Re: Need advise on R chart with PpK of 17.85 for Ford
I would be interested in knowing how you selected the 125 pieces out of the 300 plus piece run. From my perspective there are a number of ways – process, specifications and customer - to look at this situation.
This is a good question. In a machining operation, the data needs to in time-ordered sequence. My preference is taking a part every 3rd pc. Ford likes to see 5 pcs per sample, but that is only applicable when analyzing data from a normal distribution established from random, independent data - which is rare in machining if you are in control. In precicion machining, individual data - not averages - tends to be more meaningful.
First, take a look at the process as described by the data with worrying about the specifications. Steve has run the numbers. I ran them also (see attached). The first question to answer is about the stability of the process. I have included the Xbar-R, individuals, process capability and histogram charts. As Steve points out, there has been a process shift that shows up great when you split the control limits. Something has happened to cause that shift. Note that it is a “good” shift – there is less variability in the process. If you can figure out what happened and ensure it continues to happened, you have improved the process considerably. You mention that you don’t want to account for the shift. As your customer, I would want you to – it has improved your process and means less variation coming into my process.
There could very well be a "special cause" at the beginning of the process. With no other notes on the process (such as adjustments, et. al.) one might conclude that it was the special cause of warm-up/start-up. During this time, the process is not stable - but as a customer, you are getting those parts anyway. There maybe a need for additional inspection during that time - but only if there is risk of the parts getting out of specification. Can you improve the process by
eliminating the warm-up/start up? Very low probability...
Now look at the issue from specifications. Here, there is no debate. This is process is very capable of meeting specifications. That is seen easily from the high Ppk number. But there is no guarantee of keeping it there because the process is not stable.
On the other hand, if all of the special causes were captured by this sample, then it is very likely to stay there. Of course, a 300 pc sample has very little predictive power over the life of the process, so these single-number capability indices are only academic, at best.
Now look at the issue from the customer's perspective. As a customer, I want a product that is produced from a process that is in statistical control and is capable of meeting specifications. I want your process to be in statistical control because that means that the product I am receiving from you is homogenous - made from a consistent and predictable process. If your process is not in control, I cannot be sure of receiving the same homogeneous product from you in the future. And from your data, your process is not in control and you are missing an opportunity to ensure that the variation remains reduced.
Homogeneous product only comes from a process whose output is random and independent variation. Not all processes (such as tool wear) offer that output, so to expect that from all processes defies physics and statistics.
What is even more important is the basis of the specifications. If the incoming component does not fit the processes it is being used in, the specifications seem to be lacking. If they really need to be tighter for the process to work correctly, quit kidding yourself and the supplier with wide specification, tighten them up and pay the price for the correct material.
The variation displayed is already far within specification. If the variation at this point is still an issue, the specifications are fanciful at best.
So, I tend to agree with the Ford fellow. I don’t know if you need to redo it, but you need to understand what happened because I want to get product from a process that is stable. And, in the end, he is the customer.
I agree, you need to understand
why the variation is there, but I also believe all the data you need is there, there just needs to be a better understanding -or explanation - of what occurred in the process at the time.