Online chip appearance inspection remains difficult when small damage, slender bonding-wire fractures, and weak-texture scratches are mixed with specular highlights, adhesive-surface patterns, and structured backgrounds. We address this setting with SCALE-RTDETR, a real-time RT-DETR detector that reconstructs multi-scale features before they are delivered to the Transformer decoder. Rather than modifying the backbone, neck, or loss function in a general-purpose manner, the method focuses on a narrower but practically important point in the RT-DETR pipeline: after channel projection, the multi-scale features are usually passed to encoder-decoder processing with little task-specific refinement. The proposed SCALE module contains three coordinated branches. Cross performs cross-scale semantic alignment and bidirectional detail transfer to reduce the attenuation of small-defect responses; CF strengthens deep contextual discrimination so that reflections, shadows, and adhesive textures are less likely to dominate the representation; and ELA uses direction-aware spatial recalibration to improve localization of elongated structures such as scratches and wire-bond fractures. On a 5874-image chip defect dataset, SCALE-RTDETR raises mAP@50–95 from 88.67% to 90.04%, a gain of 1.37 percentage points over the baseline. With batch size = 1 on an RTX 4090, it runs at 32.1 FPS with 31.15 ms single-frame latency, remaining within the 33.33 ms budget required for 30 FPS processing. Category-level AP further improves by 2.33 and 2.11 percentage points for gold-wire bond fracture and chip scratch, respectively. These results show that pre-decoder feature reconstruction can improve weak-texture and elongated-defect representation while retaining real-time inference capability.

Wireless Sensor Networks (WSNs) operating in harsh environments are prone to the formation of isolated node clusters, leading to communication interruptions. To restore network connectivity and enhance the efficiency of multi-robot collaboration, this paper focuses on the optimization of relay node deployment within the context of multi-robot island alliance formation. The objective is to efficiently reconnect isolated subgroups while optimizing the overall network performance post-alliance formation. This paper proposes a collaborative optimization strategy based on energy balancing and load balancing. A zonal energy model is constructed to constrain robot energy consumption, and a heuristic deployment algorithm is designed. Additionally, a local load-aware model is established, and a graph search algorithm is utilized to plan disjoint paths for traffic balancing. Experimental results demonstrate that the proposed method effectively reduces the number of relay nodes, shortens deployment time, prolongs network lifetime, and improves load balancing performance. This study provides a robust solution for reliable communication in multi-robot systems operating within complex scenarios.

Object detection is an important task in computer vision, aimed at detecting and recognizing the position and category of target objects from images or videos. With the rise of deep learning, the accuracy and efficiency of object detection have significantly improved, especially the application of convolutional neural networks (CNN) in this field, which has made significant breakthroughs in object detection methods. This article provides an overview of the development history of object detection, with a focus on classic object detection algorithms, deep learning methods, and their evolution. It explores the evaluation criteria and challenges faced by object detection, and looks forward to future development trends.

Transparent objects challenge monocular perception due to refraction, reflection, and weak textures, which hinder accurate depth estimation and segmentation. To overcome these issues, we propose CESINet, a curvature-enhanced synergistic attention network for transparent object perception. CESINet explicitly incorporates surface curvature as a high-order geometric prior to strengthen spatial representation and introduces a curvature-guided synergistic attention module to enable effective cross-task feature interaction between depth and segmentation branches. A curvature consistency loss further enforces geometric coherence across predictions. Experiments on the ClearPose dataset show that CESINet achieves 94.33% mIoU and 98.27% mAP for segmentation, improving over the multi-task baseline ISGNet by 1.49% and 0.44%, respectively. For depth estimation, CESINet attains an RMSE of 0.112 and REL of 0.060, reducing errors by 8.9% and 11.8% compared with the baseline. Ablation results demonstrate that removing curvature priors or attention modules leads to performance drops of up to 3.5% in segmentation and 12% in depth accuracy, confirming the complementary benefits of explicit geometry and synergistic learning. Overall, CESINet enhances geometric consistency and boundary sharpness while maintaining computational efficiency, providing a unified and scalable framework for multi-task transparent object understanding.

Thin-walled components, known for their lightweight and high-performance characteristics, hold significant strategic importance in industries such as aerospace, radar, and transportation. However, due to their inherently low stiffness—encompassing shear, bending, and torsional stiffness—these components are highly susceptible to deformation during machining. This deformation can adversely affect the geometric integrity and machining precision of the component, including dimensional accuracy, shape accuracy, and positional accuracy. Controlling the deformation of thin-walled components has thus become a critical research focus in recent years. This paper reviews the latest developments in the types of machining deformation, deformation mechanisms, and deformation control strategies for thin-walled components, aiming to equip readers with dynamic approaches for achieving high efficiency and precision in thin-walled component machining. The first section provides an overview of the definition, classification, and factors affecting the machining accuracy of thin-walled components. The second section discusses the mechanisms behind the deformation of thin-walled components, which result from a combination of multiple factors, including deformation caused by cutting forces, cutting temperature, residual stress, fixturing, and machining chatter. The third section reviews several methods for controlling deformation, including adaptive machining and error compensation, stability lobe diagrams and chatter suppression, deformation prediction and control, and energy field-assisted machining. These methods allow for the control and prevention of thin-walled component deformation before, during, and after machining. Finally, the paper summarizes the current challenges in thin-walled component machining and outlines future development trends. The research content and methods introduced in this paper, including theoretical analysis, experimental validation, and simulation analysis, provide researchers with a clear background and research roadmap, contributing to the exploration and improvement of high-precision machining techniques for thin-walled components in future research.

In order to improve the motion control characteristics of articulated rock-drilling robot, an articulated robot chassis structure with encoder expansion link and distributed electric wheel is designed, and a steering control strategy based on push rod pose feedback is proposed. Based on Adams/Simulink co-simulation, the steering control strategy model is built. The motion of the articulated rock drilling robot in various working conditions is analyzed by co-simulation experiment. Compared with the traditional control strategy, the steering time of the articulated rock-drilling robot by the steering control strategy is increased to 13.87%, and the steering distance is reduced to 16.95%, which effectively improves the steering ability of the robot.

In close-range photogrammetry, it is difficult to meet the measurement requirements of large scenes in actual engineering due to the limited capacity of coded targets. To expand the capacity of the coded target, we propose a binary step-response serial-coded target (BSSCT). The BSSCT introduces periodic binary wave information as an additional feature in the dot-dispersing coded target. Also, a robust recognition algorithm for the BSSCT is developed, the P2-Invariant, and the step period is used for decoding. The capacity of the coded target reaches 7 magnitudes without increasing the auxiliary points. Under different lighting conditions and viewing angles, the measurement experiments show that the BSSCT outperforms other state-of-the-art coded targets. Our BSSCT is a promising standard method for large-field system calibration and object measurement.

The human body generates a sophisticated three-dimensional complex motion of the centre of mass when walking, with the majority of this vibration occurring in the vertical and coronal axes. This paper presents an analysis of the human walking motion, accompanied by the establishment of an equivalent model of the human walking COM motion. This law is integrated into the coronal plane to generate the Lissajous curve expression of the human walking COM motion trajectory. Finally, the reliability of the results was verified by comparing the COM motion equivalent model with several sets of walking experimental data. The model is capable of accurately predicting and analysing the walking amplitude and frequency of the COM at varying heights and speeds. Furthermore, a dual-degree-of-freedom power suspension backpack system has been designed based on the model, which has the potential to significantly reduce the low-frequency vibration and inertial force impacts of heavy loads on the human body's vertical and coronal axes during load-bearing walking. This could help to alleviate the resulting mechanical injuries to the human body and reduce the additional metabolic energy consumption required to resist these impacts. The device is capable of effectively reducing the low-frequency vibration and inertial force impacts in the vertical and coronal axes of the human body when walking with a load.

External impacts for Thin walled components can cause surface depressions and result in serious consequences. In order to detect these depressions , this article performs point cloud filtering and simplification prepossessing on the obtained defect point cloud, and uses the defective point cloud to fit a standard point cloud model; We used a point cloud feature description algorithm based on Fast Point Feature Histogram (FPFH) and the Nearest Point Iteration (ICP) algorithm to perform coarse and fine registration on two point clouds. Finally, we calculated the Euclidean distanceto defect the registered point clouds.

Aiming at the problem of inaccurate positioning accuracy of V-SLAM system caused by poor texture and blurred images generated by mobile robots during rapid movement and large angle rotation, an improved ORB feature extraction matching algorithm is proposed. Firstly, using the statistical form of information entropy to calculate the image features, image enhancement is used to highlight the feature information of poorly textured and blurred image blocks, improve the efficiency of feature point detection, and enhance the robustness of the system. In addition, to improve the accuracy of feature matching, the BEBLID descriptor is used instead of the BRIEF descriptor for feature description. Through experimental verification on the publicly available TUM data set, it was found that the improved algorithm has a higher matching accuracy than the ORB algorithm, and has a significant increase in algorithm time compared to the ORB algorithm. After trajectory calculation, the improved algorithm is more in line with the real trajectory in camera motion and has a good improvement in pose accuracy.

Abstract This study addresses the issues and limitations of trajectory tracking for Non-holonomic wheeled mobile robot (NWMR) under traditional PID control, including low accuracy, high response delay, poor robustness, and stability concerns. We propose a strategy that combines the Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm with fuzzy PID control to optimize PID parameter tuning for more precise robot trajectory tracking. Fuzzy logic is introduced to adjust the PID controller parameters, making them more adaptive and robust. The bidirectional long short-term memory (BiLSTM) network enhances the time series processing capability of the Actor-Critic network, improving the system’s state representation and prediction abilities. A curiosity-driven exploration method is employed to increase policy diversity and avoid premature convergence. Simulation experiments using the ROS Noetic and Gazebo platforms demonstrate that this method significantly outperforms traditional PID and TD3 algorithms in terms of trajectory alignment, training time, accuracy, and stability.

In this paper, the geometric error modeling and compensation methods of macro–micro composite five-axis turn-milling composite machining center is studied. Firstly, the machine topological structure of the branches, intermediates, and terminal bodies of the macro–micro composite multi-body system is analyzed. Then, the tool chain transformation matrix and the end position of the workpiece chain are established to obtain the geometric error model. Based on the error compensation theory of traditional machine tool structure, the compensation mechanism of macro-level compensation and micro-level sub-micron compensation is proposed. Then, the compensation model of micro-axis error is given. Furthermore, the macro–micro composite error compensation experiment is setup; the laser interferometer is used to judge the positioning accuracy and straightness before and after compensation. The results show that the accuracy of the micro-motion platform after compensation reaches the sub-micron level, which verifies the compensation method, and the machining accuracy of the micron level is achieved through the cutting experiment.

Aiming at the problem of positional accuracy degradation of industrial robots in long-term operation, a positional accuracy degradation detection method of industrial robots based on kinematic analysis is proposed. Firstly, the accuracy of the joint angles collected from the robot control cabinet and external high-precision sensors is compared and analyzed based on the MD- H method; secondly, the positional degradation detection model of the industrial robot is established based on the principle of differential kinematics; and finally, an experimental platform is built based on the KUKA KR 3 R540 industrial robot and experimental validation is carried out. The experimental results show that the root-mean-square error between the detected position error and the actual position error is 0.049 mm, and the method can effectively detect the position accuracy degradation of the robot.

针对在非结构工业场景中需要机器人对工人的行为意图做出准确的识别和快速的理解才能更加安全高效地完成人机协作任务的前提,提出一种基于3D骨骼点数据对人体动作识别的算法.该动作识别算法是以自建的数据集作为输入,通过BlazePose三维人体姿势估计算法处理得到人体骨骼点数据,提取33个骨骼点的3D坐标信息,数据预处理之后,为了减少模型的预测时间,使用主成分分析算法PCA将数据从99维降到20维,将未降维的数据和降维后的数据分别送入到BP神经网络算法中进行训练,最终用训练好的模型实现对3种人体动作的分类.在自建数据集上的实验结果表明,当使用未降维的数据进行训练时,该动作识别算法在测试集的准确率可以达到98.37％,在视频数据上的识别速度可达每秒22帧左右;当使用降维数据时,该动作识别算法的识别准确率可达到98.86％,识别速度可达每秒30帧左右.经过实验对比,当使用降维后的数据时,该动作识别算法的识别准确度更高一点,实时性更好,值得在非结构工业场景中应用推广.

针对PID控制器轨迹跟踪精度差的问题,在BP网络对PID参数拟合的基础上,利用算法融合的思想,使用改进后的遗传算法对BP网络进行参数寻优,避免BP网络陷入局部最优解,再使用BP网络对PID参数进行拟合.文章给出了控制算法的整体流程,并结合移动机器人的运动学方程进行了仿真对比实验.结果显示:改进后的融合算法相比PID算法控制误差更小,即改进的GA与BP-PID融合算法在移动机器人轨迹跟踪精度上发生了明显的改善.

视觉SLAM系统在相机快速旋转或光照频繁变化时,极易跟踪丢失.为此,提出一种基于直接法和共视图优化的紧耦合视觉惯性SLAM系统,融合IMU信息提高系统的鲁棒性,采用直接法前端提高系统的实时性,共视图优化后端提高系统的定位精度.该系统由前端和后端以及回环检测三个模块组成.跟踪线程利用IMU信息和基于稀疏图像对齐的直接法进行初始位姿估计;后端采用共视图的方法,以当前帧的二级相邻共视关键帧范围为局部优化窗口,利用光束平差法(Bundle Adjustment,BA)对系统状态变量进行优化;另外,仅对关键帧提取ORB特征点,并计算描述子信息供回环检测使用.在TUM VI数据集上的实验证明,与ORB-SLAM3和VINS-mono相比,该算法提高了系统的定位精度,且位姿估计速度提高了50％ 以上,在一帧完整跟踪任务中,比VINS-mono实时性提高了26％.

为实现《中国制造2025》的战略目标,为我国经济社会发展提供有力的人才和智力支撑,《国家职业教育改革实施方案》提出了开展"学历证书+若干职业技能等级证书"制度试点工作.职业技能竞赛作为职业院校学生技能水平评价的重要手段,对"1+X"证书制度的实施具有重要的推动作用.以协作机器人服务应用赛项的实施为例,分析了协作机器人服务应用赛项的意义,合理设计了赛项的考核内容及评分标准,并对参赛选手的决赛成绩进行了分析,为以后开展职业技能竞赛提出了合理化建议,为开展全国行业职业技能竞赛提供了借鉴.

针对当前工业机器人状态监测模式单一、实时性差、可视化差等问题,本文提出了一种基于数字孪生技术的工业机器人状态监测系统.首先,通过三维建模软件建立工业机器人的几何模型,然后利用工业机器人监测系统对物理世界的工业机器人进行数据采集,再通过监测系统实时映射工业机器人与工作平台,实现对工业机器人状态的实时监测,最后在工业机器人"减速器模型"装配生产线中进行验证.运行结果表明,本文提出的监测系统能够直观并精准地实时监测设备运行状态.

为了提高装配生产线数据的可视化和现场设备的管理透明化水平,使用云物联网平台的数据处理能力、MQTT通信技术、传感器技术以及数据采集技术,开发一种智能装配生产线的云监控系统,并进行3个月的现场测试验证.系统主要由工业机器人、PLC、智能相机、通信网络与云服务器组成,在整体架构上设置了设备层、控制层、网络层和应用层,采集生产过程数据以及设备状态数据到PLC寄存器中,网关利用MQTT通信协议将PLC中的数据传送到云平台的数据库,客户端利用HTTP通信协议完成与云平台的交互,完成装配生产线现场与客户端之间的文本、图像、视频3类数据传输,实现随时随地使用不同的IT设备查看设备状态以及监控生产状态,并实现PLC程序的上传、监控、修改及下载功能.现场测试验证结果表明,云监控系统准确性高、实时性强、操作简单、能够满足生产需求,降低企业的用人成本,具有广泛的应用前景和经济效益.

煤矿井下多水、火、瓦斯等突发性灾害事故,危害性大、波及范围广、易造成重大人员伤亡和财产损失,安全避险尤为重要.针对现有煤矿井下存在的安全避险通信联络系统自动化程度较低及可靠性较差等问题,研制了一种安全避险智能通信联络系统.首先构建了基于光纤环网、电话调度网及CAN、RS485等多种通信模式互为冗余的多网络智能通信联络平台;基于Delphi开发并研制了具有数字广播、调度指挥、录音报警及存储等功能的通信主站上位机平台,并建立了GIS系统;实现终端设备与控制中心或调度台之间的双通道语音通信、分组广播、数字对讲管理、录音管理及日志管理等功能,建立了基于矿井灾变应急预案的一键智能逃生指引系统功能以及本安型应急LED显示屏避险逃生指示功能.系统可无缝接入矿井信息化管理系统,提高了系统的通信冗余能力,有效保证了与井下复杂环境、任意区域、不同工种人员语音通信清晰准确传达,针对不同种类不同区域的突发灾害,只需通过一键操作即可完成相对应的声音报警信息、井下逃生路线指挥音频和LED显示屏逃生方向指引信息的发布,高效应对各种灾变,进一步提高了矿山处理突发事件和一体化科学管理的效率.