Gage R&R Won't Save You: The Case for Genchi Genbutsu
The production line goes down 4:30pm on a Friday. Great. Someone says
“We don’t know if the gage is reliable.”
“We need to confirm the gage with an R&R study”
Stop. It’s time to grab a hold of this situation, because I’m going to be home for dinner.
Use Cases of Gage R&R
This article is not to discredit gage R&R; at least not entirely. Gage R&R is useful to say that a gage is appropriate for a process and to measure the capability of said process. Gage R&R can also be used to quantify the performance of a gage to be compared against other gages. Don’t skip the gage R&R to set up and control a process. However, in firefighting, all too often people revert to gage R&R to confirm the measurement system. I’ll make the case for why this is wrong.
Scenarios to use Gage R&R:
-Qualifying the fitness of a gage for use in serial production
-As part of determining capability of a process
The Dangers of Gage R&R:
This may come as a shock to some but it is possible to fail gage R&R and solve an issue using the very same gage. Alternatively, it is possible to pass gage R&R and still have a gage issue.
Danger 1: Fail gage R&R and Still Solve a Parts Issue:
This comes down to interpretation of how parts for the R&R study should be selected. Often, parts are selected one right after the other at “random”. Sometimes parts are selected to represent “the full range of variation”. Almost never are the suspect parts themselves part of the study when a problem arises. That’s a shame, because the gage may be able to discriminate between good parts and the defective parts in question.
Fig 1. A case in which a gage is inappropriate for a process but is appropriate for solving the problem.
In Fig 1. gage variation accounts for 50% of tolerance and 30% of normal process variation. Obviously, this is a terrible gage for this process. Why is it on the line? Nevertheless, let’s continue to use the gage to try to figure out what’s going on with the defective parts*. Why? Because there are enough distinct categories using the gage between the nominal good parts and the defective parts.
Fig.2: Lots of distinct categories!
If the goal is to address the defective parts in red the existing gage is sufficient. If someone on your team wants to improve the gage before moving forward convince them otherwise or split ways and go it alone. If they don’t understand this you’re better off. You don’t need a ball and chain at work.
Danger 2: Pass gage R&R and have a gage issue:
Defective parts do not necessarily behave the same as parts used for a gage study. Special cause variation is uncharted territory. The gage may be measuring something people designing the gage never intended. In this case, any gage study done previously is meaningless.
Fig 3. Part on datum pegs.
Imagine a process where a part is measured for height as in Fig. 3. One side is placed on a series of pegs to create a datum and on the other a probe to measure the height. A gage R&R is conducted with samples across the normal distribution. Suppose the study results in GRR of 1%. GRREAT! Right?
However, big trouble in little China, because now, rejects are flying off the machine gage. People are in full panic mode. Measurements from the CMM lab report the reject parts are to specification. Running the master parts in the station shows no need to calibrate the gage. I swear to God if someone says gage R&R.
I won’t beat around the bush. It turns out there is an allowed burr on the edge of a part where the part rests on the peg to create the datum. Based on the drawing with the allowed burr this is an inappropriate gage. However, this poorly designed gage originally passed gage R&R with parts without an allowed burr.
The Genchi Genbutsu Solution
Genchi Genbutsu (現地現物) is a key principle in the Toyota Production System meaning “actual place, actual thing”. This concept is applied consistently in many successful organizations. As it pertains to measurement systems this means that the defective parts are included in understanding the problem. This solves the first and second dangers above.
In the first danger, it would be obvious that the gage is consistently able to tell the difference between good parts and bad parts. Additionally you would see that the bad parts are a number of distinct categories away from the good parts.
In the second danger, only by studying the parts being rejected would you find the burr and the difference between the CMM lab and the gage.
I’ve run into both scenarios where somebody wanted to do a gage R&R in problem solving mode and the problem was more swiftly solved by making logical comparisons between good and bad parts being produced at the time. Also, yes, I was home for dinner those days.
Like my analysis of Japanese Culture? Learn the fundamental concepts that I teach in Course 1 for Japanese culture.
*defective parts in this case are parts being generated outside of the normal process distribution which may include bad parts at the tails of the distribution which are technically also defective. In terms of Six Sigma we are only interested in special cause variation.