Sunday, July 22, 2012

How to start the analysis and improvement of a process with a lot of waste

On our LinkedIn Discussion group, one of our members Mr. Ian Morton asked an excellent question about processes with a lot of waste and how to start addressing the improvement of the process:

Ian's question:

A system with lots of wastage

One thing that I can’t grasp concerns the situation when there is a lot of waste in a system, and this wastage is due to numerous special causes which are inter-related with each other.

I have read summary papers about SoPK, Deming’s 14 points for management, and have a copy of Out of The Crisis, but have I missed a key point or not taken it all in properly ? (any help greatly appreciated).

How would Deming have approached the removal of these special causes ?
Would he remove one at a time ? or,
Would he remove many special causes of variation?

First of all Ian it's a great question - Thank you for asking it.

It is a question I have been thinking about for some thirty - three (33) years now since I first started applying Dr. Deming's concepts to the manufacturing of printed circuit boards at Hewlett-Packard's data Terminals Division in Sunnyvale, California. We were testing some 4000 PCA's a day with a better than 50 percent failure rate in the board test area.

Note: we did find a workable solution - a little bit of a surprise - so read all the way to the end of this discussion for my final point about data analysis.

We started collecting data and preparing Statistical Control Charts  "Xbar - R Charts" and had data coming out of out ears. That was our FIRST MISTAKE Dr. Deming made fun of during a senior manager's meeting held at HP's Cupertino site in March of 1982.

That was the eve of the "birth of the current day"  Red Bead Experiment.  Mr. Bill Boller was in the room that day.

I will never forget the laughter when I ask my question to Dr. Deming in front of the group of about 300 HP senior managers. He scorned at me and said something like...  you got too many control charts ... young man!

So my first comment about your data is to make sure you are only collecting data on the most important points and points in your processes that are impacted by and have impact on the most critical outcomes of your processes.

An excellent analogy in the manufacture of plastic molded parts is the concept of critical dimensions of the part being made. The plastic part experts can determine one or two "critical dimensions" on the part that when measured determine if the entire part is made correctly or not.

Secondly, if you don't use Xbar R charts, please consider switching to them to analyze your data. One of the first lessons we learned is that the magnitude of the Range graph is the first place to start worrying.

If the magnitude of the range fluctuates wildly from data point to data point them your process is "driving you crazy" with what I call "fire-drills" and in effect is wasting your time and taking you away from the important analysis of improving the entire process. So try to understand what is behind the fluctuations and take action to smooth the Range first. I refer to the Range Chart as my frustration index.

Referring to the sample Red / White control chart - see link:

You will notice that for the most part all of the data points plotted in both the Xbar and the R chart are within the calculated Statistical Control limits. That means that the process of drawing white beads is operating within statistical control and for the most part the managers of the process should NOT BE WORRYING.

If there would have been a data point plotted higher than the calculated limits Dr. Deming teaches us that that would be a "special cause" and most of the time we should not worry about those types of causes either. Putting effort into "chasing special causes" again takes away your resources from making the most important analysis of improving the entire process.

The magnitude of the average of the Xbar chart should now be our primary focus of analysis.  All of the data points that fall within the calculated statistical control limits are what Dr. Deming calls  "common causes".

As a manager of a process (like drawing white beads) one of the first decisions you should make is whether the average of the process Xbar is acceptable to you from a business perspective. -- as Paul refers to in his comment above.

Yes, there may be "Red Beads" defects in your process but sine the process is acting in statistical control. It will take work - sometimes a lot of expensive work to identify and remove the common causes (Red Beads) from the process.

Is all that work cost effective? Or is there another way to deal with the issues or perhaps even find someone that you can "sell the unwanted Red Beads to" ? Be creative!

That's a pure management business decision and one that should be made before spending lots of resources analyzing the root causes of all the problems.

Now assuming that management has made the decision that the average of the Xbar data is NOT acceptable, you as a data manager / analyzer need to organize your data using some sort of  Pareto Chart technique  and start looking for the 20% of the common causes (Red Beads) that when eliminated will reduce and improve the entire process by some 80% - lets NOT make work too difficult!

Each one of the common cause (Red Beads) in the process has a name engraved on it in very tiny difficult to read letters. It will take work and the involvement of the "willing workers" to help identify the names of the problem issues. Here is where the majority of Dr. Deming's 14 principles play a role.

Workers know what is going on in a process; but, if management does NOT create the proper environment; uses  ridiculous reward and punishment systems; unfair bonus and performance reward systems... all the knowledge of the "willing workers" will be LOST!

As you know, there are many points that the Red Bead Experiment makes and I won't attempt to cover them all in this discussion.

So in conclusion, create the proper environment; listen to what the willing workers have to say; carefully collect your data (not too much of it); don't worry about "special causes" outside of the control limits; use Pareto Chart techniques to find the 20% most import issues; and eliminate those 20% "Red Bead - common causes" permanently from the system. 

Don't make the fatal mistake of allowing anyone to put a "temporary fix" on the problem  - as it's like painting Red Beads with white paint and putting them back into the box. The white paint will eventually "rub off" - usually at the most expensive part of the entire process in the customer's hands!

Now for our surprise discovery that saved us  "lots and lots"  of money and  allowed me to stop working those 14 hour days. 

We discovered that the product (HP's 4X computer terminal) was made with about 13 to 15 printed circuit assemblies. When we tested each PCA to very tight specifications - many failed but even those that past failed again in final test of the entire product when a full set of boards came together.

The solution was to select boards and test them as an  entire product -- but before they were installed inside of the plastic of the terminal.

One very clever process engineer (Mr. Mark Cardella) developed an oak wood frame test fixture that would hold a complete sets of some ten or so PCA's and the oak frame would slide into a burn-in test oven as a working set of boards.

The root cause of all our problems was "tolerance stack up" at the component level and this information was eventually fed back to the R&D engineers to be used to improve their process of component selection and to improve the incoming inspection of critical components.

I am always happy to add my opinion or attempt to answer your question about the Red Bead Experimentand am available offline to do so.

email me at:

or make a POST on our BLOG:

or on our Facebook page:

If you are a consultant - consider a FREE listing on our NEW directory:

end of comment

No comments: