Accurately identifying surface defects is important to realize higher production yield of flexible integrated circuit substrates (FICS). However, surface defect detection of FICS remains an extremely challenging task due to highly blurred defect edges, random background noise, and strong visual similarity between defective and defect-free regions. To address these issues, we propose a structure-preserving model for defect detection on FICS surfaces, which is composed of three cascaded modules, namely edge-preserving smoothing, surface texture removal, and defect localization. Specifically, the edge-preserving smoothing module is developed to smooth out background random noise while sharpening defect edges by spatially varying smoothness requirement. The surface texture removal module is further proposed to remove fine surface texture by solving a nonconvex optimization problem with truncated Huber penalty function. Finally, the defect localization module is designed to separate defect and non-defect regions in the framework of sparse signal analysis. Experimental results demonstrate that the proposed surface defect detection model is more competitive than existing methods in discriminating defects from FICS images with complex background texture.

With the advancement of high-density interconnect technology in semiconductor manufacturing, the precision and complexity of integrated circuit (IC) substrates have significantly increased, placing higher demands on quality control. Efficient and accurate detection of complex etching defects, which often occur during manufacturing, has become critical to preventing potential product defects. A defect detection method is proposed that combines a lightweight network with differential geometry tools to address the issue of etching defects in IC substrates. First, an improved deformable model is used to rapidly extract regular circuit trace contours from complex metallographic images, and morphological processing is applied to enhance the details, achieving precise image segmentation. For under-etching defects between circuit traces, differential processing of the original and segmented images is performed to locate abnormal regions. Subsequently, an optimized lightweight network based on MobileNet, termed OMNet, is designed to achieve the rapid identification of under-etching defects in these regions. For etching defects occurring on circuit traces, the DGEtch method employs a high-precision discrete curvature calculation based on the Frenet frame to evaluate angular discontinuities in contours, enabling accurate detection of etching defects. Experimental results demonstrate that the proposed method achieves an average recall rate of over 95% and maintains a precision above 90%. It exhibits high accuracy and stability in detecting etching defects and consistently outperforms existing models, particularly in handling complex mixed defects. This study provides an effective solution for detecting complicated defects in high-density IC substrate manufacturing.

The quality of flexible integrated circuit substrates (FICSs) is critical to the reliability of various electronic products, making intelligent defect measurement essential for efficient manufacturing and cost-saving. However, existing solutions for substrate defect diagnosis heavily rely on human visual interpretation, which leads to poor efficiency and a high error rate. A novel vision-based detection system consisting of a multiscale imaging module and a hierarchical structure based on the deep convolution neural network (DCNN) is proposed in this article. Rapid and accurate fault diagnosis can be enabled for high-density FICS, and various defects could be located and classified in a coarse-to-fine resolution. Specifically, a new mechanism of hierarchical decision based on DCNNs is built for FICS fault diagnosis, wherein the challenge of unbalanced data is addressed in the network learning process to reach a good trade-off between detection accuracy and speed. The substantial experiments and effectiveness comparison by using the typical methods on three categories of FICS and their corresponding eight-type faults reveal that the proposed system could facilitate the solution of substrate fault measurement and achieve high accuracy and efficiency, which could provide essential information of FICS to divide its industrial acceptance quality level.

Targeting the issue that the traditional target detection method has a high missing rate of minor target defects in the lithium battery electrode defect detection, this paper proposes an improved and optimized battery electrode defect detection model based on YOLOv8. Firstly, the lightweight GhostCony is used to replace the standard convolution, and the GhostC2f module is designed to replace part of the C2f, which reduces model computation and improves feature expression performance. Then, the coordinate attention (CA) module is incorporated into the neck network, amplifying the feature extraction efficiency of the improved model. Finally, the EIoU loss function is employed to swap out the initial YOLOv8 loss function, which improves the regression performance of the network. The empirical findings demonstrate that the enhanced model exhibits increments in crucial performance metrics relative to the original model: the precision rate is elevated by 2.4%, the recall rate by 2.3%, and the mean average precision (mAP) by 1.4%. The enhanced model demonstrates a marked enhancement in the frames per second (FPS) detection rate, significantly outperforming other comparative models. This evidence indicates that the enhanced model aligns well with the requirements of industrial development, demonstrating substantial practical value in industrial applications.

As the fabrication of high-density integrated circuit (IC) substrates advances in precision, the task of sensing and visually inspecting defects becomes increasingly challenging. These challenges arise from the need for precise segmentation of metallographic substrate images, complicated by variations in gray levels, noise interference, and rich textures. To address these issues, an adaptive fractional differentiation (AFD)-based active contour model (ACM) is proposed, which integrates global and local terms for a more accurate depiction of sensed image information. Using local image statistics, we construct an adaptive fractional-order model to find the optimal order for the fractional gradient. This gradient is then integrated into the Chan-Vese model as the global fitting term. The fusion of the original image with the fractional gradient image forms the local fitting object, refined by a Gaussian kernel function. Adaptive weight parameters are designed to enhance segmentation performance and accelerate evolution, adjusting the balance between global and local terms. Additionally, a distance penalty term is introduced to prevent reinitialization and improve segmentation efficiency. Experimental results show that the AFD model achieves accurate segmentation of high-density IC substrate images, with average improvements in DSC and JS scores of 16.95% and 28.52% , respectively. Moreover, it reduces average processing times by 59.47% , demonstrating better efficiency and robustness than other models.

电极作为动力电池的重要组成部分,其质量关系到电池的性能、安全性以及使用寿命.针对锂电池制作工程定位与表面缺陷检测速度慢、精度低等问题,提出一种改进YOLOv5的特征检测算法.首先引入卷积注意力模块,对特征在通道和空间维度上进行融合增强,提高微小目标的检测精度;然后改进了损失函数达到保留有利特征和提高收敛速度的目的.最后,在自建极片缺陷数据集上实验检测,在检测速度不变的情况下,改进模型在测试集上的mAP提高了1.2％,召回率提高1.5％,能够满足极片缺陷检测要求.

In order to meet the needs of the detection accuracy and speed of lithium-ion battery chip defects, this article proposes an improved algorithm based on deep learning YOLO5. First, introduce the coordinate attention mechanism (CA) to strengthen the study of the characteristics of the image area of the lithium-ion polar tablet so that the model focuses on the extraction of the characteristics of the polar defects and weakens the effects of complex background on the test results. Further, improve the accuracy of polar defect detection, and use the CIoU loss function to replace the GIoU function, so that the regression process focuses more on high-quality anchor frames and improves the convergence speed, regression accuracy, and robustness of the model. Based on the above experimental results, the two improvements were jointly added to the YOLOv5 model, the mAP was increased by 1.48%, and the detection speed FPS was increased by 2 f/s-1.The YOLOv5-CC algorithm proposed in this article is a lithium-ion with both detection accuracy and practicality. Battery pole defect detection algorithm has high application value.

The authors wish to make the following corrections in Section 3 [...]

The goal of this paper is to estimate object’s 6D pose based on the texture-less dataset. The pose of each projection view is obtained by rendering the 3D model of each object, and then the orientation feature of the object is implicitly represented by the latent space obtained from the RGB image. The 3D rotation of the object is estimated by establishing the codebook based on a template matching architecture. To build the latent space from the RGB images, this paper proposes a network based on a variant Adversarial Autoencoder (Makhzani et al. in Computer Science, 2015). To train the network, we use the dataset without pose annotation, and the encoder and decoder do not have a structural symmetry. The encoder is inspired by the existing model (Yang et al. in proceedings of IJCAI, 2018), (Yang et al. in proceedings 11 of CVPR, 2019) that incorporates the function of feature extraction from two different streams. Based on this network, the latent feature vector that implicitly represents the orientation of the object is obtained from the RGB image. Experimental results show that the method in this paper can realize the 6D pose estimation of the object and the result accuracy is better than the advanced method (Sundermeyer et al. in proceedings of ECCV, 2018).

With the increasing precision and complexity of high-density interconnect integrated circuit (IC) substrates, automated visual inspection encounters significant challenges in accurately detecting etching defects on metallographic substrate images. Factors such as grayscale variations, noise interference, and rich textures further complicate the process. To address this issue, a novel detection method based on differential geometry theory is proposed, encompassing defect detection between circuits and on circuits. Firstly, the variational Chan-Vese model and morphological closing operation are employed to obtain highly accurate substrate segmentation images. For defect detection between substrate circuits, contour regions between circuits are extracted by differencing the original image with the segmented image. Next, a lightweight compressed MobileNet (CMNet) network is constructed using depth-weighted compression to rapidly identify defect regions between circuits. For defects on substrate circuits, the contour of the segmented image is utilized to determine candidate regions of etching defects by evaluating abrupt changes in angles between adjacent contour points. Subsequently, the proposed discrete curvature calculation method based on the Frenet frame of differential geometry theory is employed to detect and measure defect candidates on the circuits. Experimental results demonstrate the effectiveness of the proposed method in detecting etching defects, outperforming other advanced techniques in screening and identifying defect regions.

针对锂电池极片表面的痕类缺陷检测准确率低、误检率和漏检率高的问题,提出了一种基于局部最优化的随机抽样一致性(locally optimized random sample consensus,LO-RANSAC)的痕类缺陷检测算法.首先,针对锂电池极片表面存在的椒盐噪声、大噪点多的问题,提出了一种改进的自适应中值滤波和基于连通域的滤波算法.其次,针对检测痕类缺陷准确率达不到预期以及误检率漏检率较高的问题,引入一种局部最优化的RANSAC算法.最后,给出了一种基于LO-RANSAC的痕类缺陷分类方法.实验结果表明:本文所提算法相较于标准RANSAC检测准确率提高了 5.9％,相较于基于卷积神经网络算法准确率提高了 15％,达到了 98.2％;多种算法中本工作算法对于痕类缺陷的检测误检率和漏检率最低;平均检测速度较标准RANSAC算法提高了1.7倍,每秒钟检测的图片数量FPS(frame per second)达到12.49.本工作算法具有较高的检测准确率、较低的误检率及漏检率,检测速度达到实时检测要求,因此可满足锂电池极片表面的痕类缺陷检测需求,解决了锂电池极片表面痕类缺陷自动检测难题.

针对目前锂电池极片表面存在亮度不均、低对比度微小缺陷难以检测的问题,提出了一种基于小波增强与Canny算法融合的锂电池极片缺陷检测方法.首先使用K-近邻均值滤波抑制图像背景噪声,然后基于小波变换分别采用线性调整和多尺度细节增强方法处理图像的低高频分量,进行图像增强,接着利用PSO-OTSU算法自适应获取增强后图像的最佳高低阈值,最后利用哈夫检测法连接边缘点.通过测试漏金属、亮点、划痕、孔洞等缺陷各700张图片,定量分析比较了 3种算法的准确率,实验结果表明,相对于其他两种算法,本文算法可以较好地保留缺陷边缘细节,检测低对比度微小缺陷,提取精确完整的缺陷轮廓,检测准确率达97.85％,具有一定的实用价值.

针对锂电池极片表面出现的一些微小对比度低的缺陷,提出了 一种基于改进的多尺度小波变换的锂电池极片缺陷检测方法.首先使用 自适应的伽玛校正,解决锂电图像光照不均的问题,增强图像的对比度;其次对图像进行小波分解,得到图像的高频分量与低频分量;然后利用模极大值算法处理高频分量,得到高频分量的边缘信息,利用多结构多尺度的数学形态学算法处理低频分量,得到低频分量的边缘信息;再通过叠加运算融合低频和高频的边缘信息,得到原始图像的边缘信息;最后提取缺陷轮廓,并将其标记分类.实验结果表明,该算法可以更好地保护缺陷的边缘细节,完整精确地提取缺陷轮廓,具有一定的抗噪性,可有效地检测出锂电池极片表面上的缺陷.其检测准确率可达98.2％,每张图片耗时也更少,为20.2 ms.

The ankle rehabilitation robot is essential equipment for patients with clubfoot and talipes valgus to make up for deficiencies of the manual rehabilitation training and reduce the workload of rehabilitation physicians. Based on the physiological model of the ankle joint and the requirement of rehabilitation in physical therapy, an ankle rehabilitation parallel mechanism has three limbs with a universal joint, rotation joint, and spherical joint configures (3-URS ARPM), which had 6-DOF was analyzed and developed. The inverse kinematics problem of 3-URS ARPM was then solved using GRG optimization methods combined with the Banana objective function. As a result, six control solutions of the inverse kinematics of 3-URS ARPM are obtained. Furthermore, the forward kinematics problem is also analyzed through optimization approaches suitable for motor position control. Finally, the kinematic control characteristic of joints variable for 3-URS ARPM is presented in detail, comparing its motion range to the ADAMS software. The numerical simulation results showed an excellent smooth trajectory tracking in real-time control, indicating that this mechanism for ankle rehabilitation with a simple structure has precise control characteristics with the accuracy achieved is . Hence, the developed 3-URS ARPM can be applied to ankle rehabilitation widely.

针对传统的染色机助剂配送系统配送误差大的问题,提出了一种基于推荐预停值和预计用时预测的多层全连接神经网络模型.首先,使用配送过程记录的数据训练网络模型;然后,将需要配送的数据代入训练好的网络模型进行计算,得到推荐预停值和预计用时,推荐预停值与经验预停值按照可变比例算法计算最终预停值,系统根据最终预停值决定配送阀的关闭时机,利用预计用时评估配送过程是否超时.最后,使用4种预停模式各进行1000次的助剂配送验证实验.结果表明,采用神经网络模型预测模式的配送误差的标准差为23.8 g,平均绝对误差为16.1 g,优于其他3种预停模式的配送误差,取得了较好的助剂配送精度.

Aiming at the line defect detection of a flexible integrated circuit substrate (FICS) without reference template, there are some problems such as line discontinuity or inaccurate line defect location in the detection results. In order to address these problems, a line feature detection algorithm for extracting an FICS image is proposed. Firstly, FICS image acquisition is carried out by using the appearance defect intelligent detection system independently developed in our lab. Secondly, in the algorithm design of the software system, the binary image of the line image to be segmented is obtained after the color FICS image is classified by K-means, median filtering, morphological filling and closed operation. Finally, for an FICS binary image, an image segmentation model with convexity-preserving indirect regular level set is proposed, which is applied to extract the line features of an FICS image. Experiment results show that, compared with the CV model, LBF model, LCV model, LGIF model, Order-LBF model and RSF model, the proposed model can extract line features with high accuracy, and the line boundary is smooth, which lays an important foundation for high-precision measurement of line width and line distance and high-precision location of defects.

The harmonic drive (HD) is increasingly used due to its outstanding advantages, especially for equipment requiring high precision and a large gear ratio. This research is based on the Finite Element Method (FEM) and mechanical analysis to investigate stress calculation methods for the flexspline (FS), a significant component of HD transmission, providing a basis for assessing the durability and failure of the FS. First, the FS deformation and stress analysis model based on the FEM is proposed. Since the two boundary elements of the model have the form of a rotation and translational joint, which have not been provided, based on the stiffness matrix of the fixed beam to build the stiffness matrix for these elements, which completes the proposed analytical model. Then, the structures of an HD reduced are designed and applying the proposed force analysis model, the deformation and stress distributions of the FS are obtained. Finally, the theoretical safety factor is calculated and analyzed. The results of the simulation completely coincide with the theoretical calculation and show that the most concentrated stress is at the root of the teeth, which is the leading cause of the FS failure. So, the study as reference material for rapidly evaluating the stress state and calculating the safety when designing and optimizing the FS.

电极作为锂离子电池的重要组成部分,其质量直接关系到电池的电化学性能、安全性能及使用寿命.锂离子电池电极的生产环境复杂且缺陷形态多变,针对电极图像拍摄过程中易出现的亮度不均以及缺陷对比度低等问题,提出一种基于拓扑滤波与改进Canny算子的电极缺陷检测算法.首先采用基于拓扑学原理的滤波模板抑制图像背景噪声,针对图像整体灰度值过高的问题,应用灰度变换矫正图像灰度分布,增强图像对比度;其次考虑到背景细小纹理较多且边缘信息较弱,提出一种改进Canny算子,采用双边滤波降低图像噪声的同时有效避免边缘模糊,多尺度增强算法改善光照不均增强图像细节;应用四方向的Sobel算子计算梯度幅值和方向以获取更多的边缘信息,提升边缘定位的精确度;基于最大熵算法自适应确定高低阈值,避免边缘点判定的局限性,提高边缘连接的效果.实验研究结果证明相对于其他算法,该算法可以较好地保护边缘细节,精确完整地提取出缺陷轮廓,表现出更优越的准确率和抗噪性,具有一定的实用价值.

In industrial robot design, selecting a motor and reducer is one of the essential elements in the design process. The typical mechanical reducer type is a gear train or planetary gear set. Unfortunately, the gear train and the planetary gear have a meager gear ratio, while their weight is too heavy, the transmission repetition accuracy is low and has a high backlash. On the other hand, the harmonic reducer system developed from the transmission gear has an elastic strain wave, which gives a better gear ratio per weight than those two gear systems. In this paper, a harmonic gear reducer is designed according to the transmission requirements and applied rotation to the base of a Puma 560 industrial robot. The harmonic drive is designed under gear physical requirements and robot techniques properties. The required gear ratio of the rotary joint of 1/50, gear geometry design, manufacturing, calculation, and simulation of strength, the safety factor, and experimental results have verified the designs.

During the LED coating process, the flow of the phosphor colloid at the spray gun has a great impact on the correlated color temperature. It is difficult for the PID control method to use the flow characteristics of non-Newtonian fluids of the fluorescent colloid to achieve the accurate and stable control of the flow, and it has been adjusted by expert knowledge and trial and error methods. To improve the luminous performance, an RBF neural network (RBFNN) of fractional-order PID control algorithm based on sparrow search algorithm (SSA) is proposed in this paper, referred to as SSA-RBFNN-FOPID. Firstly, the parameters of the RBF neural network are optimized to improve the training speed and accuracy by using the sparrow search algorithm. Secondly, the optimal parameters of fractional-order PID are tuned online by using the optimized RBF neural network. In the end, experimental results show that the proposed algorithm has excellent dynamic and static performance, stronger robustness and anti-interference ability compared with other advanced algorithms, which is conducive to more accurate and stable control of the flow of fluorescent colloid during the coating.