基于SOM神经网络的焊球缺陷检测方法研究

摘要现如今的微型电子封装中越来越多地在运用倒装焊这种手段，不过因为焊球总是夹在芯片和基底的中间，所以要想对焊球的缺陷实行有效可靠的检验测量十分困难。这篇文章使用了超声波扫描这种手段对想要测量的芯片进行了焊球缺陷检验测量实验，得到了包含有焊球信息的图像，然后对焊球图片进行图像分析处理这方面的手段的探讨，其中包含了相关系数对比分割算法,和以数理分析等为依据的特征提取。72666

自组织竞争(Self-Organizing Maps, SOM)算法是一种同种类聚集的和多维可观测的无监督的学习算法,网络拓扑结构是由一个输入层和一个输出层组成的。经训练后的神经网络可以储存和它训练过程有联系的知识,能够直接学习曾经发生的故障所表达出来的信息,能够依据算法对象的往常的数据来锻炼神经网络,之后把这些往次的信息与现在的测量出来的数据分析对比,从而达到确定故障类型的目的;神经网络于过滤除去噪声并且在有外界干扰的情况下也可以得到正确的结果,所以可以通过锻炼人工神经网络,使其能在有外界干肉的环境中有效率地工作来识别故障信息,这种滤除外界干扰噪声的特性，使 SOM 神经网络非常合适做在线故障检验测量和识别诊断;

毕业论文关键词：超声波扫描图像处理焊球特征 SOM 神经网络

Defects inspection of solder bump using SAM technology and SOM neural network

Abstract Nowadays flip chip technology is more and more widely used in the microelectronic packaging, but it is difficult to detect the defects effectively due to the solder balls which are hidden between the chip and the substrate。 The through the technique of ultrasound scanning chip to be tested were solder ball defect detection experiments, solder ball picture information, then has carried on the image processing technology research, including to gray image gradient value as threshold segmentation algorithm and based on the mathematical statistics based feature provided take。

The self-organizing map algorithm is an unsupervised learning algorithm for clustering and high dimensional visualization。 The topology of the network is composed of an input layer and an output layer。 The trained neural network can store information about the process of knowledge, learn directly from the history of the fault information, can according to the everyday objects of historical data to train the neural network, then this information and the measured data are compared, to determine the type of fault; neural network with noise filtering and in noise ability to draw the correct conclusion, artificial neural network is trained to identify the fault information, so that it can work effectively in a noisy environment, the noise filtering ability of the artificial neural network for online fault detection and diagnosis;

keywords: ultrasonic scanning image processing welding ball characteristics SOM neural network

摘要 I

Abstract II

目录 III

图清单 V

1 绪论 1

1。1 课题的研究背景及意义 1

1。2 倒装芯片手段 2

1。3 倒装芯片缺陷检验测量方法 3

1。4 本文主要工作 6

2 缺陷检测系统基于SOM神经网络的焊球缺陷检测方法研究:http://www.youerw.com/jixie/lunwen_82696.html