月面环境三维激光SLAM技术

同步建图与定位（SLAM）可实现月球车在未知复杂月面环境下的定位与导航,月球表面由陨坑、石头等起伏地形构成,缺乏树木、建筑物等地面存有的显著特征,大量特征不显著的点云数据会对月球车定位精度和实时性造成影响。本文提出了一种针对月面环境的显著特征点云提取方法以及基于曲面定位能力估计的增量式优化算法,通过Fisher信息矩阵计算曲面定位能力指标,获取机器人位姿估计的不确定性测量,利用增量式的SLAM方案进行优化,用于提高定位精度与实时性。通过在Gazebo （物理仿真平台）仿真场景下的测试,验证了算法性能。     

Fault diagnosis and isolation of the component and sensor for aircraft engine

Aircraft engine component and sensor fault detection and isolation approach was proposed,which included fault type detection module and component-sensor simultaneous fault isolation module. The approach can not only distinguish among sensor fault, component fault and component-sensor simultaneous fault, but also isolate and locate sensor fault and the type of engine component fault when the engine component fault and the sensor faults occur simultaneously. The double-threshold mechanism has been proposed, in which the fault diagnostic threshold changed with the sensor type and the engine condition, and it greatly improved the accuracy and robustness of sensor fault diagnosis system. Simulation results show that the approach proposed can diagnose and isolate the sensor and engine component fault with improved accuracy. It effectively improves the fault diagnosis ability of aircraft engine.      

基于模糊奇偶方程的PL/INS导航系统故障诊断

提出了一种伪卫星(PL, Pseudolite)/惯性导航(INS, Inertial Navigation System)组合导航系统多故障识别方法.分析了PL的常见故障及其数学模型,采用模糊奇偶方程方法对PL的常见故障进行诊断、隔离和参数识别.采用Takagi-Sugeno(T-S)模糊模型描述PL/INS组合导航系统,对于局部线性化模型建立全解耦奇偶方程,并对奇偶方程的残差进行融合,得到全局残差.结合残差利用卡尔曼滤波进行故障参数识别,给出了参数识别的约束条件.仿真结果表明:针对导航系统中多个PL信号同时发生多种故障的情况,此方法能有效检测故障,并能准确识别出故障模型参数.      

一种结合深度学习的运动去模糊视觉SLAM方法

针对高动态环境下视觉同步定位与地图构建（Simultaneous Localization And Mapping,SLAM）系统的可靠性受运动模糊的限制,研究了一种基于生成对抗网络(Generative Adversarial Network,GAN)和AKAZE特征点的运动去模糊视觉SLAM方案。首先对因相机快速运动而产生的模糊图像进行AKAZE特征点的提取与检测,并根据特征点分布的丰富程度计算图像块权重,结合灰度图像的方差信息建立特征点与模糊程度之间的量化关系表;之后将达到模糊分数阈值的图像同步输入至改进GAN网络模型,该网络以端对端的形式恢复中心模糊帧的纹理信息,最后将输出的清晰图像重新进行位姿估计参与ORB-SLAM2后端优化过程。在公开数据集TUM上进行测试,对于受模糊影响较严重的数据集,方案可以明显降低相机轨迹估计的整体误差,同时维持较好的鲁棒性。     

导弹从发射箱热发射过程的数值分析

采用三维可压缩雷诺平均Navier-Stokes方程和k-ε湍流模型，研究了导弹出箱过程中，弹体姿态偏转后弹底周围环境及发射箱内的流场特性。首先，以超声速欠膨胀射流撞击平板实验为算例，对数值方法的有效性进行验证。其次，分别对导弹在约束期和半约束期箱内流动特性展开研究。研究表明：喷管尾流在发射箱内会形成强烈的引射效应。同时，在弹底会出现明显的回流区域，使导弹出箱时受到额外的阻力。考虑偏转后，发射装置受到的冲击载荷增大2倍以上，发射箱壁面受到的压力增加40%。而且，在半约束期，箱内的流场分布不再对称，会使得导弹受到额外的不平衡力矩。     

Adaptive sliding mode fault-tolerant control for satellite attitude tracking system

In this paper, an adaptive modified sliding mode control approach is developed for attitude tracking of a nano-satellite with three magnetorquers and one reaction wheel. A sliding variable is chosen based on finite-time convergence of the nano-satellite attitude tracking error and avoiding the singularity of the control signal. The control gain of the proposed method is developed adaptively to reduce the tracking error and improve the closed-loop control performance. The sliding variable and adaptive parameter are also employed in the reaching phase of the control law to decrease the chattering phenomenon. In addition, the finite-time convergence of attitude variables in the presence of actuator faults, inertia uncertainty, and external disturbances is proved using the extended Lyapunov theorem. The simulations are conducted to evaluate the performance of the proposed method according to different evaluation criteria. Monte Carlo simulations are also used to survey the reliability of the system in the presence of the mentioned condition.     

顾及动态物体感知的增强型视觉SLAM系统

传统的同步定位与制图（Simultaneous localization and mapping, SLAM）系统在复杂环境下工作时，无法分辨环境中的物体是否存在运动状态，图像中运动的物体可能导致特征关联错误，引起定位的不准确和地图构建的偏差。为了提高SLAM系统在动态环境下的鲁棒性和可靠性，本文提出了一种顾及动态物体感知的增强型视觉SLAM系统。首先，使用深度学习网络对每一帧图像的动态物体进行初始检测，然后使用多视图几何方法更加精细地判断目标检测无法确定的动态物体区域。通过剔除属于动态物体上的特征跟踪点，提高系统的鲁棒性。本文方法在公共数据集TUM和KITTI上进行了测试，结果表明在动态场景中定位结果的准确度有了明显提升，尤其在高动态序列中相对于原始算法的精度提升在92%以上。与其他顾及动态场景的SLAM系统相比，本文方法在保持精度优势的同时，提高了运行结果的稳定性和时间效率。