。近年来,深度卷积神经网络方法兴起,网络能自动学习图像特征且泛化能力强,在计算机视觉各领域应用广泛。但深度卷积神经网络在遥感图像飞机分类问题上应用少见。本文旨在将深度卷积神经网络应用于遥感图像飞机目标分类问题。方法 在缺乏公开数据集的情况下,收集了真实可见光遥感图像中的8种飞机数据,按大致4∶1的比例分为训练集和测试集,并对训练集进行合理扩充。然后针对遥感图像与飞机分类的特殊性,结合深度学习卷积神经网络相关理论,有的放矢地设计了一个5层卷积神经网络。结果 首先,在逐步扩充的训练集上分别训练该卷积神经网络,并分别用同一测试集进行测试,实验表明训练集扩充有利于网络训练,测试准确率从72.4%提升至97.2%。在扩充后训练集上,分别对经典传统机器学习方法、经典卷积神经网络LeNet-5和本文设计的卷积神经网络进行训练,并在同一测试集上测试,实验表明该卷积神经网络的分类准确率高于其他两种方法,最终能在测试集上达到97.2%的准确率,其余两者准确率分别为82.3%、88.7%。结论 在少见使用深度卷积神经网络的遥感图像飞机目标分类问题上,本文设计了一个5层卷积神经网络加以应用。实验结果表明,该网络能适应图像场景,自动学习特征,分类效果良好。     

Control surface failure detection and accommodation using neuro-controllers

The ability of neural networks to learn from repeated exposure to system characteristics has made them a popular choice for many applications in linear and non-linear control. In this paper, the capabilities of neural networks in detecting and accommodating control surface failures for a modified F/A-18 super-manoeuverable fighter aircraft are examined. To detect and accommodate a failure in the thrust vectoring vane during a pitch manoeuvre, a hierarchical neuro-controller is designed using thrust vectoring, symmetric leading edge flap and the throttle. This neuro- controller is then used as the fault accommodating neuro- controller. A separate neural network is trained to detect failures in the thrust vectoring vane. The performance of the controller and fault-detection networks are verified using a numerical simulation of a longitudinal model of the aircraft.     

基于新型深度神经网络的民机表面缺陷识别

为解决机务人员依靠经验来对民航飞机的表面缺陷进行识别时易发生误判的问题,开发了一种用于民机表面的缺陷识别的结合Inception-net和残差模块的新型深度神经网络。首先,通过对各机场的在修飞机表面缺陷进行采样建立数据集,手段包括使用图像处理修复不合格图像、使用数据增强缓解数据类别不平衡、使用立方卷积插值法降采样保留图像特征等图像预处理操作。然后在自建的数据集上对新型深度神经网络与其他神经网络进行对比测试。实验结果表明,新型神经网络在较少的参数下能够达到最深的网络深度,且在自建数据集的测试集上的识别率和查全率分别为74.23%和62.29%,优于进行对比的其他网络。说明在一定程度上该网络能够有效用于民机表面缺陷识别工作中。     

通用飞机验证试飞智能辅助测试系统需求研究

根据当前通用飞机验证试飞测试的实际需要,针对通用飞机体积小、重量轻、试飞周期短等特点,基于人工智能技术提出了一套适用于通用飞机验证试飞的智能辅助测试系统需求,为通用飞机验证试飞智能辅助测试系统构架研究提供支撑。智能辅助测试系统需求包括从采集记录、座舱影像记录等环节提取的机载试验设备需求和从数据传输、数据存储、数据处理等环节提取的地面智能辅助系统需求两大部分,依据该需求构建的系统可实现对测试数据进行实时综合分析评估并反馈,辅助试飞工程师和试飞员全面掌握验证飞行情况,及时调整飞行状态、操作程序,从而为试飞适航验证提供技术支持,并实现提高试飞效率、缩短CCAR23部类通用飞机型号科研试飞与适航验证试飞周期、降低试飞成本的目的。     

Application of wireless sensor networks to aircraft control and health management systems

Use of fly-by-wire technology for aircraft flight controls have resulted in an improved performance and reliability along with achieving reduction in control system weight. Implementation of full authority digital engine control has also resulted in more intelligent, reliable, light-weight aircraft engine control systems. Greater reduction in weight can be achieved by replacing the wire harness with a wireless communication network. The first step towards fly-by-wireless control systems is likely to be the ...     

Aircraft class identification based on take-off noise signal segmentation in time

Aircraft noise is one of the most uncomfortable kinds of sounds. That is why many organizations have addressed this problem through noise contours around airports, for which they use the aircraft type as the key element. This paper presents a new computational model to identify the aircraft class with a better performance, because it introduces the take-off noise signal segmentation in time. A method for signal segmentation into four segments was created. The aircraft noise patterns are extracted using an LPC (Linear Predictive Coding) based technique and the classification is made combining the output of four parallel MLP (Multilayer Perceptron) neural networks, one for each segment. The individual accuracy of each network was improved using a wrapper feature selection method, increasing the model effectiveness with a lower computational cost. The aircraft are grouped into classes depending on the installed engine type. The model works with 13 aircraft categories with an identification level above 85% in real environments.     

Real-time prediction of unsteady aerodynamics: Application foraircraft control and manoeuvrability enhancement

The capability to control unsteady separated flow fields could dramatically enhance aircraft agility. To enable control, however, real-time prediction of these flow fields over a broad parameter range must be realized. The present work describes real-time predictions of three-dimensional unsteady separated flow fields and aerodynamic coefficients using neural networks. Unsteady surface-pressure readings were obtained from an airfoil pitched at a constant rate through the static stall angle. All data sets were comprised of 15 simultaneously acquired pressure records and one pitch angle record. Five such records and the associated pitch angle histories were used to train the neural network using a time-series algorithm. Post-training, the input to the network was the pitch angle (alpha), the angular velocity (dalpha/dt), and the initial 15 recorded surface pressures at time (t (0)). Subsequently, the time (t+Deltat) network predictions, for each of the surface pressures, were fed back as the input to the network throughout the pitch history. The results indicated that the neural network accurately predicted the unsteady separated flow fields as well as the aerodynamic coefficients to within 5% of the experimental data. Consistent results were obtained both for the training set as well as for generalization to both other constant pitch rates and to sinusoidal pitch motions. The results clearly indicated that the neural-network model could predict the unsteady surface-pressure distributions and aerodynamic coefficients based solely on angle of attack information. The capability for real-time prediction of both unsteady separated flow fields and aerodynamic coefficients across a wide range of parameters in turn provides a critical step towards the development of control systems targeted at exploiting unsteady aerodynamics for aircraft manoeuvrability enhancement.     

Aircraft Classification and Acoustic Impact Estimation Based on Real-Time Take-off Noise Measurements

The acoustic impact of aircraft taking-off is an important subject for monitoring and research. It is very useful to analyze the type or class of aircraft that produces high level noises based on take-off characteristics. This paper presents a new method about aircraft classification and the acoustic impact estimation, in areas near an airport, based on real time noise measurement for each take-off. The noise measurements are made with sampling frequency of 50 ks/s (kilo samples per second) and 24-bit resolution analog-to-digital conversion, during 24 s. The aircraft identification is made through a model of two parallel feed-forward neural network combined with a weighted addition. In order to generate the inputs to the neural networks, the noise signal features were obtained from the auto-regressive model and the 1/12 octave analysis. The aircraft is grouped into categories or classes depending on the installed engine type. This system has 13 aircraft categories and an identification level above 80% in real environments. Noise signals, generated during aircraft take-off are measured in a fixed location on the airport runway end using a linear 4-microphone array. The acoustic impact is presented by means of a noise map for each take-off and displays four layers related to four take-off time intervals. Based on International Organization for Standardization, each time interval is represented by an equivalent point sound source location through the estimation of time-difference-of-arrival of the acoustic wave from aircraft taking-off.     

A MODEL FOR DIGITAL SIMULATION OF TWO-OPERATOR MAN-MACHINE SYSTEMS

A model, which had a9 its aim tho simulation of one or two oporator man-machino Bystems, is presented and discussed. Tho purpose of the tochnique is to allow prediction of Bystom offoctivonoss during the early design stage and to enable the comparative evaluation of alternative system dosigna. The modol is based on the use of a digital computer which sequentially simulates operator performance of oach subtask in a total task. As a result of calculations, output records are obtained for such items aa task success or failure, operator stress conditions, idle time, team cohesivonoss, and, in the ovent of sueeossful task completion, time available but unspent.

The model was applied to tho simulation of two tasks. Tho first is tho in-flight refuolling of an F8U receiver aircraft by an A4D tanker aircraft; the second is an attack manoouvre by an advanced supersonic aircraft.

Tho rosults from the model as reflected through tho digital simulation were compared with empirical critorion data obtained for both tasks and were further evaluated on the basis of their compatibility with logical expectation. Tho rosults from these initial applications of the model, which are summarized and evaluated, appear to conform with actual observations and are generally reasonable     

变体飞机建模及自适应动态面控制

为了描述变体飞机在变体过程中的动态特性,将变体飞机看成一个质点系,对质点系进行受力分析,得到变体飞机的力方程和力矩方程,并推导出一种变后掠翼飞机的纵向运动方程.为了确保变体飞机在变体过程中的飞行稳定性,提出一种基于动态面的自适应backstepping控制器,通过引入一阶滤波器,避免了传统backstepping设计中的计算膨胀问题.采用径向基(RBF,radial basis function)网络逼近系统不确定项,基于Lyapunov稳定性理论,给出自适应输出反馈控制器和RBF网络权值向量的自适应律,并证明了闭环系统是半全局一致有界.仿真结果表明实际系统的输出响应较好地跟随参考指令,不受变体速率的影响,满足鲁棒性要求.     

Robust identification method for nonlinear model structures and its application to high-performance aircraft

A common assumption is that the model structure is known for modelling high performance aircraft. In practice, this is not the case. Actually, structure identification plays the most important role in the processing of nonlinear system modelling. The integration of mode structure identification and parameter estimation is an efficient method to construct the model for high performance aircraft, which is nonlinear and also contains uncertainties. This article presents an efficient method for identifying nonlinear model structure and estimating parameters for high-performance aircraft model, which contains uncertainties. The parameters associated with nonlinear terms are considered one after the other if they should be included in the nonlinear model until a stopping criterion is met, which is based on Akaike's information criterion. A numerically efficient U-D factorisation is presented to avoid complex computation of high-order matrices. The proposed method is applied to flight test data of a high-performance aircraft. The results demonstrate that the proposed method could obtain the good aircraft model with a reasonably good fidelity based on the comparison with flight test data.     

Nonlinear Fly‐By‐Throttle H∞ Control Using Neural Networks

A neural network approach to gain scheduling H∞ controllers for propulsion controlled aircraft (PCA) systems is introduced. The PCA system is applied to backup control of aircraft experiencing control surface failure. The H∞ technology is applied to the problem of matching the crippled aircraft and the nominal model. Various H∞ controllers at various flight conditions are used to train radial basis function networks (RBFN), which can then be used as the nonlinear controller. Simulation on an L‐1011 under fly‐by‐throttle control demonstrates that the RBFN controller can stabilize the crippled airplane to obtain the desired model and possesses robustness against the engine delay.     

Classification System for Serial Criminal Patterns

The data mining field in computer science specializes in extracting implicit information that is distributed across the stored data records and/or exists as associations among groups of records. Criminal databases contain information on the crimes themselves, the offenders, the victims as well as the vehicles that were involved in the crime. Among these records lie groups of crimes that can be attributed to serial criminals who are responsible for multiple criminal offenses and usually exhibit patterns in their operations, by specializing in a particular crime category (i.e., rape, murder, robbery, etc.), and applying a specific method for implementing their crimes. Discovering serial criminal patterns in crime databases is, in general, a clustering activity in the area of data mining that is concerned with detecting trends in the data by classifying and grouping similar records. In this paper, we report on the different statistical and neural network approaches to the clustering problem in data mining in general, and as it applies to our crime domain in particular. We discuss our approach of using a cascaded network of Kohonen neural networks followed by heuristic processing of the networks outputs that best simulated the experts in the field. We address the issues in this project and the reasoning behind this approach, including: the choice of neural networks, in general, over statistical algorithms as the main tool, and the use of Kohonen networks in particular, the choice for the cascaded approach instead of the direct approach, and the choice of a heuristics subsystem as a back-end subsystem to the neural networks. We also report on the advantages of this approach over both the traditional approach of using a single neural network to accommodate all the attributes, and that of applying a single clustering algorithm on all the data attributes.     

Optimal transition maneuvers for a class of V/STOL aircraft

This paper focuses on the problem of computing optimal transition maneuvers for a particular class of tail-sitter aircraft able to switch their flight configuration from hover to level flight and vice versa. Both minimum-time and minimum-energy optimal transition problems are formulated and solved numerically in order to compute reference maneuvers to be employed by the onboard flight control system to change the current flight condition. In order to guide the numerical computation and to validate its results, in a first stage approximated solutions are obtained as a combination of a finite number of motion primitives corresponding to analytical trajectories of approximated dynamic models. The approximated solution is then employed to generate an initial guess for the numerical computation applied to a more accurate dynamic model. Numerical trajectories computed for a small scale prototype of tail-sitter aircraft are finally presented, showing the effectiveness of the proposed methodology to deal with the complex dynamics governing this kind of systems.     

Maximum-miss aircraft collision avoidance

Aircraft densities in terminal areas increase each year, and the risk of collision grows proportionally. The maintenance of clearance between aircraft in this environment sometimes calls for evasive maneuvers, which depend on the relative position and relative velocity of two aircraft. In this study, small-amplitude maneuvers are found for either or both aircraft in near-miss configurations. Using practical low-order dynamics, individual maneuvers are found that maximize the miss distance. These optimal maneuvers combine longitudinal (speed) and normal (lift) accelerations. The signs of the accelerations of both aircraft depend on their magnitudes. An evasive climb maneuver, for example, becomes a dive maneuver if the acceleration amplitude exceeds a certain value. The maximum-miss maneuvers appear to have practical potential, because they can be determined on-line from estimated position data for both aircraft, without consideration of detailed inertial and aerodynamic properties of the aircraft. Recommended by H. Stalford     

基于密集卷积神经网络的遥感飞机识别

传统的飞机识别方法受模糊、遮挡、噪声以及光照等多种因素的干扰时会降低识别率,且卷积神经网络主要依赖局部特征,却丢失了轮廓特征等重要的全局结构化特征,从而导致算法对于受干扰飞机图像识别效果不佳。因此,基于密集卷积神经网络提出一种结合局部与全局特征的联合监督识别方法,以密集卷积神经网络为基础得到图像特征,通过结合局部特征（卷积神经网络特征）与全局特征（方向梯度直方图特征）进行分类,分类器目标函数使用softmax损失和中心损失联合监督方法。实验结果表明,局部特征与全局特征的结合使算法更加智能化,且损失函数联合监督方法能够实现图像深层特征的类内聚合、类间分散,该算法能有效解决卷积神经网络对受到多种干扰的遥感图像识别率低的问题。     

Cyber security methods for aerial vehicle networks: taxonomy,challenges and solution

Aerial vehicle networks (AVNs) compose a large number of heterogeneous aerial nodes, such as unmanned aerial vehicles, aircrafts and helicopters. The main characteristics of these networks are the high mobility of aerial nodes and the dynamic network topology. AVNs represent attractive targets for attackers due to the fact that aerial nodes could be connected to an untrusted network and hence lead the attackers to launch lethal threats, e.g., aircraft crash. Therefore, the security of AVNs is mandatory. In this article, we examine the challenges of cyber detection methods to secure AVNs and review exiting security schemes proposed in the current literature. Furthermore, we propose a security framework to protect an aircraft (SFA) against malicious behaviors that target aircrafts communication systems. Numerical results show that SFA achieves a high accuracy detection and prediction rates as compared to the current intrusion detection for aircrafts communication system.     

基于节点向量表达的复杂网络社团划分算法

社团结构划分对复杂网络研究在理论和实践上都非常重要.借鉴分布式词向量理论,提出一种基于节点向量表达的复杂网络社团划分方法（CDNEV）.为了构建网络节点的分布式向量,提出启发式随机游走模型.利用节点启发式随机游走得到的节点序列作为上下文,采用SkipGram模型学习节点的分布式向量.选择局部度中心节点作为K-Means算法的聚类中心点,然后用K-Means算法进行聚类,最终得到社团结构.在真实和模拟两种网络上做了丰富的实验,与主流的全局社团划分算法和局部社团划分算法作了比较.在真实网络上CDNEV算法的F1指标比其他算法平均提高19%;在模拟网络上,F1指标则可以提高15%.实验结果表明,相对其他算法,CDNEV算法的精度和效率都较高.