基于新型多分类支持向量算法的发动机故障诊断

层次支持向量机(H-SVM)比通常的"一对多" (1-V-R)和"一对一" (1-V-1)等多分类支持向量算法具有更快的训练速度和分类速度.提出一种基于H-SVM的航空发动机气路部件故障诊断方法,根据特征空间中各类故障样本中心之间的距离来逐层划分子类,距离较近的故障样本归为同一个子类进行训练,得到的H-SVM层次结构合理,各层的SVM分类间隔大、泛化能力强;同时,用ν-SVM代替通常的C-SVM作为两类分类器,分类器参数意义明确、变化范围小,更容易确定.仿真实验表明,基于H-SVM的故障分类器具有良好的分类准确性和泛化性能,能够对发动机气路部件的单一故障和复合故障进行快速诊断.     

基于大规模训练集SVM的发动机故障诊断

提出了一种新的学习策略,用于解决发动机故障诊断中大规模支持向量机(SVM)的训练问题.通过保留初始SVM分类器支持向量超平面附近的样本以及错分样本,使最终得到的约减集规模明显缩小,从而可在保持较高分类精度的前提下使训练时间明显缩短;同时,由于支持向量的数量减小,分类时间也相应缩短.探讨了序贯最小优化(SMO)算法的参数选择和实现过程中的关键问题,为这种极具潜力的算法在发动机故障诊断中的实际应用奠定了坚实的基础.仿真实例表明,这种基于大规模训练集SVM的发动机故障诊断方法有效、可靠,容易实现,可以作为工程应用的基础.      

基于RDK-ELM 的航空发动机控制系统故障诊断

为保持较高诊断正确率,缩短训练时间,满足航空发动机故障诊断对于实时性和高诊断率的需求,提出1种对深度核极限学习机的简约改进方法。输入数据中随机选取部分数据作为支持向量,结合深度学习网络的多层结构,完成了对输入样本的特征提取,通过核函数实现了高维空间映射分类。数字仿真表明:算法分类正确率高,训练时间短,可应用于航空发动机控制系统的故障诊断。     

基于S-PSO分类算法的故障诊断方法

将监控数据的已知状态作为先验类别标签,构造出新的有监督的粒子群优化(S-PSO)分类算法,并对设备进行故障诊断。为提高故障诊断的准确率,降低随机性对分类算法的影响,提出了新的基于动态邻域的自适应探测更新(ADU-DN)的干预更新策略来拓展粒子搜索整个解空间的能力,引导粒子自适应地跳出局部最优区域,确保获得全局最优解;同时设计出基于最小类内距离、最大类间距离和训练样本最大分类精度的适应度函数,使得输出的最优类别中心兼顾了这3个因素,增强了分类算法在故障诊断中的通用性和容错性,提高了测试样本的分类精度。S-PSO分类算法有效克服了聚类算法只考虑数据间相似性特征、不考虑数据蕴含的物理意义以及不能很好指导样本分类的缺陷。对GE90发动机孔探图像纹理特征分类进行了对比研究,研究数据表明:S-PSO分类算法表现出了较强的鲁棒性,在故障诊断中的分类精度高于支持向量机(SVM)和常用神经网络模型。     

基于支持向量机的组合分类方法及应用

为了解决采用神经网络、决策树作为弱分类器的AdaBoost组合分类存在的不足,进一步改善组合分类效果,提出采用支持向量机(SVM)作为弱分类器的一种新的组合分类诊断方法——AdaBoost-SVM。该方法没有采用一个固定的SVM的核参数,而是自适应调整SVM中的核参数,从而得到一组有效的SVM弱分类器。通过对基准数据库的测试及航空发动机故障样本的诊断,结果表明,所提AdaBoost-SVM方法较好地解决了现有的Ada-Boost组合分类方法中存在的弱分类器本身参数选取困难问题及训练轮数的合理选取问题,并具有更好的泛化性能,更适合对分散程度较大、聚类性较差的航空发动机故障样本进行分类。     

基于Boosting-SVM算法的航空发动机故障诊断

提出了一种利用支持向量机(SVM)作为弱基分类器、Boosting算法进行加权融合的航空发动机故障诊断算法.该算法具有支持向量机的强分类能力,又具有Boosting算法适合不均衡数据集的特点.为验证算法的有效性,采用外场实测的滑油光谱分析数据针对传动系统的轴承、减速齿轮和滑油系统3类故障进行了验证.为去除实测数据之间的冗余、降低特征维数,提高算法执行效率,采用主元分析(PCA)和粗糙集理论(RST)进行故障特征压缩和提取.利用实测数据构造了Boosting支持向量机分类器.最后,实验结果表明Boosting-SVM算法可以显著提高SVM分类器的推广性能.针对实测数据,3种故障平均识别准确率由79.4%提高到了85.7%.     

Diverse AdaBoost-SVM分类方法及其在航空发动机故障诊断中的应用

 提出采用考虑到精度/差异权衡的SVM作为弱分类器的一种新的组合分类诊断方法——Diverse AdaBoost-SVM。该方法通过在一组具有适当精度的弱分类器中进一步选择具有较大差异性的弱分类器,对这些具有较大差异性的弱分类器进行组合,从而较好解决AdaBoost算法中存在的精度/差异权衡的难题;同时该方法也较好地解决了现有的AdaBoost方法存在的弱分类器本身参数选取困难问题及训练轮数T的合理选取问题。通过对基准数据库的测试及航空发动机故障样本的诊断,结果表明和其他方法相比,Diverse AdaBoost -SVM方法具有更好的泛化性能,更适合对分散程度较大、聚类性较差的航空发动机故障样本进行分类,也更适合对非对称故障样本集进行分类。     

基于支持向量机的航空发动机整机振动故障诊断技术研究

支持向量机是采用结构风险最小化原则代替传统统计学中的基于大样本的经验风险最小化原则的一种新型机器学习方法,由于它出色的学习分类能力和推广能力,广泛地应用于模式识别和函数拟合中。针对某型航空发动机整机振动过大的现象,提出了一种基于支持向量机（SVM）的整机振动故障诊断方法。首先介绍了SVM理论,然后根据SVM学习方法的结构风险最小化原则,对某型航空发动机已知的整机振动故障模式数据进行了训练和预测,并建立了基于SVM的航空发动机整机振动故障诊断模型。最后通过对已有故障模式进行诊断预测,证明该方法在航空发动机整机振动故障诊断方面具有良好效果。     

一种基于 KFCM的 SVM遥感图像机场目标分类算法

在遥感图像机场目标分类方面,支持向量机(SVM)有着广泛的应用,但由于样本不平衡问题以及不确定性数据的存在,传统SVM算法的分类精度与效果还无法令人满意。为提高传统SVM分类器的性能,文章将建立在模糊理论基础上的模糊核C-均值聚类算法(KFCM)用于处理遥感数据的不确定性问题,并通过聚类分析后的目标子图,剔除非目标样本的同时保留了目标样本,较好地解决了样本不平衡问题。将基于KFCM的SVM分类算法用于遥感图像机场目标的分类,实验结果和性能分析表明该算法分类性能优于传统SVM算法。     

基于模糊信息粒化和优化SVM的航空发动机性能趋势预测

提出采用模糊信息粒化(FIG)和优化的支持向量机(SVM)来预测航空发动机参数的变化趋势和变化空间。利用模糊信息粒化方法对性能参数进行粒化处理。以K CV验证误差最小作为优化目标,采用遗传算法(GA)实现支持向量机惩罚参数和核函数参数的自适应优化选择;训练SVM模型并进行并对模糊粒子非线性预测。利用某航空公司的某型航空发动机性能参数监测数据进行验证,结果表明:该算法可以有效实现航空发动机性能参数变化趋势和变化空间预测。在实例基础上分析了窗口大小对算法预测精度的影响以及算法多步预测的效果,得出算法最佳窗口大小为3个数据且算法3步以内预测误差小于10%。      

The effect of inlet conditions on the flow and heat transfer in multiple rotating cavity with axial throughflow

This paper discusses experimental results from two different build configurations of a heated multiple rotating cavity test rig.Measurements of heat transfer from the discs and tangential velocities are presented.The test rig is a 70% full scale version of a high pressure compressor stack of an axial gas turbine engine.Of particular interest are the internal cylindrical cavities formed by adjacent discs and the interaction of these with a central axial throughflow of cooling air.Tests were carried out for a range of non-dimensional parameters representative of high pressure compressor internal air system flows(Re up to 5×106 and Rez up to 2×105).Two different builds have been tested.The most significant difference between these two build configurations is the size of the annular gap between the(non-rotating) drive shaft and the bores of the discs.The heat transfer data were obtained from thermocouple measurements of surface temperature and a conduction solution method.The velocity measurements were made using a two component,LDA system.The heat transfer results from the discs show differences between the two builds.This is attributed to the wider annular gap allowing more of the throughflow to penetrate into the cavity.There are also significant differences between the radial distributions of tangential velocity in the two builds of the test rig.For the narrow annular gap,there is an increase of non-dimensional tangential velocity V/Ωr with radial location to solid body rotation V/Ωr=1.For the wider annular gap,the non-dimensional velocities show a decrease with radial location to solid body rotation.      

Research on Auto-detection for Remainder Particles of Aerospace Relay Based on Wavelet Analysis

Aerospace relay is one kind of electronic components which is used widely in national defense system and aerospace system. The existence of remainder particles induces the reliability declining, which has become a severe problem in the development of aerospace relay. Traditional particle impact noise detection (PIND) method for remainder detection is ineffective for small particles, due to its low precision and involvement of subjective factors. An auto-detection method for PIND output signals is proposed in this paper, which is based on direct wavelet de-noising (DWD), cross-correlation analysis (CCA) and homo-filtering (HF), the method enhances the affectiv-ity of PIND test about the small particles. In the end, some practical PIND output signals are analysed, and the validity of this new method is proved.     

Preparation of Highly Dispersed Antimony-doped Tin Oxide Nano-powder via Ion-exchange Hydrolysis of SnCl4 and SbCl3 and Azeotropic Drying

Antimony-doped tin hydroxide colloid precipitates have been synthesized by hydrolysis of SnCl4 and SbCl3 using: (1) an ion-exchange hydrolysis to remove chlorine ions, and (2) isoamyl acetate as an azeotropic solvent to obviate water. The obtained dried powder is of high dispersivity without any need for further grinding. The size and dispersivity of the final particles are investigated with the aid of TG-DTA, BET, XRD and TEM. After having calcined, the antimony-doped tin oxide nanopowder possesses a tetragonal rutile structure with high dispersivity, uniform particles and low hard agglomeration.     

Effects of Mo on the Microstructure and Hydrogen Sorption Properties of Ti-Mo Getters

The effects of Mo on the microstructure evolution, porosity and hydrogen sorption properties of Ti-Mo getters are investigated in this work. The results show that the addition of Mo prolongs the densification process of Ti-Mo getters and results in a significant amount of sintered pores. With the Mo content increasing, the porosity of getters firstly increases reaching the maximum value as it at- tains about 7.5wt.%, and then drops. At the room temperature, the hydrogen sorption property of getters increases progressively with the Mo content increasing, but the tendency is not very clear before its content lies below 2.5wt.%. When the Mo content achieves about 7.5wt.%, the hydrogen sorption property proves to be the best. The discussion is made about the above mentioned phenomena inclusive of hydrogen sorption properties of getters under different activation conditions (from 500-750 ℃).     

超声速燃烧数值模拟中的湍流与化学反应相互作用模型

高精度数值模拟有助于理解超声速湍流燃烧中湍流与化学反应的相互作用,可为发动机燃烧室等工程应用设计提供可靠的预测模型。除直接数值模拟外,目前在湍流燃烧应用中使用的大涡模拟和雷诺平均Navier-Stokes模拟均需要借助模型模化发生在湍流小尺度上的流动与化学反应过程对湍流大尺度运动的影响。现有的湍流与化学反应相互作用模型大致可分为:火焰面类模型和概率密度函数类模型,2类模型在不同的应用中各自具有优势和局限性。此外,现有模型大都基于低马赫数燃烧,而超声速燃烧中通常会伴随快速混合、局部熄火和再着火以及激波等复杂过程,这为发展其中的湍流与化学反应相互作用模型提出了更多的挑战。     

适于低轨卫星IP网络的单核共享树组播算法(英文)

为了解决低轨卫星IP网络中现有典型源组播算法的信道资源浪费问题,本文提出了一套单核共享树组播算法,即核心群合并共享树(CCST)和加权核心群合并共享树(w-CCST)算法。CCST 算法包括动态近似中心(DAC)选核方法和核心群合并组播路径构建方法。DAC方法专为周期、规律运动的低轨卫星网络提出,不需要复杂的星上计算。在核心群合并方法中,以核节点作为初始核心群,通过核心群和剩余组成员的最短路径方法逐步扩展直至整棵组播树构建完成,从而使得组播树的树代价最小,大大提高了网络的带宽利用率和组播传输效率。w-CCST 算法中所提出的加权因子可以调整树代价和端到端传播时延之间的折衷程度,因此,可以通过调整加权因子来适度增大树代价、降低端到端传播时延以支持某些端到端时延要求苛刻的实时组播业务。最后,与低轨卫星 IP 网络中典型算法进行了性能比较,仿真结果说明,CCST 算法的平均树代价比其它算法显著降低,w-CCST 算法的平均端到端传播时延小于 CCST 算法。     

Jet Vectoring Control Using a Novel Synthetic Jet Actuator

A primary air jet vectoring control system with a novel synthetic jet actuator (SJA) is presented and simulated numerically. The results show that, in comparison with an existing traditional synthetic jet actuator, which is able to perform the duty of either “push” or “pull”, one novel synthetic jet actuator can fulfill both “push” and “pull” functions to vector the primary jet by shifting a slide block inside it. Therefore, because the new actuator possesses greater efficiency, it has potentiality to replace the existing one in various appli- cations, such as thrust vectoring and the reduction of thermal signature. Moreover, as the novel actuator can fulfill those functions that the existing one can not, it may well be expected to popularize it into more flow control systems.     

Nonlinear Adaptive Slewing Motion Control of Spacecraft Truss Driven by Synchronous V-gimbaled CMG Precession

The slewing motion control of a truss arm driven by a V-gimbaled control-moment-gyro (CMG) is a nonlinear control problem. The V-gimbaled CMG consists of a pair of gyros that must precess synchronously. The moment of inertia of the system, the angular momentum of the gyros and the external disturbances are not exactly known. With the help of feedback linearization and recursive Lyapunov design method, an adaptive nonlinear controller is designed to deal with the unknown items. Performance of the proposed controller is verified by simulation.     

An SOR Implicit Time-accurate Scheme in Calculating Unsteady Flows

A new time-accurate marching scheme for unsteady flow calculations is proposed in the present work. This method is the combination of classical Successive Over-Relaxation (SOR) iteration method and Jacobian matrix diagonally dominant splitting method of LUSGS. One advantage of this algorithm is the second-order accuracy because of no factorization error. Another advantage is the low computational cost because the Jacobian matrices and fluxes are only calculated once in each physical time step. And, the SOR algorithm has better convergence property than Gauss-Seidel. To investigate its accuracy and convergency, several unsteady flow computa- tional tests are carried out by using the proposed SOR algorithm. Roe’s FDS scheme is used to discritize the inviscid flux terms. Un- steady computational results of SOR are compared with the experiment results and those of Gauss-Seidel. Results reveal that the numerical results agree well with the experimental data and the second-order accuracy can be obtained as the Gauss-Seidel for unsteady flow computations. The impact of SOR factor is investigated for unsteady computations by using different SOR factors in this algorithm to simulate each computational test. Different numbers of inner iterations are needed to converge to the same criterion for different SOR factors and optimal choice of SOR factor can improve the computational efficiency greatly.