Development of a new cluster index for wafer defects

Defect number and defect clustering are two key determinants of wafer yield. Preventing and detecting wafer defects thus is an important issue in integrated circuit manufacturing. Defect clustering tends to grow with increasing wafer size. Methods have been developed for assessing defect clustering on wafers. However, these methods require either statistical assumptions regarding defect distribution or complex computations. This study develops a new cluster index, utilizing the rotating axis technique from multivariate analysis to accurately quantify defect clusters on a wafer. The developed defect-clustering index does not require making assumptions regarding defect distribution. Thus, the proposed method can be efficiently used by engineers with little statistical background. A simulation experiment is conducted to demonstrate the effectiveness of the proposed defect-clustering index.     

Testing local dependence of spatial structures on images

Associations between two spatial processes can be due to a real dependence between the two processes or to the dependence on common underlying variables. We propose to test the existence of a real dependence by use of local tests, leading to a global test of real dependence and a map of local interactions. We present first how classical interaction tests based on random rotations between completely observed processes such as those developed by Berman (Berman. Appl. Statist . (1986) 35 , 54–62), can be integrated in local analyses. For this purpose, tests are first performed locally, and the distribution of their p -values is then compared to the corresponding value under the null hypothesis. A similar approach is proposed to test non-stationarity of a point pattern by using distance statistics popularized by Diggle (Diggle. Statistical Analysis of Spatial Point Patterns . (1983) Academic Press, New York). The problem of testing the interaction between a random field and a censoring area pattern process is discussed and an approach similar to the preceding ones is then proposed. The methods are mainly applied to agricultural examples but they can be applied to any microscopical images for which one wishes to analyse the spatial structure.     

Multiple defect diagnostics of gas turbine engine using SVM and RCGA-based ANN algorithms

An artificial neural network (ANN) based on the real coded genetic algorithm (RCGA) has been used with the support vector machine (SVM) for developing the defect diagnostics of the turbo-shaft engine of an aircraft. Nonlinearity increases due to the ascending number of input data in the off-design region. If the ANN algorithm is used by itself to determine defects under this condition, the possibility of falling in the local minima becomes high because of the large amount of learning data. To solve this problem, the expanded multi-class SVM has been used to reduce nonlinearity of input data. The RCGA, which is effective to search the global minima, has been applied to the ANN algorithm to obtain the magnitude of defects. As results, the number of learning data has been decreased and convergence and accuracy have been improved.     

基于集成自适应欠采样的铝管表面缺陷检测方法研究

铝管作为一种常见的传输零件,对其表面缺陷进行检测是保证其生产质量、运行安全的必要措施。基于机器视觉的铝管表面缺陷检测方法因其检测精度高、速度快等优点,已取代人工检测,成为主流检测方法之一。但由于缺陷样本与背景样本之间分布不平衡,导致分类器决策边界偏移、检测精度下降,限制了其应用范围。针对这一问题,提出一种基于集成自适应欠采样的铝管表面缺陷检测方法,首先利用支持向量描述方法对数据分布间的重叠区域进行识别,其次通过构建样本局部密度关系自适应确定欠采样对象及数量,最终利用随机空间生成技术同时对数据样本空间和特征空间进行优化。试验结果表明,所提方法在铝管表面缺陷数据集上识别精确率达到98.52%,优于其他先进检测方法。     

增材制造材料缺陷表征及结构完整性评定方法研究综述

金属增材制造过程中不可避免地会产生气孔和未熔合缺陷。尽管采取参数优化和后热处理能够在一定程度上降低缺陷水平,但至今尚无有效方法予以完全消除。这些缺陷作为典型的应力集中源,会诱导疲劳裂纹形核,从而大幅降低材料的疲劳强度和寿命,被视为增材构件可靠性服役的“顽疾”。从静态缺陷表征、动态缺陷演化、缺陷分级、缺陷-疲劳强度设计方法以及缺陷-疲劳寿命评估技术等五个方面论述增材制造缺陷与疲劳行为的研究进展。重点介绍借助X射线成像技术开展缺陷特征及演化的三维、无损、可视化表征与定量统计方法;进一步地,论述基于同步辐射光源的原位力学和疲劳测试系统及表征方法及其在原位、无损、实时、动态追踪缺陷或者裂纹演化机制方面中的应用;增材缺陷具有全域分布、形态多样、尺寸跨度大等特征,总结六种缺陷等级判断方法;在缺陷容限和损伤容限框架内,建立基于材料表面/亚表面/内部缺陷特征的疲劳强度和寿命评价方法。最后,指出借助数据驱动的高通量试验平台和机器学习算法、多尺度多物理场数值模拟是实现增材制造材料工艺设计-缺陷表征-性能评价一体化研究的重要研究课题。     

稳定运动物体视频的特征方法

将背景特征块配对和直方图聚类运动矢量滤波相结合,提出了一种稳定运动物体视频的方法。用得到的全局运动参量补偿摄像机的帧间运动,运用帧差法分割出前景块和背景块,把参考帧背景上的特征块与当前帧上的特征块配对估计下一轮的帧间运动。用一块对多块的匹配策略,把参考块和以参考块为中心的搜索窗内的当前帧上的最佳匹配块配对,建立稀疏运动矢量场。然后,运用直方图聚类方法滤除矢量场中未分离出的前景块矢量和误匹配矢量。最后,用多个包含运动物体的实际视频序列对提出方法进行测试,并与其它先进稳像算法和技术仿真比较。结果表明:提出方法的全局平均帧间变换保真度可达31.05dB,接近或优于上述先进算法和技术。提出的方法对包含运动物体的视频具有较强的鲁棒性,可以去除帧间高频抖动并有效改善视频质量。     

A hybrid defect detection for in-tray semiconductor chip

The requirements for high-speed and high-precision defect inspection in semiconductor chip are growing rapidly because of the complicated surface in semiconductor chip. Due to manufacturing tolerance of IC tray, the misalignment from the chip positioning shift and rotation are always presented for the application of in-tray inspection. In the beginning, this paper focuses on compensating the positioning shift and rotation of in-tray chip by using the proposed image alignment algorithm before the defect detection. After applying the process of image alignment, a hybrid approach of defect detection is applied to detect the defects of in-tray chip. Furthermore, this hybrid approach simultaneously detects the defects based on its surface by the following two categories: (1) the complicated surface in the circuit and (2) the primitive surface on the bump. As mentioned above, the image alignment strategy and the adaptive image difference method are applied in the detection of complicated surface, and the design-rule strategy is adapted to detect the defects on bumps. Finally, the experimental results show that the proposed image alignment strategy and hybrid approach can accurately and rapidly inspect the defects of in-tray chip. This approach is superior to the traditional template matching in defect detection. In addition, the computational complexity can be efficiently reduced by the proposed hybrid strategy.     

Deep Learning Based Data Fusion for Sensor Fault Diagnosis and Tolerance in Autonomous Vehicles

Environmental perception is one of the key technologies to realize autonomous vehicles.Autonomous vehicles are often equipped with multiple sensors to form a multi-source environmental perception system.Those sensors are very sensitive to light or background conditions,which will introduce a variety of global and local fault signals that bring great safety risks to autonomous driving system during long-term running.In this paper,a real-time data fusion network with fault diagnosis and fault tolerance mechanism is designed.By introducing prior features to realize the lightweight network,the features of the input data can be extracted in real time.A new sensor reliability evaluation method is proposed by calculating the global and local confidence of sensors.Through the temporal and spatial correlation between sensor data,the sensor redundancy is utilized to diagnose the local and global confidence level of sensor data in real time,eliminate the fault data,and ensure the accuracy and reliability of data fusion.Experiments show that the network achieves state-of-the-art results in speed and accuracy,and can accurately detect the loca-tion of the target when some sensors are out of focus or out of order.The fusion framework proposed in this paper is proved to be effective for intelligent vehicles in terms of real-time performance and reliability.     

空间目标识别技术与威胁评估算法综述

简要介绍了现有的空间目标探测技术地基／天基的雷达、光电等,指出目标识别的难点在于高效、准确、实时的数据融合。针对空间环境的复杂性,指出空间目标威胁评估的关键在于属性权值向量的确定,详细分析了现有的威胁评估算法：层次分析、专家系统、灰色聚类、Bayesian网络、云模型、神经网络、支持向量机等;通过对比研究,着重探讨其优势与不足,最后给出此类算法在空间环境下的普适性。     

A honeybee-mating approach for cluster analysis

Cluster analysis, which is the subject of active research in several fields, such as statistics, pattern recognition, machine learning, and data mining, is to partition a given set of data or objects into clusters. K-means is used as a popular clustering method due to its simplicity and high speed in clustering large datasets. However, K-means has two shortcomings. First, dependency on the initial state and convergence to local optima. The second is that global solutions of large problems cannot be found with reasonable amount of computation effort. In order to overcome local optima problem lots of studies done in clustering. Over the last decade, modeling the behavior of social insects, such as ants and bees, for the purpose of search and problem solving has been the context of the emerging area of swarm intelligence. Honeybees are among the most closely studied social insects. Honeybee mating may also be considered as a typical swarm-based approach to optimization, in which the search algorithm is inspired by the process of marriage in real honeybee. Neural networks algorithms are useful for clustering analysis in data mining. This study proposes a two-stage method, which first uses self-organizing feature maps (SOM) neural network to determine the number of clusters and then uses honeybee mating optimization algorithm based on K-means algorithm to find the final solution. We compared proposed algorithm with other heuristic algorithms in clustering, such as GA, SA, TS, and ACO, by implementing them on several well-known datasets. Our finding shows that the proposed algorithm works better than others. In order to further demonstration of the proposed approach’s capability, a real-world problem of an Internet bookstore market segmentation based on customer loyalty is employed.     

Measuring the Quantitative Characteristics of Local Vortices on a Plane of a Deformed Material

A problem for determining the center, size, and spatial distribution of displacements of a vortex on the basis of a plane vector field is under consideration. A seed algorithm is proposed instead of typical contour scanning, and specific circulation is used as a criterion for determining the center of a vortex flow. In order to increase the stability of this algorithm, a parameter playing the role of a threshold is introduced. A Monte-Carlo method is used to study the influence of measurement errors on the accuracy in determining the coordinates of a hard rotation center, and the relationship between the false alarm probability and the threshold level is established. The use of the least squares method for calculating the spatial distribution of displacements is shown. This approach is demonstrated on the example of a real vector field measured in the vicinity of a main crack tip.     

木材表面图像的缺陷分割与类型识别方法

为了识别死节、活节、虫眼三种木材表面缺陷,采用Gabor变换和模糊C均值聚类进行缺陷分割;采用数学形态学运算对分割图像进行了后处理;获取了木材缺陷区域的12维频率能量参数和2维几何形状参数;用支持向量机进行木材表面缺陷类型的识别。采用Gabor变换和模糊C均值聚类方法对死节、活节、虫眼三种木材表面缺陷的分割精度都达到94%以上,支持向量机对缺陷类型分类正确率达到93%以上,这说明本文的方法对木材表面缺陷的分割与识别是可行的。     

A hybrid clustering approach for multivariate time series – A case study applied to failure analysis in a gas turbine

A clustering problem involving multivariate time series (MTS) requires the selection of similarity metrics. This paper shows the limitations of the PCA similarity factor (SPCA) as a single metric in nonlinear problems where there are differences in magnitude of the same process variables due to expected changes in operation conditions. A novel method for clustering MTS based on a combination between SPCA and the average-based Euclidean distance (AED) within a fuzzy clustering approach is proposed. Case studies involving either simulated or real industrial data collected from a large scale gas turbine are used to illustrate that the hybrid approach enhances the ability to recognize normal and fault operating patterns. This paper also proposes an oversampling procedure to create synthetic multivariate time series that can be useful in commonly occurring situations involving unbalanced data sets.     

Improved color feature arrangement for mean shift tracking

In order to reduce redundant empty bin capacity in the probability representation,we present a new color feature arrangement mechanism for mean shift tracking objects.In the proposed mechanism,the important optimal color,or we call it optimal color vector,is clustered by closing Euclidean distance which happens inside the original RGB color 3-D spatial domain.After obtaining clustering colors from the reference image RGB spatial domain,novel clustering groups substitute for original color data.So the new color substitution distribution is as similar as the original one.And then target region in the candidate frame is mapped by the constructed optimal clustering colors and the cluster Indices.In the final,mean shift algorithm gives a performance in the new optimal color distribution.Comparison under the same circumstance between the proposed algorithm and conventional mean shift algorithm shows that the former has a certain advantage in computation cost.     

软性电路板金面缺陷的无监督检测

为实现软性电路板(FPC)金面缺陷的准确自动检测,提出了一种以Gabor滤波器和Mean Shift聚类算法为基础的完全无监督FPC金面缺陷检测方法。首先,用Gabor滤波器组、数学形态学与Gaussian平滑处理抽取待检测图像的多维特征;然后,使用主元分析(PCA)将每个像素特征维数降为二维;最后,使用Mean Shift方法对二维特征数据进行聚类并将聚类的结果转化为二值图像。整个检测过程无需预先知道缺陷的类型和FPC金面的纹理类型,是一种完全无监督的检测方法。对带有各种缺陷的FPC金面进行检测实验,结果表明,该方法能够准确地将各类缺陷区域从背景区域中分离出来,具有自动缺陷检测系统所要求的识别能力强、稳定性高的特点。     

Automatic defect inspection for LCDs using singular value decomposition

Thin film transistor liquid crystal displays (TFT-LCDs) have become increasingly popular and dominant as display devices. Surface defects on TFT panels not only cause visual failure, but result in electrical failure and loss of LCD operational functionally. In this paper, we propose a global approach for automatic visual inspection of micro defects on TFT panel surfaces. Since the geometrical structure of a TFT panel surface involves repetitive horizontal and vertical elements, it can be classified as a structural texture in the image. The proposed method does not rely on local features of textures. It is based on a global image reconstruction scheme using the singular value decomposition (SVD). Taking the image as a matrix of pixels, the singular values on the decomposed diagonal matrix represent different degrees of detail in the textured image. By selecting the proper singular values that represent the background texture of the surface and reconstructing the matrix without the selected singular values, we can eliminate periodical, repetitive patterns of the textured image, and preserve the anomalies in the restored image. In the experiments, we have evaluated a variety of micro defects including pinholes, scratches, particles and fingerprints on TFT panel surfaces, and the result reveals that the proposed method is effective for LCD defect inspections.     

Application of Bayesian networks for diagnostics in the assembly process by considering small measurement data sets

The ability of variation source diagnosis in the auto body assembly process plays an essential role in the success of the manufacturing enterprises. However, it is more challenging to identify the process faults associated with the compliant sheet metal parts based on small measurement data sets. A new Bayesian networks (BN) modeling approach under the condition of small data sets is proposed. The main causal links are identified based on mapping of the variation sensitivity matrix. The interaction effects are detected according to the conditional mutual information tests. After the network structure is determined, the Bayesian approach is used to obtain the conditional probability tables by incorporating prior probability distributions. The evaluation of diagnostic performance concerning evidence number and log-odds noise levels is also presented. A real bracket assembly case was used to illustrate the whole procedures for fixture fault diagnosis. The examined test cases demonstrate the proposed BN approach is practical and effective, even when incomplete evidences are observed and a medium-level noise is present.     

基于形状模板匹配的印刷品缺陷检测

针对传统的模板匹配算法无法满足工业检测中的实时性要求,提出了一种基于形状模板的匹配算法,并应用于印刷品的缺陷检测。算法以Sobel算子滤波后得到的边缘点的方向向量为基础,定义相似性度量,同时结合图像金字塔分层搜索策略,利用形状信息进行模板匹配。实验结果表明,所提出的匹配算法速度快、精度高,而且匹配鲁棒性高,不受印刷品缺陷检测中可能出现的遮挡、非线性光照变化和全局对比度反转等情况的影响。     

Manufacturing performance evaluation for IC products

As we known, the product diversity and complexity in the production line will gradually increase. When the multiple products were alternately produced at the same line, the manufacturing performance will be difficult to evaluate. In particular, traditional process capability analysis and related process capability indexes cannot be directly employed to the IC manufacturing process. As we know, the yield has a direct effect on the manufacturing cost. Hence, yield is frequently used by most IC manufacturers to evaluate manufacturing performance. The diversity of function will become another analytic consideration due to that the component density, wafer area and product complexity of an IC product rapidly increase. Hence, the diversity of function can be regarded as the evaluated factor. Additionally, the defects on a wafer will begin to cluster as the wafer area gradually increases. Therefore, only using the yield to represent manufacturing performance may not lead to an appropriate judgment. In particular, only using the yield to evaluate the process’s stability and the product’s maturity can not provide a meaningful resolution. The primary reason is that the inherent features in the processes or products are not included into analyzing. For instance, even though the defect count, defect size and defect distribution are the same, the yield loss of the complicate manufactured product will be less than that of the simple manufactured product. In this study, we propose a simple performance evaluation index to assess the manufacturing performance in the IC manufacturing industry. This evaluation index is constructed according to a modified Poisson yield model, and the related parameters regarding the process or product (e.g., the minimum linewidth, the area of a die, the number of manufactured process or layer, the degree of defect clustering, and so on.) are taken into consideration. In addition, an integrated evaluation procedure is also suggested to evaluate the performance of the manufacturing of multiple IC products. According to the result obtained from the illustrative example, the index and the procedure can overcome the drawback of separately using yield or defect count in the analysis. The rationality and the feasibility of the proposed evaluated index and the procedure can be verified by demonstrating the illustrative example.     

改进的FCM聚类法及其在行驶工况构建中的应用

针对FCM聚类法对初始聚类中心比较敏感、迭代容易陷入局部极值、难以取得最优聚类的问题,提出了一种改进的FCM方法,即利用SOM网络对主成分数据进行聚类,将得到的权值作为FCM聚类的初始聚类中心,从而使聚类结果更加接近最优聚类。将改进的FCM聚类方法应用于合肥市道路行驶工况的构建中,理论分析及试验结果表明,该方法有效地提高了聚类精度,构建的行驶工况与实际道路的交通状况吻合很好。