机器视觉系统在电子行业中的应用已经有很长一段时间了，它的应用历史几乎可以跟机器视觉行业的历史相媲美了。早期时候，只是用来检验孔中是否存在引线。不过由于线路板组装而成设备通常配有检测引线是否进孔的传感器，而且这种机器视觉的方法无法检测出引线是否进入了正确的孔洞之中，所以这种应用没有被推广。
    几乎同时，众多公司顺应焊接检测的要求提供了能够保证焊接存在而不缺失的系统。不过这只是部分回应。原始的3D技术可以检验焊接的湿度角与表面情况，但是行业没有采用这类系统。原因包括高错误拒绝率，以及专家们在对焊接点评估中视觉品质是否能够真正反映焊点的好坏存在争议，另外一点是所有这些系统都很昂贵。
    这样下来电子组装行业采用了表面贴装设计。凭直觉一些早期的专门应用于线路板组装的机器视觉系统是可用的，但后来实践证明系统需要添加新的性能以提高可靠性同时降低错误拒绝率。表面贴装生产线为机器视觉提供了一系列应用的场所：后焊点焊膏的回流检测、后部件定位、在回流前进行的存在/缺失判断，正确部件识别，精确定位，为了保证焊点或焊接处在经过焊接后没有发生漏焊或焊接不充分而进行的后焊点回流检测。在后焊点的相关应用中，基于光学以及X射线的视觉系统均可使用。
    基于2维和3维的视觉系统的产生是电子组装市场不断增长的需求的产物。在网印中，3维系统具有检测焊膏定位精度以及容量特性的功能。大部分部件定位系统都是2维的，有一些是3维的。视觉和基于X射线的系统均可以对那些应用3维方法检测的后焊点进行检测。在视觉系统中可以对3维表面特性进行检测，而在X射线检测中，焊接处的内部特性可以被检测出来。
    当今，至少有两个原因促使正规的组装印制板的生产商安装机器视觉系统。由于部件本身和内部连接器的体积越来越小，想要获得一个较好的焊点的话，定位精度问题就变得越来越重要。根据RoHS国际标准焊点不允许有导线穿过，这对焊点本身的粘度构成了挑战，同时，考虑到部件的内部连接，对焊点的湿度特性以及相关定位问题的要求也越来越高。
    考虑到生产管理的重要性，机器视觉的应用就更广泛了。3维系统可以用来监测印刷板的平整度，对平整度的要求随着连机器以及相关衬垫体积的减少变得越来越苛刻。在网印中系统可以通过检测屏幕图像来确保图样的正确性与完整性。
    重要的是，2维与3维机器视觉技术的基础技术在电子组装行业的应用变得越来越广泛，功能也越来越强大。相机现在可以提供更强的分辨率、速度以及更好的图像质量。光源随着LED的发展功能也越来越强，而且多方向与时序光源的费用也在降低。而微处理器与计算机技术的进步更使得机器视觉产品的功能变得越来越丰富。
机器视觉产品的功能包括精确度量、色彩识别、光学字符识别、更强大的表面情况描述功能以及CAD文件兼容性。

    了解了这么多之后，我们将从一些为电子行业提供机器视觉系统的供应商那里了解到更为详尽的信息。

1. 你们公司在电子行业的哪些特殊应用中用到了机器视觉技术？
[Stacy Johnson_Agilent]我们为电子行业里的各类不同应用提供了大量的检测方案。我们的光学检测组合能提供3维焊膏检测系统，前、后回流部件检测系统以及3维X射线后回流焊接检测。我们的系统在包括移动电话、服务器、汽车等不同类别的电子部门中都有广泛应用。
[Jim Gibson_Landrex Technologies]我们是全球印刷电路板行业检测方案的提供商。Landrex所提供的应用于电子行业的产品中包含有机器视觉技术主要是下面几种：Optima系列前、后回流检测系统。Optima的7300系列主要对印刷电路板的部件、焊接处、线路板缺陷以及插针顶端等处进行后回流检测。这个机器家族有两类产品：OptimaII7301 Express是低成本高性能的系统；Optima II 7310 Extreme是Express的超高流量版本。
[Shavi Spinzi_Orbotech]我们为SMT生产线的全过程提供检测方案：3维2维相结合的超高速焊膏容量检测、3维定位检测、后回流及焊接处的波检测。我们的产品能够满足大多数装配检测的要求,在计算机、便携式装置等设备的高速、大规模生产中均有广泛应用。

2.你们的哪些应用中既用到了3D技术，又用到了2D技术？为什么？
[JIM] Optima II 73xx系列应用光源和相机的灵活组合来处理焊点的某些3D特性。这一系列有五个相机，分别位于四个角和一个中心上，LED光源也同样能以多种方法放置。相机角度和光源方向角度的结合使得73xx系列能够处理平面和非平面的反射特性。而且系统可以用来检测焊接点是否具有凹、凸或平坦的3D特性。目前，73xx系列不沿着焊点重组点的高度。然而硬件所具有的通用特性使得未来的软件不仅能处理3D特性，还可以重构它们。在不久的将来系统很有可能开发这一特性。
[SHAVI]针对焊膏的检测，Orbotech提供了Symbion P36，它结合了3D/2D传感器可以对焊锡膏进行精确的体积测量。众所周知，高度和体积信息在焊接过程中是关键参数，并对离线生产有着重大影响。在这个模式下，3D传感器实际上补充了2D传感器所能提供的基本信息使得检测范围得以扩大，并为使用者提供了过程的全面信息。对于部件定位，用2D的方法来捕捉部件的有无及位置信息已经足够。对于焊接点的检测，Orbotech提供了带有一个高清晰度顶端相机和四个角度的相机的系统。结合多角度白炽灯，我们可以准确描述焊点的特性，并且看到一些隐藏着的或是部分隐藏的焊点，例如那些在J-脚芯片或是搞密集度的边缘连接器上的焊点。
[STACY] Agilent的焊膏检测和X射线系统都使用3D检测技术。对于焊膏检测来说包含3D体积和高度信息是很重要的，因为体积与连接处的长期稳定性有着密切的关系。如果测量的是2D信息则无法获得那些关键信息。应用X射线的一个主要原因是隐藏焊点的存在，3D技术之所以有用是因为我们的用户创建了双面板。

3．您所提到的每个应用的关键的机器视觉系统性能指标是什么？
[SHAVI]下面几条是至关重要的：
1） 检测的任何阶段，所有机器都要符合SMT生产线的速度并且对所有板子做到100%的检测（非抽样）。
2） 在焊膏检测的应用中，测量的精确度和可重复性是很关键的，要保证Gage R&R<10%。
3） 低故障报警——后回流FPY>90%，粘贴/定位FPY>95%
4） 检测水平>95%——这意味着能够看到包括后回流状态下的上升管脚在内的所有问题。
5） 快速而简便的安装过程。
[STACY]在自动光学检测（AOI）中用户通常用精确度和可重复性作为标准来判断光学检测系统是否符合他们的测量需要。对于自动X射线检测（AXI）来说，低故障报警率是采用和实施该系统的主要驱动因素。
[JIM]机器视觉应用有五个主要的标准：错误拒绝，错误接受，编程时间，速度，当然还有费用。大多数用户在考察市场上的产品时将速度和费用两个参数作为他们的首选指标。然而，错误拒绝，错误接受和编程时间对用户的生产成本的影响不亚于速度和系统价格。
    用于前、后回流光学检测的Landrex Optima系列机的目标是将错误接受和错误拒绝率控制在百万分只二十至五十之间，并且在用户长时间在线使用该系统的过程中保持不变。Optima系列机器速度很快，特别是用于后回流分析的Optima II 7310速度更快。每一种类型的机器有2-4小时的编程时间。最后，每台机器的价格在10万以下。
    Optima 7210 OPT系统有附加性能指标。由于它既是一个测量系统又是一个故障显示器，它遵循严格的精确度和重复性标准。大体上来说，测量系统的精确度和可重复性必须是生产扳子的机器的十倍以上，依据这种标准选择和定位系统，以保证提供过程有用的测量信息。例如一个0402的系统，精确度必须在+-100微米之间而可重复度在正负10微米之间。对于一个0201，要求则更为苛刻。总的来说，这类标准就是所谓的10%GR&R。

4．那些在电子行业中应用的机器视觉系统的基本技术经历了怎样的变革才使得系统的性能有了显著的提高？
[STACY]在我看来，最引人注目的是相机技术的变革，它使得AOI产品的检测速度和精确度的得以不断提高。
[JIM]在过去的5-8年中，我们看到机器视觉系统的硬件水平有了显著的提高。二十年来，机器视觉领域的工程师们一直被相机性能的有限性、照明的选择以及慢速计算机存储空间的局限性等问题困扰着。由这些硬件组装起来的机器只能用最基本图像处理技术来解决问题。从2000年开始，硬件部分出现了更新，更快、更便宜、更可靠、更大的矩阵相机。另外，彩色相机由于具有了与黑白相机相同的性能指标也逐步成为一种可选方案。计算机速度和内存容量在以惊人的速度提升，而价格又在飞速下降。最终导致光源行业的迅速发展。随着LED技术的到来，特别是白光LED，它从成本和速度的角度来讲很可行。所有这些硬件方面的进步使得机器视觉得以更好的发展。
[SHAVI]AOI系统是一个融会了光学，硬件，软件等多方面知识的机器。这些技术的进步使得我们可以满足电子行业检测不断增长的需求。具有更快的图像速率的相机保证了图像的高质量，同时高性能的图像采集硬件使得数据读出速率和之后的执行时间（这曾是3D AOI的瓶颈）有了惊人的改善。光源技术的进步也使得照明强度和照明环境控制能力有了飞速提高。最后，像JAVA这样新的技术平台对我们在硬件上运行程序以及减少产品从研发到市场所需的时间方面有着极为深远的影响。

5．电子行业中使用的机器视觉系统的哪些基础技术的重大突破将会促使机器视觉系统在未来三年中的进一步发展？
    下一场革命将会发生在这些系统的中枢部分，也就是说由计算机视觉算法引发。新的算法使我们能够更轻松更有把握地完成现有任务，同时还能处理那些以前根本无法想象的新任务。例如，凭借计算机视觉算法，我们现在能够提供有关部件定位的多样化的数据，而非仅仅给出故障分析。多样的数据能够使得PCB生产的过程控制领域发生了根本性的变革。在过去五年里硬件的演变将导致软件在未来五年中的不可阻挡的重大变革。
[SHAVI]多核处理器将使数据处理发生重大革新，并且为机器性能水平的改进提供驱动力，某些原来只能顺序执行的任务可以并行完成。
[STACY]与上面看法相同，我相信相机技术的革新将会进一步推动AOI在未来三年中的技术发展。

6 电子行业市场的哪些变化对机器视觉的应用具有推动做用？例如，RoHS规范有什么影响？
[SHAVI] AOI正以7-10%的增长率平稳健康地增长。RoHS的引入无疑影响了AOI的应用，但还不是主要的决定性影响因素。而扮演决定性因素的是施加于公司之上的与日俱增的包装的微型化和法律方面的压力。微型化的底线是人眼无法应付的，工业依靠机器视觉提供的可靠方法来完成或取代人工视觉检测和电子测试。立法，包括RoHS，正促使制造商们的操作更为透明化。在这个领域，AOI在视觉领域和它在SMT过程中提供的质量数据上有着很大的优势。
[STACY] SMT板上的包装/仪器的尺寸的缩减推动了AOI和AXI的应用。另外，对于AXI，可用性是个问题。内部含有隐藏结点的BGA和CSP包装的普遍性和使用率的增加驱动了AXI的需求。我相信无引线是对于机器视觉应用是一个驱动因素，如果你使用机器视觉（AOI 或AXI），可以使进程更快达到稳定。
[JIM]结果是检测印刷电路板变得越来越困难。随着装置的沟槽变小，板子变得紧密复杂，材料从有引线变为无引线，仅仅依靠人工检测，ICT，和功能测试将会降低效率。另外，大多数公司从每个板子上获得的利润正在减少，结果是，大多数的公司看到了通过使用100%光学检测来提高效率，降低生产成本，提高质量，并提高市场占有率的重要性。

7．这些改变对于应用机器视觉系统进行的检测提出了哪些新的要求？
[STACY]我不相信无引线装置是促使机器视觉检测要求变化的驱动因素，小型化的确起到了这个作用。对于精确性和可重复性的要求随着装置的变小而不断提高。
[JIM]光学检测系统需要通过进一步的发展来满足这些要求。硬件和软件都需要有更小型的部件，了解新材料的变化，有足够的马力来满足生产线的速度，并且能够在没有维护的条件下可靠地运行。机器需要在无人干预的情况下产生较少的错误报警和错误接受。错误接受意味着板子有可能在现场出故障。错误拒绝会导致对机器输出的昂贵的人工确认成本。错误标志和错误逃逸可能会对许多公司的经济收益有重大影响。底线是机器必须自己适应正在生产之中的产品的视觉特性的变化。虽然这些听起来很简单也很符合逻辑，实际上想要设计这样一个对新环境有高度的自适应性并提供可靠的检测结果的系统却很难。
[STAVI]这样就存在一个矛盾，机器必须变得更加复杂并具有满足这些要求的功能，而另一方面使用者却没有时间和图像处理的知识来利用和开发这种功能。检测机器因而需要具有高度的智能化，最关键的是使具有不同技术水平的操作者能够轻松地学会控制机器。与数据处理及管理的功能相同，AOI会采用一个具有预测功能的方法来进行检测，这种方法通过内建的智能功能来识别和预测过程的变化。我们正进入一个自动化检测时代。

8．作为电子行业的机器视觉系统提供商，你们在机器视觉系统营销中遇到哪些挑战？
[JIM]我们所面临的最大的挑战就是让顾客明白我们的产品与其他对手的产品之间的区别。实际上，任何一个人拿一个相机和一个闪光灯就可以组装成一个AOI系统。每年都有“自家生产”的AOI系统投放市场，而且也有人因为这些产品低廉的价格而购买它们。令人惊奇的是，人们几乎可以用任意一个AOI系统进行最基本的检测。然而，如果要求不论应用于哪种板子，无论小批量还是大批量，都能达到国际标准，且错误率控制在20-50ppm，则需要做大量的设计工作。要想使机器在99.9%的时间里都能自动完成检测任务是件很难的事。我们人类可以在不同的环境下完成视觉识别任务，不过，我们的眼睛可是经历了数万年的演变的，而且我们的大脑更是如此。
    而在顾客方面，他们对机器所做的短期评估很难反映出技术的优劣，因为几乎所有的机器在短期内的性能都会比较好，而长期的性能是很难在短期测试中显现出来的。我们的产品是20年设计、科研与实践经验的结晶。我们对产品的检测延伸到了顾客方。我们希望看到我们的产品在顾客实际使用中仍然能够保持优良的性能。我们将顾客方面得到的数据作为我们销售理念的一部分，因为让人们理解测试所得的结果与长期使用得到的结果的差别所在是很重要的。
[Shavi]基于当今检测的要求以及未来的挑战，我们将重心放在创新以及新技术的采用上。市场上存在的问题是如何区分两种技术，一种是有局限性并且会对将来的应用产生瓶颈作用的技术，而另一种则是建立在新技术基础上并且可以适应各种新的需求的系统。在我们的营销策略中，我们将重心放在AOI的四个基本元素上，分别是：易用性、检测能力、速度和精确度。
[Stacy]一个最大的挑战存在在意识层面上。有趣的是，即使Agilent有组合方案，很多人并不知道我们卖3D焊接检测系统。

9. 你对机器视觉在电子行业未来的发展有何想法？
[SHAVI]我相信机器视觉会在生产线以及各种过程中占据越来越重要的地位，并且会成为电子制造业中常见的设备。由于电子产品开始向硅衬底型转变，组装线上的可视部件越来越小，虽然在传感器技术没有显著突破之前X射线可以解决这个问题，但是对于AOI检测来说X射线还一个很昂贵的选择。如果费用继续成为主要的限制因素，用机器视觉控制前焊接过程。这就需要对从不同机器获得的数据进行智能化的整合，从而更好的优化过程。
[STACY]随着无引线技术的采用以及更新、更小的封装的产生，我们将会很快看到闭环系统在挑拣、定位、AOI、显示屏打印与焊接检测中的普遍性应用。诸如AOI与AXI之类的机器视觉应用将成为生产线的标准。AOI与AXI功能的不断提升将会影响故障预防环节，从而影响整个过程控制。故障包含保证机器视觉能够帮助产品升值而不仅仅是一个昂贵的检测系统。
[JIM]电子行业中有四种趋势正在流行着。第一种趋势是可以建立更多更复杂的系统。第二种趋势是这些机器的要价正在直线下降。第三种趋势是第二种趋势产生的结果，就是工厂一直在寻找最廉价的劳动力资源。最后一种，为了与快速更新的产品相适应并且使工厂能够满负荷运转，每个工厂正在生产更多的产品组合。结果就是我们看到那些没经验的工厂正在以超低价格生产更为复杂的产品线组合。这样的情形需要在每一级上都设有自动系统，包括电子器件的光学检测。因此，人工干预的情况越来越少。另外，机器视觉系统不仅被用于故障检测，还被整合起来为工厂提供正确生产产品的过程信息。
10. 你打算给那些考虑购买用于电子行业应用的机器视觉系统的公司什么建议吗?
[JIM]很令人惊奇的是，人员成本是AOI机器的隐藏费用。最大的问题是人们没有意识到印刷电路板、部件以及焊点在板子上的视觉形状会发生巨大改变。光学检测系统必须能够处理这个自然发生的可以接受的改变并且能够准确地将其与真正的缺陷区分开。大多数AOI系统要求编程人员列出所有可能发生的形变并且指导它如何检测出这类形变。如果有新的形变类型产生，编程人员必须将其添加进去。我们假设总有新的形变发生，大多数机器需要全天候监控并且要有一个熟练的编程人员维护从而保持错误拒绝与错误接受率保持在可接受的范围之内。如果编程人员不能够解决问题，公司就会因为派专门人员辨别错误警报而花掉一部分钱。大多数人没有意识到购买机器后所需付的员工成本。这种人员配置可以将AOI系统的成本提高100%。我们有一个建议提供给顾客们，那就是评估一下AOI设备在没有编程人员的辅助之下自动处理PCB板上的可接受的形变的能力如何。
[Shavi]公司通常将精力放在能够帮助他们处理临时问题的测试方面，而很少关注未来环境的变化。当我们跟用户聊天时，经常会碰到这样的情形——之前购买的AOI由于最终无法满足需要而被扔在一边。考虑一下如果平台有能力适应新的问题，例如新的部件形式或者板子的设计，结果又将如何。再想想你们的员工流动与技能。似乎去研究个别部件的算法的确有优势，但是这种知识能延续下去吗？最后，如果正处于扩张阶段，你们的机器能满足不断增长的要求吗？
[Stacy]寻找支点。像Agilent公司所提供的能够制定测试战略并加以执行的完整的组合方案，将会为你的生产线提供最好的测试方案。从Agilent公司获得一个全程解决方案还有一些隐性的成本减少，例如不需要培训，因为在组合系统操作起来与一般系统没什么两样。较低的基本设施费用与一般性花销可以帮助制造商省钱。与其为了某个方案费尽心思在诸多提供商之间进行评估到头来省的钱又不多，不如投资于一个组合方案，这样你的费用节省可以最大化并且具有很好的适应性。

英文原稿（摘自AIA官方网站）：

Machine Vision-Based Inspection for Electronic Assembly
By Nello Zuech, President, Contributing Editor
(Posted 06/27/2006)
Application-specific machine vision-based systems addressing applications in electronic assembly have been around for a long time, almost as long as the machine vision industry itself. Early on, the requirement was to verify the presence of a lead in the appropriate hole. This was not widely accepted because the board populating equipment often came with sensors able to detect if the lead did not go into a hole. Significantly, it did not detect if it went into the correct hole.
About the same time companies responded to the requirement for solder joint inspection by offering systems capable of assuring the solder was present and not shorting. This, however, was only a partial response. Primitive 3D techniques emerged capable of verifying wetting angle of the joint and surface conditions but the industry did not embrace these systems. High false rejects were experienced and, in the case of solder joint evaluation, experts debated whether the visual conditions really reflected a good joint. Furthermore, all these systems were very expensive.
Along the way the electronic assembly industry embraced surface mount designs. While intuitively some of the early application-specific machine vision systems targeting board assembly applications might be suitable, over time it has been determined that the systems had to be adapted with new features to improve reliability and reduce false rejects. The surface mount line offers a number of opportunities for vision systems: post solder paste verification and inspection; post component placement and before reflow presence/absence verification, correct component verification, placement accuracy; post solder reflow inspection to assure correctness of placement and presence/adequacy of solder and solder joint itself and no shorts caused by soldering; and post-wave to verify solder joint integrity especially on lead-through-hole-based components. In the case of post-solder applications both optical and x-ray-based vision systems are now available.
Both 2D and 3D-based vision systems have emerged in response to the requirements of the electronic assembly market. In the case of screen-printing, 3D systems offer the ability to inspect the solder paste both for placement accuracy and volumetric properties. While most component placement systems are 2D, some are 3D. Both visual and x-ray-based systems are available for post-solder inspection that perform 3D inspection. In the case of visual systems, the 3D surface properties are inspected and in the case of x-ray, the 3D internal properties of the solder joint are inspected.
Today there are at least two drivers that require all serious assembled board manufacturers to deploy these application-specific machine vision systems. The components themselves and the interconnects are becoming smaller making placement accuracies to obtain a good solder joint ever more demanding. And the response to RoHS international standards eliminating lead from the solder has resulted in changes to the viscosity of the solder itself presenting challenges to wetting properties and corresponding placement of the solder with respect to the component interconnect.
Given the importance of yield management, machine vision is finding more applications. 3D systems exist to monitor board flatness, which is ever more critical given the smaller interconnects and corresponding pads. Systems are also now available to inspect the screens used in screen-printing to assure pattern correctness and completeness.
Significantly, the underlying technology that is the basis of both 2D and 3D machine vision is becoming ever more relevant for applications in the electronic assembly industry as well as more powerful. Cameras now offer greater and greater resolution, speeds, and better color properties. Lighting is more capable with the developments associated with LEDs making multi-directional and sequential lighting more cost effective. Needless to say the underlying microprocessor and compute technology makes it possible to perform more compute intensive algorithms leading to more rigorous performance.
The result is that the capabilities now available include precision metrology, color-based recognition, optical character recognition, better characterization of surface conditions and CAD file compatibility. The result is better detection with fewer false rejects and systems easier to train on new board designs and interface to rework stations.
With all this in mind we canvassed input from many of the suppliers of application-specific machine vision systems addressing assembled board applications in the electronic industry.
The following provided the responses to our questions:
· Stacy Johnson, Product Marketing, Agilent Technology
· Jim Gibson, President, Landrex Technologies, Inc.
· Shavi Spinzi, Director of Marketing, Orbotech
1. What are some specific applications in the electronic industry that your company addresses with machine vision technology?
[Stacy Johnson ?Agilent] - We offer a broad range of inspection solutions for a broad range of applications in the electronics industry.  Our optical inspection portfolio supplies 3D solder paste inspection systems, pre and post reflow component inspection systems and 3D X-ray for post-reflow solder inspection.   Our systems are utilized in many different electronic sectors including mobile phones, servers, automotive, etc.
[Jim Gibson - Landrex Technologies] We are a global test solution provider for the printed circuit board industry. Landrex抯 offerings in the electronic industry that are addressed with machine vision technology are the Optima series of pre- and post-reflow inspection systems. The Optima?7300 series is targeted at post-reflow inspection of all aspects of a printed circuit board including parts, solder joints, board defects, and pin-tips. This machine family has two offerings, the Optima II 7301 Express? which is a low cost, high performance system and the Optima II 7310 Extreme? which is an extremely high throughput version of Express. On the in-process side of the oven, Landrex offers The Optima?7210 Optical Process Test system (OPT). This system performs post-place and 2D post-paste inspection and measurement and provides statistical process control information.
[Shavi Spinzi - Orbotech] We offer inspection solutions for all stages of the SMT production line: ultra fast combined 3D/2D for paste volumetric measurement, 3D for placement inspection, post reflow and wave inspection of solder joints. Our products are designed to cover the demands across the whole assembly inspection spectrum from high-speed mass production of gaming, computer and portable devices to highly flexible production in the mid-sized EMS segment.
2. Which of your applications uses 3D techniques as well as 2D techniques? And why?
[Jim] The Optima II 73xx series uses a very flexible combination of lights and cameras to assess some characteristics about the 3D profile of solder joints.  This line of machines has five cameras, four angled and one center, as well as a 揹ome?of LED lights that can be configured in any number of ways. The combination of angles of cameras and angles of lighting directions allow the 73xx series to assess the reflectance properties of both flat and non-flat surfaces. In particular, the system is able to check solder joints to see if they have certain 3D properties such as being convex, concave or flat. At present, the 73xx series does not reconstruct the height of points along the solder joint. However, the generic composition of the hardware allows for future software development that would not only assess the 3D properties, but also to reconstruct them. At some point in the future, the system may exploit this ability.
[Shavi] For solder paste inspection, Orbotech offers the Symbion P36 with a combinational 3D/2D sensor for accurate volumetric measurement of solder paste. It is widely accepted that height and volume information is a critical parameter in the paste process that has a strong influence on end-of-line yields. In this mode, 3D actually complements the basic area data that 2D provides to give excellent test coverage and the user gives the most comprehensive view of the process. For component placement a 2D approach to catch presence/absence and the position of components is sufficient. For solder joint inspection, Orbotech offers a 3D system that employs a high-resolution top camera with four angled cameras. Combined with multiple angled white lighting, we are able to accurately characterize the profile of solder joints, and of course get visual access to otherwise concealed or partially concealed solder joints such as those on J-legs ICs or increasingly high-density edge connectors.
[Stacy] Agilent's solder paste inspection and X-ray system both utilize 3D inspection techniques.  For solder paste inspection having the 3D volume and height information is important because volume has been linked to long-term joint reliability.  If you only measured 2D you would not have that critical information.  A main driver for X-ray is hidden joints and the 3D is needed because of the double-sided boards that our customers build.
3. What are critical machine vision system performance criteria for each of the applications that you address?
[Shavi] The following are vital:
1) Irresepctive of the inspection stage, all machines have to meet SMT line speed with 100% board inspection (not sampling).
2) In paste inspection applications accuracy and repeatability of measurement is critical with Gage R&R <10%
3) Low false alarms - FPY >90% for post reflow and >95% for paste/placement
4) Detection level >95% - this means see all defects including lifted legs in the post reflow stage
5) Fast and easy set-up.
[Stacy] In AOI (automated optical inspection) customers often utilize accuracy and repeatability to determine if an AOI system will meet their measurement needs.  For AXI (automated X-ray inspection - in our case it's 3D) low false call rates is a main driver for adoption and implementation.
[Jim] For all machine vision applications there are five main criteria ?false fails, false accepts, programming time, speed, and of course, cost. Most customers consider speed and cost parameters in their preliminary rankings of the products on the market. However, false fails, false accepts and programming time can have just as big an impact on the customer抯 economic bottom line as speed and system price.
The Landrex Optima series machines for pre- and post-reflow optical aim to have exceptionally low false fails and false accepts in the range of 20-50 ppm which remain stable over long periods of use of the systems in line at real customer sites. The Optima series machines also have exceptionally high speeds, especially the Optima II 7310 Extreme for post-reflow analysis. Each type of machine has a programming time of 2-4 hours. Finally, each machine starts at an asking price below 100K.
The Optima 7210 OPT system has additional performance criteria. Because it is a measurement system as well as a defect screener, it adheres to very rigorous accuracy and repeatability specifications. In general, a measurement system must be 10x more accurate and repeatable than the machines that manufacture the boards, in this case the pick and place system, in order to provide useful measurement information about the process. For an 0402, for instance, the system must have an accuracy of +- 100 microns and a repeatability of +- 10 microns. For an 0201, the requirements are even more stringent. In general, this type of requirement is known as a GR&R of 10%.
4. What changes have been taking place in the technologies that are the basis of the machine vision systems used in the electronic industry that has resulted in improved performance?
[Stacy] In my opinion, the most noteworthy would be the evolution in camera technologies, which are assisting with the continuous improvement of the AOI product inspection speed and accuracy.
[Jim] In the past 5-8 years we have seen dramatic improvements in the hardware components of machine vision systems. For twenty years in the machine vision field, engineers suffered with extremely limited camera and lighting choices and also relatively slow computers with limited memory. Machines constructed of these hardware pieces could only use the most rudimentary image processing techniques to solve their vision problem. Since 2000, the hardware area has exploded with newer, faster, cheaper, more reliable, bigger array cameras. In addition, color cameras are now an option with all the same parameters as the gray scale versions. Computer speeds and memory capacities have increased exponentially, while prices have decreased at shocking rates. Finally, the most dramatic advance has been in the lighting industry. With the advent of LED technology, especially white LEDs, it is now practical from a cost and speed perspective to provide white strobable light in many different shape configurations. All of these hardware advances give the machine vision the capability to 搒ee better?and the basic computing platform to process information in ways we couldn抰 have previously imagined.
[Shavi] An AOI system is a multidisciplinary machine comprising optics, hardware and software. Advancements in all of these technologies enable us to create products to meet the growing demands placed on inspection by the electronics industry. Cameras with faster frame rates that ensure high image quality coupled with high-performance frame grabber hardware enable amazing data read-out rates and a subsequent speed of operation that was traditionally the bottleneck especially for 3D AOI. Advancements in lighting technology - the cornerstone of image processing ?such as LEDs, are also giving huge benefits in terms of intensity and the ability to control the lighting environment at camera frame rates up to 200 FPS. Finally, new software platforms such as JAVA have had a profound effect on how we can run routines on hardware and dramatically reduce time-to-market for new developments.
5. Where do you see breakthroughs coming in the specific technologies that are the basis of machine vision systems used in the electronic industry that will result in further improvements in the near future ?next three years?
[Jim] The next wave of breakthroughs will be in the advances in brains of these systems, namely provided by the computer vision algorithms. New algorithms will allow us to do existing tasks easier and more reliably and to do new tasks that were previously unimaginable. For instance, thanks to computer vision algorithms we are now able to offer variable data about part position rather than just 揼o/no-go?defect analysis. This variable data will truly revolutionize the field of process control for PCB production. The hardware revolution from the past five years will lead to an unimpeded software revolution in the field in the next five years.
[Shavi] Multiple core processors will create a breakthrough for data processing and give impetus to new levels of machine performance where tasks that are typically executed sequentially can be performed in parallel. State-of-the-art classifiers for true multi-dimensional planar separation of interrelated characteristics will drive a new level of defect detection with near to zero false calls.
[Stacy] Same as above. I believe the evolution of camera technologies will drive further improvement in AOI in the next 3 years. 
6. Are there market changes in the electronic industry that are driving the adoption of machine vision? E.g.. What impact does RoHS compliance have? What about new IC packages and interconnect densities?
[Shavi] AOI is experiencing a steady and healthy growth rate in the area of 7 ?10%. The introduction of RoHS has no doubt influenced the decision to introduce AOI but is not a major deciding influence. Increasing package miniaturization and legislative pressure on companies, however, is. The bottom line with miniaturization is that the human eye cannot cope and the industry is resorting evermore to the reliable alternative that machine vision poses to complement or replace both manual visual inspection (MVI) and electrical test. Legislation ?which includes RoHS ?is putting pressure on manufacturers to become more transparent and provide more comprehensive documentation on the whole process. In this area, AOI offers major benefits in terms of both the visual and quality data that it provides on the SMT process.
[Stacy] For AOI and AXI the ever shrinking dimensions of packages/devices on SMT boards is driving adoption.  Additionally for AXI, access is an issue.  The prevalence and increased usage of BGA and CSP (ball grid array and chip scale package) packages that inherently have hidden joints are driving AXI needs.  I believe that lead-free is a driver in that if you use machine vision (AOI and/or AXI) you can bring your processes to stability faster.
[Jim] The truth is that it is harder and harder to test printed circuit boards. With device pitches getting smaller, boards denser and more complex, and materials changing from lead to lead-free, a strategy of relying solely on human inspection, ICT, and functional test will result in lower yields. Additionally, the margins most companies get per board are decreasing. As a result, most companies see a compelling reason to do 100% optical inspection to increase yields, decrease manufacturing cost, increase quality, and in the long run grow market share.
7. What impact do these changes have on the inspection requirements for machine vision systems? And how will machine vision systems have to change to address these more demanding requirements?
[Stacy] I do not believe lead-free material set is driving the change in requirements for machine vision but the drive towards smaller devices is.  The requirements for accuracy and repeatability are continuing to get more challenging as the devices get smaller.
[Jim] Optical inspection systems have to evolve to meet these demands. The hardware and software has to see tiny parts, understand variations in the new materials, have enough horsepower to meet the speed of the line, and be reliable enough to run 24/7 for long durations without maintenance. The machines have to deliver minimal false calls and false accepts without a lot of human intervention. A false accept can mean that the board may fail in the field. A false reject results in expensive, labor-intensive human validation of the machine抯 output. False flags and false escapes can have a very big impact on the economic returns for many companies. The bottom line is that the machines have to be able to adapt on their own to changing visual aspects of the products being built. While this sounds simple and logical, it is quite difficult to engineer a system that can both adapt automatically to new conditions and provide reliable inspection results given the existing conditions.
[Shavi] There is paradox here as machines must become more complex and packed with more power to deal with these demands. The user ?on the other hand ?has neither the time nor the image processing know-how to harness and exploit this power. Inspection machines, therefore, will continue to offer a high degree of intelligence but the focus is to make it easily accessible and controllable for operators of varying skill levels. As well as the power to process and manage data, AOI will adopt a predictive approach to inspection by having built in intelligence that is able to both characterize and predict process variance. We are entering the era of unattended AOI.
8. As a supplier of machine vision systems for the electronic industry what are some challenges you face in marketing machine vision systems?
[Jim] The biggest challenge that we face is getting the customer to understand the differentiation between our products and other competing products in the market that offer a continuum of capabilities. The truth is that anyone with a camera and a flashlight can cobble together an AOI system. Every year we see a few ''home-grown'' AOI systems put on the market, and we also see people actually buying them because of their usually low price tag. Surprisingly, one can get a limited base level of performance with almost any AOI system. However, to get to world-class levels of performance, with error rates in the range of 20-50 ppm, that stay stable from a small run to high volumes and across different types of boards, one has to do a significant amount of engineering. Making a machine that automatically understands what it is looking at 99.99% of the time is a very hard task. We as humans perform visual recognition tasks all the time in different situations. However, we have had millions of years of evolution on our eyes and more importantly our brains.
During a short evaluation at a customer site, it is difficult for the customer to understand the difference in technologies that will give them excellent performance in both the short and the long run because most machines fair well in the short run and the long run is not easily demonstrable. Our products have 20 years of engineering, science and practical experience behind them. We test them extensively in line at customer sites on real boards. We want to see our excellent performance last well beyond the evaluation to running 24/7 on millions of millions of parts in real customer use environments. We provide this customer data as part of our sales literature because it is important for people to understand that performance in a demonstration is very different from performance over the long haul.
[Shavi] Orbotech抯 focus is on innovation and adoption of new technology for both the inspection requirements of today and the challenges of the future. There are many systems on the market and most are set up to demonstrate good initial performance and tick all the feature boxes. The issue for the market is to distinguish between technology that has limitations and will create a bottleneck in the future and systems built on new technology that are well equipped to adapt to their needs. In our marketing approach, we focus, therefore, on the four basic elements of AOI, namely: ease-of-use, detection, speed and accuracy.
[Stacy] One large challenge is awareness.  Even though Agilent has a portfolio solution it is interesting that many people do not know we sell 3D solder paste inspection systems, for example.
9. What are your thoughts on the future of machine vision in the electronic industry?
[Shavi] Rather than a necessary evil, I believe that machine vision will establish itself and enjoy higher penetration into lines and processes and almost become a commodity in electronics manufacturing. The biggest challenge facing our industry is how to deal with the transition of electronic functions onto the silicon substrate. The result is less visible components on the assembly and although X-ray may be the obvious answer unless there is a breakthrough in sensor technology X-ray will remain a cumbersome, restricted and more expensive alternative to AOI. With cost continuing to be the biggest driver a more effective solution is to use machine vision to control the pre solder process. This will include intelligently combing data from different machine sources in order to better understand process behavior and predict non-conforming states.
[Stacy] With the adoption of lead-free and the advent of new and smaller packages we will soon see closed loop being more prevalent for pick and place and AOI and for the screen printers and paste inspection.  In the future, machine vision applications such as AOI and AXI will be 憇tandard?in production lines.  The increasing abilities of AOI and AXI to influence process control through defect prevention and defect containment is ensuring that machine vision will soon be considered a value add and must have rather than an 慹xpensive test system?
[Jim] There are four trends happening in the electronics industry. The first is that we are able to build more complicated sophisticated machines. The second is that the asking price for these machines, almost irrespective of fields, is dropping rapidly. The third, which is almost a consequence of the second, is that factories are constantly on the move to find the cheapest labor source. Finally, in order to keep up with rapidly changing products and to keep the factories at full capacity, each factory is making a larger mix of products. As a result, we see that inexperienced factories are making more complicated and mixed lines of products for dramatically low prices. A scenario like this one demands automation at every level, including optical inspection of the electronics. Thus, we will see much less reliance on human intervention. In addition, not only will the machine vision systems being used for defect inspection, they will also be used in an integrated way to provide information about the process in order to help the factories build product right the first time.
10. What advice would you give to a company investigating the purchase of a machine vision system for an electronic industry application?
[Jim] Surprisingly, personnel costs are really the hidden costs of AOI machines. The big issue is that people do not realize that the visual appearance of printed circuit boards, pasted pads, components, and solder joints can change dramatically across boards and even within one board. An optical inspection system must be able to handle this naturally occurring, acceptable variation and to distinguish it from real defects reliably and robustly.
Most AOI systems ask the programmer to show it all the types of variations it can expect to see, and to give it a recipe for how to inspect for defects given these types of variations. If a new variation occurs, the programmer must 揹ial in?this new occurrence. If we assume that new variations happen all the time, most machines require almost constant attention and a highly skilled programmer to keep them at acceptable false fail and false accept rates. If a programmer is not available to fix the issues, the companies pay for the problem in having humans verify the false calls. Most people don抰 realize the staffing commitment they抣l need after they purchase the machine. This level of staffing can increase the cost of an AOI system by 100%, which is way too high.
If we had one piece of advice to offer customers, it is to evaluate how well an AOI machine can automatically handle acceptable variation on PCBs without the aid of the human programmer.
[Shavi] Companies often make the mistake of focusing on the test aspects that will solve their immediate problem with little consideration for the future or the changing environment. When we talk to customers, it is not uncommon to find a previous AOI purchase standing unused in the corner because it eventually failed to meet expectation. Consider if the platform has the potential to adapt to meet new challenges such as new component formats or board designs. Give thought to your employee turnover and skills. It may seem to be an advantage to drill down to individual component algorithms but can that knowledge be passed on? Lastly, if your enterprise is on course for expansion will your machine meet the growing demand. Often, the simple fact that the machine cannot be connected to a central shop floor data collection system can seriously diminish its effectiveness as an asset.
[Stacy] Look for leverage.  An entire portfolio solution with the ability to help tailor the test strategy and implementation, like Agilent offers, will give you the best test plan for your production line.  Leveraging a full line solution from a company like Agilent also brings hidden cost savings from things like training because the portfolio systems operate similarly or the same in some cases. The lower infrastructure costs and the general overhead in addition to the data cross-talk among machine vision products is guaranteed to save manufacturers money.  Rather than engaging in long laborious evaluations with many suppliers to save a few $K on a point solution, why not invest in a portfolio solution where your cost savings can be customized and maximized and continuously evolving over time?