To address the challenges of entity-relation extraction in the unmanned aerial vehicle power system assembly domains-particularly annotation scarcity and Chinese semantic complexity, this paper constructs a task-adapted dataset, AM_NER, and proposes an entity relationship extraction model-MTB_Proto, which integrates a pretrained language model with a prototypical network. Within the pre-trained language model (the bidirectional encoder representations from the transformers) framework, the method introduces an entity position matching mechanism and a masked language model loss to enhance relation representation learning, while leveraging the prototypical network for few-shot classification. Experimental results show that MTB_Proto significantly outperforms baseline models on multiple few-shot tasks in the AM_NER dataset, achieving an F1 score as high as 76.19 %. On the public datasets SemEval-2010 Task 8, MTB_Proto achieved an F1 score of 91.90 % under the corresponding test setting, and reached 92.90 % on FewRel_process. These results demonstrate that MTB_Proto excels not only on our proprietary AM_NER dataset but also exhibits strong generalization and effectiveness on public benchmarks. Overall, the model effectively addresses few-shot relation extraction challenges in unmanned aerial vehicle power system assembly and mitigates the long-tail problem in vertical-domain Chinese relation extraction.

In recent years, unmanned aerial vehicles (UAVs) have had excellent performance in various fields, but their frequent component faults often lead to damages and serious accidents, so it is crucial to carry out timely fault diagnosis for them. Deep learning is widely used in the field of UAV fault diagnosis due to its superior feature extraction capability, but the increasing complexity of UAV faults and the scarcity of data have limited the development of deep learning in this field. To address the above problems, this paper proposed an Attention-based Joint Multi-Spatial Shared Knowledge Network (A-MSKN) for multi-objective fault diagnosis of UAVs under small samples. A-MSKN considers both complementary relationships between different tasks and intra-task dependencies within the same task for individual fault samples in different time intervals. Firstly, a single fault sample is divided into multiple sub-samples based on different time slices, and different sub-samples are coded to obtain different feature sub-spaces. Then, a sharing unit based on attention is designed to share not only the different feature subspaces within a task but also the features related between different tasks in a more fully shared way, to obtain more fault information for fault diagnosis under small samples. Finally, the effectiveness of the A-MSKN in the case of small samples was verified by testing it on real faulty flight data.

The manufacturing processes of unmanned aerial vehicle (UAV) power systems generate large amounts of data and knowledge. The extraction of useful information or patterns from redundant data and knowledge texts has become a challenge in intelligent manufacturing. Unfortunately, graphics processing unit (GPU)-based parallel computing is limited, and the inference speeds of the available named entity recognition (NER) models for Chinese text datasets are low because they are mainly based on the long short-term memory (LSTM) algorithm. Herein, first, the flat-lattice transformer (FLAT) model was optimized by using a stochastic gradient descent with momentum (SGDM) optimizer and adjusting the model hyperparameters. Compared with the existing NER methods, the proposed optimization algorithm achieved better performance on the available dataset. Then, an NER method named the TL_FLAT model based on transfer learning and the abovementioned optimization model was introduced. Finally, a Chinese text dataset from a UAV power system created by the authors was used to validate the proposed method. The F1 score was 76.26%, the precision value was 76.98%, and the recall value was 75.56%, indicating that the TL_FLAT model was suitable for Chinese text entity recognition for UAV power systems.

Fault diagnosis methods based on deep learning have progressed greatly in recent years. However, the limited training data and complex work conditions still restrict the application of these intelligent methods. This paper proposes an intelligent bearing fault diagnosis method, i.e., Siamese Vision Transformer, suiting limited training data and complex work conditions. The Siamese Vision Transformer, combining Siamese network and Vision Transformer, is designed to efficiently extract the feature vectors of input samples in high-level space and complete the classification of the fault. In addition, a new loss function combining the Kullback-Liebler divergence both directions is proposed to improve the performance of the proposed model. Furthermore, a new training strategy termed random mask is designed to enhance input data diversity. A comparative test is conducted on the Case Western Reserve University bearing dataset and Paderborn dataset and our method achieves reasonably high accuracy with limited data and satisfactory generation capability for cross-domain tasks.

目前航空装备制造企业在设计、制造相关流程中积累了大量数据,基于知识图谱技术可以对这些数据进行有效融合与管理,对不断更新的制造知识进行挖掘,为航空制造企业智慧化升级提供有力的知识支撑.为探明知识图谱在航空制造领域的理论支撑体系与实际应用情况,通过文献调研分析航空制造知识图谱架构、定义及特点.阐明知识图谱领域构建过程中的核心技术并进行研究综述,对比航空制造知识图谱与通用知识图谱构建技术上的异同,并提出了三个切合实际的航空制造知识图谱应用方向及其解决方案.最后对未来航空制造知识图谱的挑战进行了分析及展望,为后续该领域的研究提供一些思路.

针对产线设备运行与管理存在的实时性差、精度低等问题,构建了一种针对产线设备状态预测准确、误差小、鲁棒性高的基于数字孪生的PHM系统,面向空调、化成箔等行业设备生产线进行了试点应用,验证了该系统在多行业产线设备状态预测中的可行性,总结了系统在产线设备中应用的经验,分析并提出了下一步发展思考.

The content, distribution and size of retained austenite (RA) affect its mechanical stability in carburized layers. The stability of RA plays a decisive role in cold work hardening and strain-induced martensitic transformation during sliding friction; these changes determine wear resistance. In this study, a database was established based on laser confocal metallographic images of carburized layers on 23CrNi3MoA steel after different carburizing treatments. Eight algorithms were used to identify and calculate the amounts of RA in the carburized layers. The tribolayers and wear on the surfaces that underwent three carburizing processes, P13, P15, and P17, were characterized and tested. The results showed that the U-Net algorithm with an attention module and drop block regularization was the most suitable for the database. Predictions of the RA contents of surfaces after P13, P15, and P17 treatments were 19.9%, 28.1%, and 40.1%, respectively. The errors of the predictions compared with experimental results were within 5%. The surface carburized by the P15 process contained moderate amounts of RA and had the best wear resistance because the friction strain induced the formation of nanolamellar structures and the transformation of RA to martensite. The results of this study support the use of deep learning to identify and calculate the amounts of RA in carburized layers and optimize carburizing processes of mild steel.

In order to improve the accuracy of manipulator operation, it is necessary to install a tactile sensor on the manipulator to obtain tactile information and accurately classify a target. However, with the increase in the uncertainty and complexity of tactile sensing data characteristics, and the continuous development of tactile sensors, typical machine-learning algorithms often cannot solve the problem of target classification of pure tactile data. Here, we propose a new model by combining a convolutional neural network and a residual network, named ResNet10-v1. We optimized the convolutional kernel, hyperparameters, and loss function of the model, and further improved the accuracy of target classification through the K-means clustering method. We verified the feasibility and effectiveness of the proposed method through a large number of experiments. We expect to further improve the generalization ability of this method and provide an important reference for the research in the field of tactile perception classification.

Researchers all over the world are aiming to make robots with accurate and stable human-like grasp capabilities, which will expand the application field of robots, and development of a reasonable grasping strategy is the premise of this function. In this paper, the improved deeplabV3+ semantic segmentation algorithm is used to predict a triangle grasp strategy. The improved model was trained on the relabeled Cornell grasp datasets and tested on self-collected datasets. Compared with the existing rectangular grasp strategy, the proposed algorithm and triangle grasp strategy have achieved outstanding performance in stability, accuracy, and speed. Finally, based on the ROS platform, this paper deploys the trained model and verifies the real effect of the trained grasping strategy prediction model, and achieves excellent grasping effect.

Deep neural networks are widely used in the field of image processing for micromachines, such as in 3D shape detection in microelectronic high-speed dispensing and object detection in microrobots. It is already known that hyperparameters and their interactions impact neural network model performance. Taking advantage of the mathematical correlations between hyperparameters and the corresponding deep learning model to adjust hyperparameters intelligently is the key to obtaining an optimal solution from a deep neural network model. Leveraging these correlations is also significant for unlocking the “black box” of deep learning by revealing the mechanism of its mathematical principle. However, there is no complete system for studying the combination of mathematical derivation and experimental verification methods to quantify the impacts of hyperparameters on the performances of deep learning models. Therefore, in this paper, the authors analyzed the mathematical relationships among four hyperparameters: the learning rate, batch size, dropout rate, and convolution kernel size. A generalized multiparameter mathematical correlation model was also established, which showed that the interaction between these hyperparameters played an important role in the neural network’s performance. Different experiments were verified by running convolutional neural network algorithms to validate the proposal on the MNIST dataset. Notably, this research can help establish a universal multiparameter mathematical correlation model to guide the deep learning parameter adjustment process.

以数控机床用的深沟球轴承为研究对象,分析其运转中的振型及结构疲劳寿命.首先通过Solidworks建立了深沟球轴承的三维模型,利用ANSYS对轴承进行模态及谐响应分析,获得轴承在固有振动频率以及在各阶次频率下的固有振型变化,同时借助有限元谐响应分析,确定了对轴承影响最大的模态频率.其次利用ANSYS中的Faigue模块对轴承进行了疲劳寿命研究,通过结合材料的S-N曲线理论以及Hertz接触理论,对轴承的疲劳寿命进行了预估分析.仿真结果表明:在满足轴承强度工况下,固有频率1125Hz为结构共振最大点,且轴承的寿命范围为5937.7～1×106次,为后续轴承结构优化提供参考.

<span id="ChDivSummary" name="ChDivSummary" class="abstract-text">果树农学参数分别指树叶稀密程度、冠幅、叶面积指数（LAI）,通过同时对应检测三个时期不同健康状况的175株果树的三个农学参数。以SPSS为平台进行果树农学参数模型构建,利用最大轮廓矩形法检测的果树树叶稀密程度作为自变量,冠幅和叶面积指数分别为因变量来构建果树农学参数模型:树叶稀密程度与冠幅之间的最优关系模型为对数模型:y=1.839+1.306lnx,树叶稀密程度与LAI之间的最优关系模型为S模型:z=7×e（-0.137-0.336 x）。该农学参数关系模型的构建将为实现果树农学参数的实时检测提供算法依据。</span>

为了将已研究的树叶稀密程度检测算法——最大轮廓矩形法应用于DSP采集系统和对应硬件设备,以OpenCV为平台,将最大轮廓矩形法进行移植.试验证明:树叶稀密程度检测算法在OpenCV上是可以正确有效实现,OpenCV可以作为研究出的算法应用于以C语言为汇编语言的DSP采集系统的平台,从而实现果树农学参数的实时检测.

<span id="ChDivSummary" name="ChDivSummary" class="abstract-text">变量喷雾是现代农药精确喷施技术之一。为了实现果树变量喷雾,需要检测果树的形态结构参数。图像处理技术是获取果树形态结构参数的重要方法。本文根据果树图像的特征,分别基于灰度模型、RGB模型和HSV模型对果树图像进行图像分割试验。试验结果表明,有效的果树图像分割方法是基于灰度模型的图像分割与基于RGB模型的（R+G+B）、2R、2G等分量的图像分割。其中,基于灰度模型的图像分割采用固定阈值128,基于RGB模型的图像分割采用Otsu算法。比较各种有效图像分割方法的准确率和耗时,结果表明采用基于RGB模型的2R分量、Otsu算法可以满足果树图像分割处理的要求,该方法分割准确率为79.67%。</span>

Through analyzing the characteristics of cold extrusion for thin-wall piston,a process scheme of cold extrusion for thin-wall steel piston was proposed and the parameters of the process were calculated.Its technology was numerically simulated by Qfrom-3D software.And according to the simulation results,reasonable dies were designed.This process can significantly improve the efficiency,save materials and reduce the casts.

In order to solve the problem that the existing method for detecting the leaf density of fruit trees is advisable to consider the adoption of standard whiteboard or fixed imaging distance when grab images. This paper presents a new method for detecting the leaf density which based on image processing- maximum profile rectangle technology. Images effectively segmented by using ultra- green method,Ostu,median filter algorithm,erosion and dilation and other image processing technologies; Detecting the largest share of fruit trees area of binary image after processed,then test the entire image area occupied by the leaves and tree trunks,the leaf density can be calculated according to its definition. Experimental results show that it doesn't need to use the fixed imaging distance and the white calibration image,the detection of fruit trees image area due to the actual image varies in the maximum profile rectangle technology,there is no uniform application of the conventional method which set the image size in the camera as a result of the maximum outline of the all images,it avoid the problem that results is too small. Among the 20 tested images,the maximum difference is 0. 2950 and minimum difference is 0. 0027 compared with the existing method.

According to the cold extrusion process for steel thin-wall piston,the slip line field of solving forming problems of axis symmetrical cup-shaped thin-wall pieces for the pistons was drawn. A mathematical model which contains wall thickness,friction factor,material shear strength and unit contact stress was first established; the process of cold extrusion forming for pistons was numerically simulated by Qform-3D. It was confirmed that the above mathematical model is useful to pressure calculation of cold extrusion forming for pistons.

In order to access image data in digital image processing based variable spray system,A SDRAM controller with data,address and control bus interface is designed using FPGA chip.Finite state machine and modular design was used,and the controller was divided into the initialization state machine and the reading and writing command control state machine.Modlesim and quartus was used for the design verification,the results show that the controller can work at a max frequency of 135 MHz.Because of its simple interface and high speed access characteristic,the controller can meet the variable spray system image data real-time access requirements.

<span id="ChDivSummary" name="ChDivSummary" class="abstract-text">设计一种采用数字信号处理器（DSP）TMS320F2812为微控制器的直流电机控制系统,分析了直流电机的调速原理,阐述该控制系统的硬件结构、软件流程和控制算法,并使用Matlab/Simulink搭建控制系统的原理图,对控制系统进行仿真试验。结果分析表明:以DSP为控制器并结合参数自适应模糊PID控制的控制系统,具有响应速度快、稳定性高、超调量小等特点,较好地满足变量喷雾的控制要求。</span>

A variable-rate spray system of fruiter needs real-time detection for the fruiter crown diameter. Existing digital image processing technology can not realize real-time detection for the fruiter crown diameter. The problem is that the detection error is large or no error can be detected. A reference standard plate with fruiters when capturing images is used to obtain the fruiter crown diameter by calculating image unit pixel width in this paper. Comparing with the reference crown diameter detected by the handheld laser rangefinder, test results show that 74% of the test sample of 50 fruiters have a relative difference of less than 20% between the crown diameters detected by the image processing method and the reference values. The regression analysis of fruiter crown diameter on digital image detection value and the reference one shows R2=0.7406.