SHAININ DOE 七工具介绍Multi-Vari Chart(多层图)B vs .C (B与C比较)Paired Comparisons(成对比较)Components Search(组件寻找)Variables Search(变数寻找)Full Factorials(全因子效果)Realistic Tolerance Parallelogram (scatter plots)(散布图定公差)目的:降低变异MUTI-VARI CHART多层图:变异之掌握时间面变动(Temporal Variation)在不同的时段、生产班次、生产日期、生产周别等等,由於时间不同制程会发生的品质变异,是一种非随机性的要因,只要能掌握到它们的存在,伴生的品质变异就可望全数消除。
空间面变动(Position Variation)在相同时间里,在不同的部位、机台、人手或工厂所发生的品质变异,就是所谓的空间要因所产生的。
经过恰当对策後,空间面要因所产生的品质变异可望消除大半。
以下列举了各类的空间面要因:单品的内变异,如一件铸品因不同部位孔隙度有差异。
组品内各单件之间的差异,譬如一块含千、百只零组件电路机板,各点之问焊锡品质有差异。
全品之内相同各件之间的差异,譬如一片晶圆上数百粒晶体之间品质出入很大。
同模或同次生产,各件产品之间的品质差异。
譬如在IC的封胶制程,,乙付模具上通常有数十处相同的穴位,但产出的各个胶体之间也有所差异。
不同的作业手、生产机台、或生产工厂投入相同的生产要素,但产品之间也有品质差异。
重覆面变动(Cyclic Variation)在同一机台,用同批材料、由同一作业手、按相同程序生产,产品之间仍有品质差异。
这种随机性要因是会再度出现的,所以它们有反覆性。
只有在技术上、材料上或设备上等等有所突坡,此类反覆性品质变异才可以减少。
讨论:请举出在LCD之制程中,时间之变异有哪些。
讨论:请举出在LCD之制程中,空间之变异有哪些。
讨论:请举出在LCD之制程中,重覆之变异有哪些。
Multi-Vari个案研究:转子轴某制造厂生产圆柱的转子轴,需求直径为”±”,制程能力研究显示”的(标准差)散布,C=。
领班准备废弃此老式的生产转子轴的六角车床设备(TURRET PKLATHE),买一个新的价格为$70,000,能保持±”的车床,即 Cpk=,然而,顾问说服工厂经理先行Multi-Vari 研究,即使在买进新车床前,它的回收只是九个月图表6-2显示Multi-Vari图的结果.空间面变动(PositionVariation)轴四个位置的(轴内)变动,显示如方格内,每个轴的左边到右边,上下为轴的最大的直径和最小的直径重覆面变动(Cyclic Variation)循环性的变动,一方格到这下一个方格时间面变动(TemporalVariation)从周期到这下一个,以小时显示.结论:图中显示,最大的变化似乎是时间到时间,变化发生於10上午和11上午,这提供这领班一个强的线索,上午10什麽呢休息时间!。
而在下一个三轴样本是取在11上午,这些读数是类似於最初8上午生产。
变异要因检讨解析例某家瓷砖制造商磁砖褙纸之褙纸黏度品质不易控制,搜集数据如下表(1)横条之内(每条5片瓷砖)(2)横条之间(3)时间,另外,将以上数据绘制成 multi-vari charts(包括每条中最高黏度每时段平均黏度、每条平均黏度),如图( 问题)1那一方面的变因有最大的变异2你可以找到什麽端倪包括非随机的趋势。
Multi-Vari Chart 之制作计算各组最大,最小,小平均,大组平均B8格 =MAX复制B8,至B9格 =MIN复制B9,至B10格 =AVERAGE复制B10,至B11格 =AVERAGE复制B11,选择贴上(值)至复制B11,至E11复制E11,选择贴上(值)至复制E11,至H11复制H11,选择贴上(值)至画图分隔之做法练习为了解0402印刷宽度之变异,取3个MASK,每MASK作4JIG,每JIG取上下两PACK,每PACK,X方向与Y方向等距离取3点共9点,量测印刷宽度,1.若规格在240 10,制程能力C PK=,显然不足,X = ,s =成对比较成对比较类似组件搜寻方法,藉由成对"良品"和"坏品"单位的比较,找出两者之间差异,进而根据其差异分析重要要因。
使用时机:单位元件或子装置不能够分解或重新组装(不像组件搜寻)有多数良品和少数的坏品成对单位出现有适当的参数来发现与区别良品与从坏品此技术可适用在组装站、制程、测试仪器,等具有类似的单位,组装,或工具。
同时,它也是失败故障分析的有力工具。
成对比较制作步骤:1.选出一良品单位和一坏品单位(尽可能的,接近相同的制造时间)。
2.称此为一对,详细地观察记录在二单位之间的差异。
差异可能来自外观的﹐尺寸﹐电性﹐机械性质,化学性质等,观察技术包括眼睛,X光,扫描电子显微镜,破坏测试等。
3.选择第二对良品和坏品单位。
如同第2步骤,.观察且记录此对差异,4.重复此搜寻步骤,第三,第四,第五,和第六对,直到观察的差异显现出有重覆的模式。
5.去掉每对中有矛盾方向的差异。
通常,到第五或第六对,一致性的差异将降至少数几个要因。
为差异的要因分析提供强列的线索。
成对比较个案研究:不良两极管DO-35两极管,汽车里的在-那之下-头巾电子学组件用,有无法接受的失败率。
一些被失败的两极管被从领域向後地带来和反对没有有缺点的好的单位比较。
被的成对比较结果,当在扫描的电子之下检查的时候仔细检查,是依下列各项:号码分对观察差异1良品-坏品良品没有缺点坏品Chipped die,oxide defects,copper migration2良品-坏品良品没有缺点坏品Alloying irregularities,oxide defects3良品-坏品良品没有缺点坏品Oxide defects,contamination4良品-坏品良品没有缺点结论:1.Four repeats in oxide defects, probable Red X family2.Two repeats in chipped die, probable Pink X familySolution: Working with the semiconductor supplier (who, up to this analysis, had resisted responsibility), the following corrective actions were instituted:1. For oxide defects:* Thicker photo resist* Mask inspection* Increased separation between mask and die2. For chipped die:* Reduced oxide thickness in scribe gridB VS. CB表示Better,C表示Current,就是比较好条件与现有条件是否有差异。
在过去常用的方法为两组母平均差之检定,但计算较为复杂,再过去统计方法中有很多简易之计算方法,其中SHAININ提出两种容易之方式(1)Lord Test及(2)Tukey Quick Test。
SHAININ使用Lord test 之步骤例:HOUREMETERAn hour meter , built by an electronics company , had a 20-25 percent defect rate because several of the units could not meet the customer's reliability requirement of perfect operation at -40 C .The worst units could only reach 0 C before malfunction .The hour meter consists of a solenoid cell with a shield to concentrate the electrical charge which pulses at regular intervals .The pulse triggers a solenoid pin , which in turn causes a verge arm , or bell crank , to trip the counter , advancing it by one unit .The counter is attached to a numeral shaft containing numeral wheels .These numeral wheels are separated from each other by idler gears , which rotate on an idler gear1参考Lord's test for two independent samples.此处Lord 采用平均,SHAININ采用中位数,较方便计算,判断值在5%下,Lord值为,SHAININ为,可能是为方便记忆。
shaft .Both the idler gear shaft and the numeral shaft are attached to the mainframe ,units is determined by the formula : D: R>:1,The Red X and Pink X are among thecauses being considered and there is good repeatability in the disassembly / reassembly process .Lord's test for two independent samples.In this test the sample ranges R1,R2replace s1, s2. This is a quick test, no more robust under nonnormality than the t test, and even more vulnerable to erroneous sample extreme values.Table A 7(ii) applies to two independent samples of equal size. The mean of the two ranges, w = (R1 + R2)/2, replaces the w of the paired test and X2—X1takes the place of D. The test of significance is applied to the numbers of worms found in two samples of 5 rats, one sample treated previously by a worm-killer. See table We have X2—X1 = and w = (219 + 147)/2 = 183. From this, t R = (X2—X1)/ R = 183 = , which is beyondthe 1% point, , shown in table A 7(ii) for n = 5. TABLE OF WORMS PER RATTreated Untreated12337814327519241240265259286Means, XRanges, R219147To find 95% confidence limits for the reduction in number of worms per rat due to the treatment, we use the formula(X2—X1)—t R R≤μ2-μ1≤ (X2—X1) + t R R- (183≤μ2-μ≤ + (183)60 ≤μ2-μ1≤284The confidence interval is wide, owing both to the small sample sizes and the high variability from rat to rat. Student's t, used in example for these data, gave closely similar results for the significance level and the confidence limits.For two independent samples of unequal sizes; Moore (7) has given tables for the 10%, 5%, 2%, and 1% levels of Lord's test to cover all cases in which the sample sizes n. and n1,n2 are both 20 or less.The range method can also be used when the sample size exceeds 20. With two samples each of size 24, for example, each sample may be divided at random into two groups of size 12. The range is found for each group, and the average of the four ranges is taken. Lord (3) gives the necessary tables. This device keeps the efficiency of the range test high for samples greater than 20, though the calculation takesa little longer.To summarize, the range test is convenient for normal samples if a 5% to 10% loss in information can be tolerated. It is often used when many routine tests of significance or calculations of confidence limits have to be made.SHAININ使用Tukey Quick Test之步骤1找出联合的那些二组样本里的最大和最小值,称为联合最大,及联合最小2如果联合最大值和联合最小值两者在相同组的样本发生,我们没有足够证据证实二组样本是不同的。