注册 登录  
 加关注
   显示下一条  |  关闭
温馨提示!由于新浪微博认证机制调整,您的新浪微博帐号绑定已过期,请重新绑定!立即重新绑定新浪微博》  |  关闭

论老子

道,领导也。领导必需要不断呼唤,教导下属以及以身作则。下属的过和错皆因领导懒惰。

 
 
 

日志

 
 

Chapter 30: Misused of Statistical Process Control  

2012-06-24 11:06:10|  分类: Buffer Mentality |  标签: |举报 |字号 订阅

  下载LOFTER 我的照片书  |

“Visual Management is not complete with a complete understand of the quality issue. One of the most powerful tool use to detect a quality issue as and when it is about to produce significantly larger number of defects is Statistical Process control (abbreviation SPC). This abnormal situation must be caught fast enough. Other wise, the deteriorating underlying condition will produce a large number of defects. However, SPC Let me continue to explain to you the possible misused of statistical process control (abbreviation, SPC),” I told John, “Following is a true story.

I spoke to a staff engineer, Xavier who works in Seagate Singapore, the world largest manufacturer of hard disk used in computers. This company pays US$15,000 to train each and every one of its engineers in 6-sigma workshop. And it employs more than 500 engineers. That is millions of US dollars invested in the 6-sigma program. So, should you expect it produces close to 3.4 defects per million kind of quality level?

This was how I interviewed Xavier during one of my visit to his home.

Based on my twenty years of implementing SPC, I went in straight in and said, ‘I know at least 40% of the control charts (interchangeable with SPC charts) are not effectively. How to tell whether a control chart is effective?

When there is an out-of-control point, do your engineers react to it immediately?’ Back in my mind, I would like to say 100% of the time no one would react to the out-of-control points.

Xavier replied, ‘Yes, we do. Our equipment is fixed with an auto-alarm system that whenever an out-of-control point occurs, the few engineers in-charge of that piece of equipment and their immediate bosses are informed through a computerized system linked to their personal computers. The engineers will therefore, rush down to the production line and act on it.’

I said, ‘Okay! Let’s assume they responded immediately all the time. What do you think is in these engineers’ mind? At that pressing moment to avoid prolonged equipment downtime, in his mind he thinks of two things. One, he must first quickly decides ‘Go’ or ‘No go’. ‘Go’ in this context it means to release the equipment to continue production. ‘No go’ means to down the equipment for further investigation. I strongly believed almost all the time, the decision was a “Go”.

Xavier replied, ‘Yes, I fully agree with you. But what is wrong with that decision?’ 

Figure 30-1: Distorted SPC principle

 

I explained, ‘The basic tenet of SPC is to down the equipment for an immediate investigation after an out-of control point occurs. Upon completing the investigation, an engineer then decides what to do with the piece of equipment. Your engineers are doing things in the reverse order. That is they make the decision whether to continue production or not before carrying out the investigation. Under this kind of erroneous practice, I can safely conclude that your engineers never carry out a proper investigation.’

‘No! That is no true,’ said Xavier ‘We do carry out our investigation. Our SPC system is very advance. It allows us to tabulate the recent occurrences of out-of-control situations. We do a Pareto Analysis[1] and we work on the top 3 most frequently occurring problems.’

‘Wow!’ I exclaimed, ‘That is not the true spirit of SPC which calls for immediate investigation. In addition to allowing a delay investigation, it also tells the whole engineering community there is no need to investigate most of the issues. The correct principle should be investigate first and then decide whether to stop or continue with production.’

 

Figure 30-2: Correct SPC principle

 

Let’s not drill further into the actual practice that had deviated from the basic tenet of SPC. What is the cumulative process yield? Is it around 98%? That is 2% yield loss. At 20,000 defect per million, it is a far cry from 3.4 defect per million.

I am very sure none of the company that had spent a fortune in training their engineers in the 6-sigma black belt program is achieving a quality level anything near zero defects. Perhaps, at 2% defect rate, that is the best that they can achieve. Why am I so certain of my deduction?”

John shook his head.  

 

Illustration #1: Making an incapable control chart looks great with respect to the specification limits

 

I continued, “I am an expert on statistical process control. There are many ways points plotted on a control chart could reveal a looming problem with either an incorrect way of setting the control charts or the process is simply not capable of producing good parts consistently. Of course, to the untrained eyes an incapable process that produces a significant amount of rejects could depict itself as a process that is well in control when in fact it is not. Let me share a few examples with you.

In HP, the attach machine picks and places orifice plates onto the dies that form the printheads of the inkjet pen cartridges. The orientation of the orifice plate and the die must be precisely superimposed on top of one another. The distance between the center of the orifice plate position marker and the die position marker was measured.

A total of sixty-four points are measured and its average value plotted onto the control chart. All the points are seen hugging around the center line. The upper and lower control limits are a short distance above and below the center line. There are two other lines drawn many times further away. They are the upper and lower specification limits. Visually the points plotted are very close to the center line and none nearing the control limits. This is an obvious sign that the control limits were not computed correctly.

 

Figure 30-3:    Specification limits must not be plotted onto a normal control chart



I casually asked the process engineer if the attach machine was doing well. He replied, ‘Of course, it is doing very well. It is very capable. You look at the control chart. All the points are near the center line and they are very far away from the specification limits.’

I pulled the engineer into a meeting room and explained to him that he should not draw the specification lines on the control charts. With a sample size of sixty-four readings taken to calculate the average value for one point, the spread of the points plotted on the control chart is naturally one-eighth the distance between the specification limits if the process is marginally capable at a process capability index of one. The one-eighth reduction is derived from the square root of sixty-four; the sample size taken to compute the average value of one point plotted on the control chart.

Visually all the points hugging around the center line and very far away from the specification limits do not necessarily mean that the process is very capable. In this case it was a visual trick done by the vendor who designed the piece of die-attach equipment. If the control chart is seen as very capable he will sell many more pieces of the die-attach equipment. It was pure marketing gimmick! I ordered for the specification limits to be removed and the control limits revised. The yield went up to better than 99.9% in less than 2 months.

 

Illustration #2: Large variations in the quality of purchased parts

 

In the pen assembly line the engineer felt continuing to plot the control chart was not useful. On and off the control chart shows an out of control situation but when they checked the output from the assembly lines every single unit was good. The control chart was giving false alarm causing them to go for wild goose chase for nothing. It was a complete waste of time.

Over time the process engineers didn’t believe in the chart and thus, didn’t respond to the out of control situations anymore. The charts were still being plotted but nobody would want to pay any attention to it. It had become a complete failure of implementing process control in the assembly line. Though the line yield was quite good at more than 98 percent, certainly this could be closer to 100 percent yield.

I queried the process engineers and found that one significant assignable cause of the sudden shift in the points plotted on the control chart was whenever there was a new batch of raw plastic bodies was phased in after the old batch of materials run out. Therefore, the engineer felt there was no need to do anything.

In normal practices, the procurement engineer should have gone back to the vendor who produced the plastic bodies to improve its process control to reduce the variation in the dimension of the plastic bodies. But wanting to be a likeable person, none of the procurement engineers ever told the vendors they had to do something about it.

 

Illustration #3: Response plan

 

In the Hewlett-Packard Singapore inkjet factory I saw the introduction of an innovative piece of document that was devised by engineers in Hewlett-Packard. It is called the response plan. What does this piece of document do? It tells the production operators exactly what to do when an out-of-control (OOC) situation occurs.

Usually the out-of-control situation is depicted as a point plotted beyond the control limits. There are two control limits; the upper control limit and the lower control limit. Based on the studies of statistics, the law of large numbers says if a process is in control the chance of a point to be plotted outside of either the upper or lower control limits is 0.27%. In other words calling for a false alarm action to bring the process back into control is unnecessary 0.27% of the time.

If the process had drifted away from the process mean or targeted value the chance would have increased considerably. If left unattended it would mean the process will continue to produce significantly higher number of rejects.

The industrial practice is to react to the situation whenever a point is outside the control limit. That means all out-of-control situations need to be investigated. Based on the result of the investigation the engineer decides whether he needs to take corrective action or not. Most likely he will have to investigate for the root cause first. If not, the process will continue to produce excessive rejects of more than 0.27%.

The statistical process control guideline says every time an out-of-control situation occurs the process engineers have to rush down to the production floor to investigate for an assignable cause. The engineers find it very frustrating to stop doing a task on hand to attend to the out-of-control situation. It would be worse if the process is not capable of producing quality products. The frequency of the engineers required to run down to the shop floor will be excessively frequent.

So in order to beat the common industrial practice that says that he has to respond immediately to an out-of-control situation, the engineers innovatively add a diamond box in the response plan flowchart. The diamond box reads, ‘3 Consecutive point OOC?’ If ‘Yes’, call engineer. If ‘No’, resume production.

 

Figure 30-4: HP’s unique response plan


Assume the process has shifted away from the process mean and is already producing significantly more rejects at say, 10% rate. This means there is only a 10% chance a point plotted on the control chart is out-of-control. The chance for the next point to fall outside the control limits is also ten percent. So is the third point.

What are the chances of three consecutive points in a row would lie outside the control limits? The answer is 0.1% (10% x 10% x 10%); which is much lower than 0.27%. The response plan has effectively mitigated the need to call the engineers to come down to the production floor to do an investigation even though the process is indeed already running out-of-control. This rule was widely copied over to many response plans throughout the factory.

However, the production operators noticed the number of rejects went up and they have to do a lot of unnecessary re-screening and subsequent rework. The statistician was called in to provide an answer to whether it is alright that repeated situations where the process control charts are highlighting the out-of-control situations are alright not to be followed up with any investigation. Could the production of more rejects be stopped in time?

She knew there was nothing wrong with the control charts. She knew it was the response plan that fails to get the operators to alert the process engineers. But she kept quiet. If she pointed out this fact, she would incur the wrath of all the process engineers. She chose to be a likeable person and opt to play along with all the engineers who shared the same buffer mentality.”

John’s eyes were wide opened. He said, “Eric, you can tell the presence of buffer mentality by just looking at the control chart. You are simply marvelous.”



[1] Please go back to read chapter 5, ‘Misused of Pareto Analysis’.

  评论这张
 
阅读(194)| 评论(0)
推荐

历史上的今天

在LOFTER的更多文章

评论

<#--最新日志,群博日志--> <#--推荐日志--> <#--引用记录--> <#--博主推荐--> <#--随机阅读--> <#--首页推荐--> <#--历史上的今天--> <#--被推荐日志--> <#--上一篇,下一篇--> <#-- 热度 --> <#-- 网易新闻广告 --> <#--右边模块结构--> <#--评论模块结构--> <#--引用模块结构--> <#--博主发起的投票-->
 
 
 
 
 
 
 
 
 
 
 
 
 
 

页脚

网易公司版权所有 ©1997-2017