基于机器学习的二次电子发射唯象模型

基于机器学习和深度人工神经网络(artificial neural network, ANN)提出一种二次电子发射唯象模型。利用Vaughan模型生成先验数据集,用于训练生成描述二次电子发射一般规律的先验知识ANN模型,并在不同参数条件下验证了先验知识ANN模型的正确性。然后,分别利用银和铝合金材料的二次电子发射系数实验数据修正先验知识ANN模型,分别得到了描述两种材料的特异ANN模型。测试结果表明,特异ANN模型计算结果与实验结果相比的平均绝对误差较Vaughan模型和Furman模型降低了30%以上,与复合唯象模型精度相当或更高。在小样本条件下测试了二次电子发射ANN模型的正确性,验证了分步训练方式的有效性和二次电子发射ANN模型对于小样本集的适应性。提出的基于机器学习的二次电子发射唯象模型能够避免复杂的参数修正过程,能够基于先验知识提升模型对于小样本的适应性,能够实现二次电子发射系数的连续插值,适于在数值模拟软件中使用。     

Imaginary filtered hindsight experience replay for UAV tracking dynamic targets in large-scale unknown environments

As an advanced combat weapon, Unmanned Aerial Vehicles (UAVs) have been widely used in military wars. In this paper, we formulated the Autonomous Navigation Control (ANC) problem of UAVs as a Markov Decision Process (MDP) and proposed a novel Deep Reinforcement Learning (DRL) method to allow UAVs to perform dynamic target tracking tasks in large-scale unknown environments. To solve the problem of limited training experience, the proposed Imaginary Filtered Hindsight Experience Replay (IFHER) generates successful episodes by reasonably imagining the target trajectory in the failed episode to augment the experiences. The well-designed goal, episode, and quality filtering strategies ensure that only high-quality augmented experiences can be stored, while the sampling filtering strategy of IFHER ensures that these stored augmented experiences can be fully learned according to their high priorities. By training in a complex environment constructed based on the parameters of a real UAV, the proposed IFHER algorithm improves the convergence speed by 28.99% and the convergence result by 11.57% compared to the state-of-the-art Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm. The testing experiments carried out in environments with different complexities demonstrate the strong robustness and generalization ability of the IFHER agent. Moreover, the flight trajectory of the IFHER agent shows the superiority of the learned policy and the practical application value of the algorithm.     

基于深度强化学习的无人机栖落机动控制策略设计

无人机栖落机动飞行是一种无需跑道的降落方法，能够提升无人机在复杂环境下执行任务的适应能力。针对具有高非线性、多约束特性的无人机栖落机动过程，提出了一种基于模仿深度强化学习的控制策略设计方法。首先，建立了固定翼无人机栖落机动的纵向非线性动力学模型，并设计了无人机栖落机动的强化学习环境。其次，针对栖落机动状态动作空间大的特点，为了提高探索效率，通过模仿专家经验的方法对系统进行预训练。然后，以模仿学习得到的权重为基础，采用近端策略优化方法学习构建无人机栖落机动的神经网络控制器。最后，通过仿真验证了上述控制策略设计方法的有效性。     

一种无人机自主避障与目标追踪方法

针对无人机自主避障与目标追踪问题，以深度Q网络(DQN)算法为基础，提出一种多经验池深度Q网络(MP DQN)算法，使无人机避障与追踪的成功率和算法的收敛性得到优化。更进一步，赋予无人机环境感知能力，并在奖励函数中设计了方向奖惩函数，提升了无人机对环境的泛化能力以及算法的整体性能。仿真结果表明，相较于DQN和双重DQN(DDQN)算法，MP DQN算法具有更快的收敛速度、更短的追踪路径和更强的环境适应性。     

Force measurement using strain-gauge balance in shock tunnel based on deep learning

When a force test is conducted in a shock tunnel, vibration of the Force Measurement System (FMS) is excited under the strong flow impact, and it cannot be attenuated rapidly within the extremely short test duration of milliseconds order. The output signal of the force balance is coupled with the aerodynamic force and the inertial vibration. This interference can result in inaccurate force measurements, which can negatively impact the accuracy of the test results. To eliminate inertial vibration interference from the output signal, proposed here is a dynamic calibration modeling method for an FMS based on deep learning. The signal is processed using an intelligent Recurrent Neural Network (RNN) model in the time domain and an intelligent Convolutional Neural Network (CNN) model in the frequency domain. Results processed with the intelligent models show that the inertial vibration characteristics of the FMS can be identified efficiently and its main frequency is about 380 Hz. After processed by the intelligent models, the inertial vibration is mostly eliminated from the output signal. Also, the data processing results are subjected to error analysis. The relative error of each component is about 1%, which verifies that the modeling method based on deep learning has considerable engineering application value in data processing for pulse-type strain-gauge balances. Overall, the proposed dynamic calibration modeling method has the potential to improve the accuracy and reliability of force measurements in shock tunnel tests, which could have significant implications for the field of aerospace engineering.     

考虑移动目标不确定行为方式的轨迹预测方法

针对现有方法难以预测出符合飞行移动目标不确定行为方式轨迹的问题,提出基于逆强化学习的飞行移动目标轨迹预测方法,通过学习目标行为偏好以及模拟目标行为决策过程的方式预测目标的移动轨迹。首先基于深度神经网络建立目标的行为决策模型与行为偏好模型,然后通过最大熵逆强化学习方法交替地学习模型参数。为了有效地学习目标的不确定行为特征,采用监督学习的方法学习出目标示例轨迹概率分布模型,用于指导目标行为偏好模型的训练以及初始化目标行为决策模型,同时通过对目标行为偏好模型进行预训练的方式提高其训练质量。仿真结果表明,提出的飞行移动目标轨迹预测方法可通过学习到的目标行为决策模型较为准确地模拟目标的行为方式,预测的目标轨迹分布与真实的目标轨迹分布在Kullback Leibler(KL)散度下的相似度可达0.24。     

基于深度学习的非合作航天器姿态估计

针对空间非合作航天器姿态测量时受光照和地球背景影响大的问题,提出了一种基于卷积神经网络的端到端姿态估计方法.在该方法中,主干网络采用AlexNet与ResNet.首先,移除主干网络末端的全连接层,并列连接3个全连接层,采用三分支网络分别对姿态角进行估计.然后,设计了将分类问题与回归问题相结合的损失函数,通过分类方法将姿态估计限定在某一范围内,再使用回归方法进一步微调姿态.姿态分类损失函数确定姿态角度基准点,姿态回归损失函数对估计角度进行微调.相较于仅采用回归方法进行姿态估计,此方法能够有效减小姿态估计平均绝对误差、标准差与最大误差.实验对比了不同主干网络的测量精度,平均绝对误差在0.376°～0.746°之间,最优标准差为0.474°.     

Rotating machinery fault detection and diagnosis based on deep domain adaptation: A survey

In practical mechanical fault detection and diagnosis, it is difficult and expensive to collect enough large-scale supervised data to train deep networks. Transfer learning can reuse the knowledge obtained from the source task to improve the performance of the target task, which performs well on small data and reduces the demand for high computation power. However, the detection performance is significantly reduced by the direct transfer due to the domain difference. Domain adaptation (DA) can transfer the distribution information from the source domain to the target domain and solve a series of problems caused by the distribution difference of data. In this survey, we review various current DA strategies combined with deep learning (DL) and analyze the principles, advantages, and disadvantages of each method. We also summarize the application of DA combined with DL in the field of fault diagnosis. This paper provides a summary of the research results and proposes future work based on analysis of the key technologies.     

An automatic isotropic/anisotropic hybrid grid generation technique for viscous flow simulations based on an artificial neural network

Based on the author’s previous research, a novel hybrid grid generation technique is developed by introducing an Artificial Neural Network(ANN) approach for realistic viscous flow simulations. An initial hybrid grid over a typical geometry with anisotropic quadrilaterals in the boundary layer and isotropic triangles in the off-body region is generated by the classical mesh generation method to train two ANNs on how to predict the advancing direction of the new point and to control the grid size....     

Supervised learning with probability interpretation in airfoil transition judgment

Transition prediction has always been a frontier issue in the field of aerodynamics. A supervised learning model with probability interpretation for transition judgment based on experimental data was developed in this paper. It solved the shortcomings of the point detection method in the experiment, that which was often only one transition point could be obtained, and comparison of multi-point data was necessary. First, the Variable-Interval Time Average (VITA) method was used to transform the fluctuating pressure signal measured on the airfoil surface into a sequence of states which was described by Markov chain model. Second, a feature vector consisting of one-step transition matrix and its stationary distribution was extracted. Then, the Hidden Markov Model (HMM) was used to pre-classify the feature vectors marked using the traditional Root Mean Square (RMS) criteria. Finally, a classification model with probability interpretation was established, and the cross-validation method was used for model validation. The research results show that the developed model is effective and reliable, and it has strong Reynolds number generalization ability. The developed model was theoretically analyzed in depth, and the effect of parameters on the model was studied in detail. Compared with the traditional RMS criterion, a reasonable transition zone can be obtained using the developed classification model. In addition, the developed model does not require comparison of multi-point data. The developed supervised learning model provides new ideas for the transition detection in flight experiments and other experiments.