基于人工神经网络的预腐蚀铝合金疲劳性能预测

 通过对BP神经网络算法分析和收敛性改进,从获得的预腐蚀和疲劳试验数据中通过训练建立了LY1 2CZ铝合金腐蚀性能和疲劳特性与预腐蚀温度和时间的映射模型,从而可预测铝合金在一定预腐蚀环境谱下的最大腐蚀深度和疲劳特性。神经网络算法采用 BP算法 ,网络结构采用2-4-2形式。结果表明 ,神经网络用于预腐蚀铝合金的腐蚀状况和疲劳性能预测是可行的     

以科研为关键加快向研究型大学转变的步伐--庆祝南京航空航天大学建校50周年

首先回顾了建校50年来的科研工作,着重介绍了“七五”－“九五”期间我校承担国防型号任务,加强基础研究,跟踪高新技术等方面取得的成绩;总结了取得科研业绩的经验和体会;队伍是根本,基地是关键,团队是基础;并提出为实现向研究型大学转变的战略目标,必须进一步巩固完善科研管理体制和机制,加快向研究型大学转变的步伐。     

SOM人工神经网络在客机零部件故障诊断中的应用研究

现代客机在使用过程中不可避免地会发生各种故障，故障诊断对保证飞行安全十分重要，本文将基于概率的数学方法与专家经验相结合对其进行故障诊断。Kohonen的自组织特征映射（Self-organizing map,SOM）人工神经网络在输出上可反映出输入学习样本的概率密度分布，且无需知道样本的概率分布的先验知识，兼具函数逼近功能。本文将SOM引入这一领域，用于计算飞机零部件发生故障的概率，以及实现数学方法计算结果与专家经验的结合，实际应用说明了该方法的可行性。     

某型无人机拦阻网回收系统设计

介绍了无人机拦阻网回收系统的组成和工作原理，论述了系统的网体设计、涡轮阻尼器设计以及拦阻过程计算等关键技术问题，并给出某型无人机拦阻回收系统的性能计算结果。本文的研究成果为某型无人机拦阻网回收系统的工程设计提供了理论依据，并已成功应用于某型无人机系统的研制。     

用于建立机械制造系统模型的扩展着色Petri网

着色Petri网是描述和分析并行系统的有力工具,近年来在机械制造系统中得到了广泛的应用。由于机械制造系统的自身特点,在表示着色Petri网结构的关联矩阵中含有许多零元素,从而导致着色Petri网占据数据空间大的缺陷,难以对复杂机械制造系统的规划设计与生产调度进行分析研究。 本文基于着色Petri网的基本原理,提出一种适用于建立机械制造系统模型的扩展着色Petri网理论,用于解决着色Petri网占据数据空间大的问题,为Petri网理论在机械制造系统建模方面的广泛应用提供理论基础。最后给出实例,说明减少数据空间的结果。     

电子干扰对低可观测飞行器飞行路径规划的影响

为了提高巡航导弹的低空突防能力,在分析电子干扰对雷达网威胁区域的影响后认为:在多重干扰条件下,雷达探测空间会顺着干扰方向产生内凹和偏移,因此,建立了新的雷达探测空间模型.采用空间对象间拓扑关系推理方法建立了飞行器飞行路径规划目标函数中雷达威胁指数模型,然后改进了飞行器飞行路径规划模型,并在此基础上进行了巡航导弹飞行路径规划的软件仿真.仿真结果表明:考虑电子干扰对雷达威胁区域的影响情况下规划的飞行路径能够回避雷达威胁,有效提高飞行器低空突防能力.     

无人机4D制导系统研究

基于逆运动学方法设计了无人机四维制导系统。基本姿态控制器采用基于在线神经网络的非线性动态逆控制器。动态逆控制器用来对消无人机的非线性，在线神经网络补偿对消不精确引起的状态误差。制导系统的任务就是由给定的三维航迹，解算出航迹角，并生成三个姿态角指令，同时设计发动机控制回路以控制航迹速度，最终实现精确的四维航迹跟随。     

Online scheduling of distributed Earth observation satellite system under rigid communication constraints

This paper focused on online scheduling of distributed Earth observation satellite system in a dynamic environment. The objective was to maximize the total profit of the overall system by efficiently coordinating the different satellites with stochastic arrival of urgent tasks, subject to rigid communication and observation time window constraints. We formulated this problem as a single-task, single-robot, time-extended assignment problem with intra-schedule dependency using the multi-robot task assignment taxonomy and formulated the sub-problem after releasing urgent tasks in a mixed-integer linear programming model. We first described the online scheduling algorithm for a single satellite, then we proposed the modified consensus-based bundle algorithm (m-CBBA) and modified asynchronous consensus-based bundle algorithm (m-ACBBA) with synchronous and asynchronous communication, respectively. Compared with initial versions of CBBA and ACBBA, the modified versions added the communication loop prediction phases to efficiently utilize scarce communication opportunities and reduce the communication requirements. Additionally, we introduce two contract net protocol (CNP)-based algorithms for comparison, respectively SingleItem-CNP-based (SI-CNP) algorithm and Batch-CNP-based (BA-CNP) algorithm. Computational experiments indicated that both the total profit and percentage of scheduled urgent tasks achieved by the m-ACBBA and m-CBBA algorithms were much higher than those achieved by both SI-CNP and BA-CNP. Additionally, the number of communications needed by either m-ACBBA or m-CBBA algorithm was lower than that by SI-CNP. When the communication cost in the system is high, the m-CBBA algorithm is preferred because it balances the profit and the required number of communications. When the communication cost is low, the m-ACBBA algorithm is preferred because it achieves high total profit and high percentage of scheduled urgent tasks.     

Plastic deformation analysis and forming quality prediction of tube NC bending

Plane strain assumption and exponent hardening law are used to investigate the plastic deformation in tube bending. Some theoretical formulae including stress, curvature radius of neu-tral layer, angle of neutral layer deviation, bending moment, wall thickness variation and cross-section distortion, are developed to explain the phenomena in tube bending and their magnitudes are also determined. During unloading process, the springback angle is deduced using the virtual work principle, and springback radius is also given according to the length of the neutral layer which remains unchanged before and after springback. The theoretical formulae are validated by the experimental results or the validated simulation results in literature, which can be used to quickly predict the forming quality of tube numerical control (NC) bending.     

致密金属零件的激光近形制造

介绍了激光近形制造系统的组成,分析了该技术的特点和应用场合,指出了目前存在的问题和解决办法.