基于深度神经网络的像素级别可见光图像配准

现有图像配准算法中,借助图像采集设备参数的方法存在硬件内参难以获得或精度不够的问题,采用匹配图像特征计算图像单应性的方法存在对场景深度信息利用不全的问题。针对这一现象,提出了结合可见光图像与其深度信息来生成更具有真实性的配准图像对数据,用以训练得到一个可以进行像素级别图像配准的深度神经网络PIR-Net。建立了一个大规模、多视角、超仿真的图像配准数据集:多视角配准(MVR)数据集,该数据集包含7 240对含有深度信息的待配准图像及其像素级别的坐标对准真值;基于编码器-解码器的深度神经网络结构,训练得到一个能以全分辨率形式对2幅输入图像之间的坐标变化矩阵进行重建的PIR-Net。通过实验验证了PIR-Net能够在未知相机内参的情况下实现不同视角的可见光图像配准,并比传统算法具有更高的配准精度。在MVR数据集上,PIR-Net的配准误差仅为通用的特征匹配对准算法(SIFT+RANSAC)的18%,同时减少了30%的时间消耗。      

亚声速无人机S弯进气道的多点多目标优化设计

为提高亚声速无隔道式S弯进气道的整体气动性能,本文以提高总压恢复系数和降低畸变指数为设计目标,结合高精度数值模拟方法与第二代非劣排序遗传算法（NSGA-II）,开展了无隔道式S弯进气道在马赫数0.25和0.7时的多目标优化设计。整个优化流程基于400个样本,最终得到四幅有效Pareto前沿图。从总压畸变Pareto前沿图中选取出优化算例并与原始进气道进行对比,结果表明：优化后的进气道中心线斜率入口段小、出口段大,而横截面面积分布的曲线斜率恰好相反;优化后的进气道低压区缩小、流动分离得到有效的控制;虽然总压恢复系数提高有限,但是总压畸变得到大幅降低,在马赫数为0.25和0.7时,分别降低15.86%和23.61%。优化后的进气道在马赫数0.25~0.7范围内的整体性能得到有效改善。本文把优化设计方法进一步推广应用于3个马赫数下的多点多目标优化设计,并得到了三维Pareto前沿图。     

不同类型搭接网格对周向槽处理机匣模拟结果影响研究

摘 要: 为深入研究不同类型搭接网格对周向槽处理机匣数值模拟结果的影响,以某单级跨声速压气机及其转子叶顶的周向槽处理机匣作为研究对象,划分了多种计算网格并采用不同类型搭接网格关联周向槽与压气机主流道。详细的三维定常数值模拟和分析表明,传统完全非匹配搭接网格,在计算上会导致流场数据通过搭接网格传递时明显失真,使转子尖部复杂流动呈现显著间断特征,给压气机内流和性能模拟结果带来较大不确定性;增强搭接网格匹配度,有利于提高流场数据传递精度,弱化转子尖部复杂流动的不连续性,降低模拟结果的不确定性;采用完全匹配类型搭接网格,可在计算上完全避免搭界网格上的流场数据传递失真,从而使压气机内流及性能模拟结果具有更高的可信度。     

针对网络时延的发动机PI 控制系统稳定性分析和控制器设计

针对航空发动机分布式控制系统中的通信时延问题,进行了PI控制系统稳定性及控制设计。基于Nyquist稳定性判据 分析比例系数Kp的稳定域,进一步采用等效开环传递函数分析积分增益Ki的稳定域,获得时延上界依赖型PI控制器参数稳定域。 基于参数稳定域设计某涡扇发动机分布式控制系统的时延PI控制器,并进行了数字仿真。结果表明：在时延上界为0.3 s时,基于 参数稳定域设计的PI控制器可保证被控系统的稳定性,系统无稳态误差,转速控制系统超调量不超过0.38%;反之,基于控制器参 数非稳定域设计的控制器使被控系统的性能显著恶化。     

基于八叉树的简化构型三角片等值面削减算法

采用简化构型的SMC算法相比标准MC算法能够有效减少构成等值面的三角片的数量,但因为其仅是在体元内部的简化,所以不能较好地利用数据集表面局部形态特征。针对这一问题进一步提出OSMC算法。其根据简化构型的特点,首先采用八叉树结构组织体元,然后采用自底向上的合并策略合并节点,最后实现局部区域三角片合并。实验证明:OSMC算法能够实现比SMC算法更多的三角片削减,尤其对于具有较多平坦区域的数据集效果显著,其对公开数据集数据的平均削减率为55.1%,而SMC算法为29.7%,在面对高分辨率的地质数据时其最高削减率达到了80%,平均也超过了50%,同时OSMC算法能够更好地适应数据集分辨率的增长。      

计及储能寿命的风储电站分层优化控制策略

为了有效地平抑风电场的功率波动,提高风电场的并网特性,增强风电场并网运行的经济性和可靠性,提出了一种计及储能寿命的风储电站分层优化控制策略。首先,从经济学角度分析储能系统的充放电特性,考虑不同控制方式对储能系统寿命的影响;其次,采用分层-分区控制方式对风储电站进行协调控制,上层根据电网调度需求,考虑分时电价政策对集中式储能和整个风电场级进行协调功率控制,下层考虑区域内分布式储能和多台风电机组间的协调功率控制;最后,采用改进遗传算法对分层优化控制模型求解。结果表明:所提出的基于储能寿命损耗的风储电站优化控制方法实现了风储电站中储能系统及风机出力的合理分配,在延长储能寿命的同时提高了电网对风电的消纳能力,使得整个风电场运行经济性显著提高。     

Effects of the radial blade loading distribution and B parameter on the type of flow instability in a low-speed axial compressor

Previous studies showed that an axisymmetric hub-initiated disturbance defined as partial surge may initiate the stall of a transonic compressor; to reveal the instability evolution under full-span incompressible flow for different levels of hub loading and B parameter, an experimental investigation is conducted on a single-stage low-speed compressor. Experimental results show that under a uniform inflow condition without inlet flow distortion, a modal-type stall inception dominates in this low-speed compressor. When an inlet screen introducing hub distortion is used to increase the hub loading, a compressor stall is initiated by a modal wave, but large disturbances are present in the hub region before the compressor stall, which become stronger as the hub loading increases. Under high hub loading and large B parameter (implemented by adding hub distortion through an inlet screen and enlarging the outlet plenum volume, respectively), a compressor stall is triggered by an axisymmetric hub-initiated disturbance, which is much different from the modal-like disturbances. The beginning of this axisymmetric disturbance may be captured over 800 rotor revolutions prior to the onset of stall, and the amplitude grows with time. The disturbance is hub-initiated because the disturbance signal at the hub is detected much earlier than that at the tip; meanwhile, the frequency of this axisymmetric disturbance changes with the length of the inlet duct. The characteristics of instability evolution in the low-speed compressor are also compared with those in a transonic compressor.     

涡轴发动机部件特性修正及更新方法

针对由航空发动机零部件制造、装配及性能退化引起的发动机模型与实际发动机之间的性能不匹配问题,提出1种基于粒子群优化算法(PSO)的发动机部件特性自动修正及更新方法。根据发动机部件级模型的输出数据和发动机性能分析软件GasTurb计算结果,以发动机关键测量参数所定义的目标函数最小为优化目标,利用PSO获取不同相对换算转速下的部件特性修正因子,并在线完成特性图的自动更新。并以某型涡轴发动机为对象进行仿真验证,结果表明:该方法可有效提高涡轴发动机部件级模型的精度,并直接输出更新后的部件特性。     

Materials Experiment on Tiangong-2 Space Laboratory

During the China's Tiangong-2 (TG-2) flight mission, the experiments of 18 kinds of material samples were conducted in space by using a Multiple Materials Processing Furnace (MMPF) mounted in the orbital module of the TG-2 space laboratory. After the experiments of 12 kinds of samples of the first and second batches were completed successfully, astronauts packed and brought them back to the ground by ShenzhouⅡ spacecraft. By studying processing and formation on semiconductor and optoelectronics materials, metal alloys and metastable materials, functional single-crystal, micro-and nano-composite materials encapsulated in sample ampoules both in space and on Earth, we expect to explore some physical and chemical processes and mechanism of the materials formation that are normally obscured and therefore are difficult to study quantitatively on the ground due to the gravity-induced convection, to obtain the processing and synthesis technology for preparing high quality materials, and lead to the improvement and development of materials processing techniques on Earth, and also develop the experiment device and comprehensive ability for materials experiment in microgravity environment. This report briefly introduces the main points of each research work and preliminary comparative analysis results of 12 samples carried out by scientists undertaking research task.      

关于甚低频无线电波在电离层中衰减的计算

对甚低频无线电波在电离层中由碰撞引起的衰减进行了分析计算．利用二维傅里叶变换的方法,导出了衰减量随电子的碰撞频率、波的工作频率和地磁场的方向变化的解析表达式．给出了不同地磁场方向下衰减量随工作频率变化的数值结果．