The following words of an anonymous poet as he (or she) immortalized the lessons from Deming’s funnel experiment.
“Tamper, tamper is the game, try to make all the same. Squeak and tweak it every day, off we go to the Milky Way.”
—Anonymous
I offer a corollary that may help understand the underlying problem:
ADVERTISEMENT |
“We love to tamper with our processes. We cannot help it. We are hard-wired to tamper with our processes!”
For many years, I have examined in-house and customer-generated specifications. Often, I have seen something curious called “action limits.” Action limits always fall somewhere between the target value and upper and lower specification limits.
The following is an example:
Brightness Specification |
||||
Lower Spec. |
Lower Action Limit |
Target |
Upper Action Limit |
Upper Spec. |
The target, upper spec, and lower spec are easy to understand. They should represent the voice of the customer (or, in some cases, engineering), but what about the action limits? What “action” is to be taken, and by whom? I have never seen “action limits” operationally defined in the body of a specification agreement. I have always been afraid to ask these questions aloud for fear of appearing ignorant in front of customers or colleagues.
Thus, I have concluded that when action limits are reached, someone does something with an intent to steer the process back toward the target.
However, after much study and thought, I have arrived at the following conclusion:
Action limits = Tampering limits
The only proper way to define action limits as other than tampering limits is to assure that the action limits represent the upper and lower control limits of the process based on a capability study of that process, while in a “reasonable degree of statistical control.” Control limits can only be determined from the voice of the process via a process behavior chart. Donald Wheeler, Ph.D., has written many articles on this subject for Quality Digest and Quality Digest Daily.
In my experience, specifications are seldom set with an understanding of the capability of the processes to meet the specifications. A typical scenario is that the customer meets with the sales engineer to discuss product specifications. Sometimes the specifications are necessarily tight, and at other times they are artificially tight, as when the customer tightens specs in order to make wiggle room for any performance problems from the vendor. The sales engineer should always determine whether the process is capable of meeting the customer’s specs.
The same scenario holds true for specifications that come from in-house engineers, who can design products with specifications that are beyond what the manufacturing process is capable of producing.
The most accepted method of properly setting specifications involves the assurance that the process capability index (Cpk) has a value of at least 1.33. This minimum value is a generally accepted rule of thumb.
There is no convention for determining so-called action limits. In my opinion, the term, action limits, invokes a feeling of comfort for the customer that the supplier will do something (who knows what?) if either specification is approached by the process.
Thus, tampering is virtually assured if action limits are adhered to in the operation of the process.
Case study
A paper mill producing lightweight coated offset papers was experiencing picking/linting at the customers’ press rooms due to low internal bond strength. Press room trials demonstrated that at an internal bond strength average of 90 the picking/linting was resolved. The paper mill accomplished this by adding cationic starch (20 pounds per ton of paper) to the papermaking process. Operators were given specifications as follows:
Internal Bond Specification
|
||||
Lower Spec. |
Lower Action Limit |
Target |
Upper Spec. |
|
The specification and action limits were set by use of “tribal knowledge.” Operators were instructed to adjust the starch flow when the action limits were reached in order to stay close to the target.
The mill confidently produced paper with the new specifications but, unfortunately, continued to receive picking/linting complaints from the customers. Further investigation showed that the operators were adjusting the starch flow as instructed, but were averaging 24 pounds per ton of paper, four pounds higher than the trials showed as necessary.
A capability study was performed on the process. The study revealed that when the process average internal bond was 90, the upper and lower control limits of the process were 111 and 69, respectively. Thus, the Cpk was 0.71, and the process was not capable of meeting specifications. Thus, the operators were doomed to not being able to meet specification. Any changes they made to the starch flow based on action limits led them to tampering with the system and increasing variability of the internal bond.
Upon learning this, the method of introducing starch into the papermaking process was changed to reduce the variability of the test by more than 50 percent. This change to the system changed the capability index to 1.45, which was quite acceptable. The “tampering limits” were removed and a process behavior chart was developed for the operators to use as a method to monitor the average internal bond and make changes only when it was appropriate to do so. Eventually, starch usage was reduced to 16 pounds per ton of paper and picking/linting complaints from the field were virtually eliminated.
Summary
Action limits are almost assuredly tampering limits, because action limits are meaningless outside the context of knowledge of the process, which can only be gained via a properly conducted capability study. Without this knowledge, we are doomed to tamper with our systems, increase variability, and disappoint customers.
It is indeed time to get rid of action limits.
Comments
Labeling Action Limits
This is interesting...I've run across the same thing. I've even know people to have rules for their action limits...three points in a row trending in the direction of the action limit triggered action. Did action limits come from ECP? Most of the people I've seen use them made that claim, anyway.
We should require anyone who wants to use them to label them in some meaningful way..."Tampering Limits" might help, if they understood tampering. Maybe just "act above this line to really screw the system up" or "abandon hope, all ye who enter here..."
Right on!
I've experienced non-statistical, Action Limit triggers with no defined subsequent process too, in my work as a Quality Technical Writer/ISO 9001&14001 Internal Auditor. Generally speaking, it would only make sense to have Action Limts if they were synonomous with the calculated process Control Limits on a control chart AND the action to be taken was pre-defined, well documented, and consistently followed (e.g., Out-of-limit data point trigger has specified person/s leading the investigation, analysis to be conducted, due date for initial findings, corrective/preventive action plan). Good job with your insight and explanations!
Action Limis
Not that I've ever used them but I thougt they were the same as "Warning Limits" (2 sigma limits). A British convention I think.
Rich DeRoeck
ex-tweaker
"Warning Limits"
"Warning" of what? That the process is operating with common cause variation? Labelling the 2 sigma limits as anything other than "2 sigma limit" is a demonstration of non-statistical thinking and will lead to tampering.
Add new comment