V
vanputten
I thought that SPC is the application of statistics for understanding variation in the process and that SPC could be the application of any statistical tool and not just run charts with control limits.
In fact, in many people's mind they believe that SPC specifically refers only to Shewhart charts. Can't get much more myopic than that..yet, it exists and persists.
E
Effocus
Hi all,
Sorry that can not reply to you earlier due to the current work load and in fact I need some time to read your messages for understand your idea. I try to use the "in reply to ..." Bev D & Bobdoering separately but do not know how, so I will put here in same space.
In reply to Bev D "There are many valid chart types available for use - not all processes can or should be monitored using the Xbar R chart":
Yes, we do aware that SPC tool sill have many other as 7 tools or even "7 new SPC tool" and I do realized those benefit for using such. But come back to the benefit of very original tool is Shewhart chart and Control charts - I'm still wonder so much energy spent for the little benefit received.
Come back to your explanation and Bobdoering's comment: Xbar-R charts are nightmare to the "precision machines" so can see that this original tool is not so useful - taking into consideration that, now a day, more and more precision machine is used!!! (CNC, CMM machines are replacing those cutting machines in the past already so more and more precision machines are used).
Take another control charts such as p, np or c / u chart in the IAIG manual, you can also see the the "condition" to apply such charts? - onformity per inspection lot need to be >=5 ..."! So you can see that with the current production with the level of "Zero defect" or 6 sigma ... - it is also not so practical to apply such a conditions?
In reply to Bob doering:"
It takes me some time for reading your X/Hi-lo R chart, and I found that It is quite interesting, when you point out the square distribution for tool wear! The normal distribution is not correctly applied. However, I have some questions:
1. In some slide you said that "the natural in-control condition mean the variation to be within the Spec limit" - why? as I know that the in-control or out-control saying only about the special cause exist or not. You may have in-control status even all the value is out-of-spec! As I said control limit determined by the process itself but Spec limit defined by the client!
2. As you define for the diameter measurement is "indefinite number of measurement" - it means you never reach to the Max / Min value yet so the interval that you define - is also only a sample! So we need to come back again to the starting point" - sampling with transforming to the normal distribution?
3. As your Hi-lo chart is solve only for the "tool wear" issue and assuming other source of variation are insignificant, why you do not use very simple tool = "time" or "count"? That is based on the experimental / experiences that the operator will change his blade (cutter) based on time or the number of cuts. In fact, now in those CNC machines those information can be easily obtained and in reality (as I know) the production people do like that for tool changes! Make live easier
for all your sharing!
Sorry that can not reply to you earlier due to the current work load and in fact I need some time to read your messages for understand your idea. I try to use the "in reply to ..." Bev D & Bobdoering separately but do not know how, so I will put here in same space.
In reply to Bev D "There are many valid chart types available for use - not all processes can or should be monitored using the Xbar R chart":
Yes, we do aware that SPC tool sill have many other as 7 tools or even "7 new SPC tool" and I do realized those benefit for using such. But come back to the benefit of very original tool is Shewhart chart and Control charts - I'm still wonder so much energy spent for the little benefit received.
Come back to your explanation and Bobdoering's comment: Xbar-R charts are nightmare to the "precision machines" so can see that this original tool is not so useful - taking into consideration that, now a day, more and more precision machine is used!!! (CNC, CMM machines are replacing those cutting machines in the past already so more and more precision machines are used).
Take another control charts such as p, np or c / u chart in the IAIG manual, you can also see the the "condition" to apply such charts? - onformity per inspection lot need to be >=5 ..."! So you can see that with the current production with the level of "Zero defect" or 6 sigma ... - it is also not so practical to apply such a conditions?
In reply to Bob doering:"
It takes me some time for reading your X/Hi-lo R chart, and I found that It is quite interesting, when you point out the square distribution for tool wear! The normal distribution is not correctly applied. However, I have some questions:
1. In some slide you said that "the natural in-control condition mean the variation to be within the Spec limit" - why? as I know that the in-control or out-control saying only about the special cause exist or not. You may have in-control status even all the value is out-of-spec! As I said control limit determined by the process itself but Spec limit defined by the client!
2. As you define for the diameter measurement is "indefinite number of measurement" - it means you never reach to the Max / Min value yet so the interval that you define - is also only a sample! So we need to come back again to the starting point" - sampling with transforming to the normal distribution?
3. As your Hi-lo chart is solve only for the "tool wear" issue and assuming other source of variation are insignificant, why you do not use very simple tool = "time" or "count"? That is based on the experimental / experiences that the operator will change his blade (cutter) based on time or the number of cuts. In fact, now in those CNC machines those information can be easily obtained and in reality (as I know) the production people do like that for tool changes! Make live easier
for all your sharing!
E
Effocus
Normally (as theory), I know that the control chat Xbar-R can help you to find out the special causes when your process is in control with only common causes. For the capability - normally we use the index Cp / Cpk? So for your example - how much your Cp/Cpk?
Here, possible you have changed some conditions so, the common causes are also reduced?
It is too roughly to me so could you please more specific? Thanks
For the continuous uniform distribution that exists from adjustment due to tool wear, the variation is tied directly to the specification. the variation is not describing random variation as is necessary for a Shewhart chart. So, if the variation is from adjustment, what would you adjust to? Yes...the specifications. That provides the direct connection that you do not find in Shewhart charts. Your statement "you may have in-control status even all the value is out-of-spec!" is true for Shewhart statistics, but not for the uniform distribution. That is one reason why X-bar R charts do not work for tool wear.
It can be a sample, or you can rotate the part and detect the highest and lowest diameter. In any even, a high and low value much better describe the variation than a single point. It is absolutely insignificant, an can not describe a circle. Go back to GD&T. What describes a circle there? Diameter and roundness. Transforming to a distribution that does not describe the process is mathematically possible, but its endpoint is meaningless. As Shewhart sates: "The total information is given by the observed distribution.” The observed distribution is the continuous uniform distribution (when controlled correctly). Transformation is a dangerous tool that should be used as a last resort. Its major flaw is that it masks signals that are important in the process.
Anybody that has machine a part knows that tools do not break according to tool counts. Using tool counts can also have you throwing away tools prematurely - which increases cost of tools and downtime to change them. Tool counts are a guess based on historical perspective. Watching the roundness on an X hi/lo-R chart is a real-time evaluation of that tool, with that lot of material, under its current conditions. Much more powerful and effective. Tool change numbers may seem easier, but actual "pulse of the process" is of more interest of people who chose knowledge over the easy way out.
Last edited:
J
jasonb067
I have not gotten into all of the detail of the conversation but to paraphrase Deming the thermometer does not tell us why we have a fever, only that we have a fever.
We can implement SPC but unless we know how to relate that data to our process and respond accordingly we are all just standing over the process looking at the numbers saying, "yes, it is sick" at best.
At worst we are taking the temperature and do not know how to read the thermometer (SPC data).
I think that the lack of understanding of what SPC chart and data are telling us about our process is the biggest set back we face.
Those most knowledgeable in SPC are doing very little good if they do not understand the process to which it is applied, (unless they are teaching or showing others the concepts).
The most knowledgeable about the process tend to not have the most in depth of SPC tools and all that they are capable of.
When you get both in one person then we can see the benefit fully.
"Quality folks" learn the details of your processes - how it works and why.
"Process folks" learn the details of SPC - how it works and why.
These are not things that can be learned in an 8 hour power point training lesson. These are not things that can be learned by reading a book. There are no short cuts, we must know the data, data analysis and the process.
If SPC is not working then either the process is not understood or data/analysis of the data is not understood. That is it, no short cuts!
No substitute for knowledge!
We can implement SPC but unless we know how to relate that data to our process and respond accordingly we are all just standing over the process looking at the numbers saying, "yes, it is sick" at best.
At worst we are taking the temperature and do not know how to read the thermometer (SPC data).
I think that the lack of understanding of what SPC chart and data are telling us about our process is the biggest set back we face.
Those most knowledgeable in SPC are doing very little good if they do not understand the process to which it is applied, (unless they are teaching or showing others the concepts).
The most knowledgeable about the process tend to not have the most in depth of SPC tools and all that they are capable of.
When you get both in one person then we can see the benefit fully.
"Quality folks" learn the details of your processes - how it works and why.
"Process folks" learn the details of SPC - how it works and why.
These are not things that can be learned in an 8 hour power point training lesson. These are not things that can be learned by reading a book. There are no short cuts, we must know the data, data analysis and the process.
If SPC is not working then either the process is not understood or data/analysis of the data is not understood. That is it, no short cuts!
No substitute for knowledge!
A
Aaidada
Why normality test required while study SPC.
Normality tests (at least if you follow Dr. Shewhart's original work, and as expanded upon by Dr. Wheeler and Dr. Deming) are NOT required for use of SPC.
Beat me to it!
I was thinking maybe we should make this a banner heading on this forum
Last edited:
Similar threads
