Control Charts + Ishikawa

[Your Soap]

Hi again,
In attachement I have added a sample of the bad piece of bolt by its ''construction'' that belongs to Throw Out. This kind of defects doesn't belong to dimension errors. I think I found a good example in one of the Throw Out boxes. Please ignore the brightness which isn't equal on the whole surface - its because of pictuing from my cell phone.
What I would like to know is the following; When this or any other ''not-dimension-related'' bolt appears on the machine, the worker will always put in the box with bad pieces the whole part of range but only the part after the latest quality check has already been made.
In the attachment is just one example, there are many others that can appear.
My question is; how do I put the datas of those kind of samples in the control chart? From the technical drawing specifications (the specifications that we measure - they are given on the drawing with tolerances), I cannot use any value. Actually in this case the defect cannot be determinated by wrong value from measuring because it is determinated by ''free-eye'' without any measuring instruments. When I take the samples for control charts, I need to type the values but what should I type in this or any other similar example? There is nothing to type.

 

[jasonb067]

I know that you are looking for a direct answer. I am sorry but I can not give one, hopefully this post will allow you to answer some specifics that will allow us to help.

1) What is wrong about the part in photo?

2) What is wrong about other non-dimensional "Throw Out" parts?

3) What is wrong about dimensional "Throw Out" parts?

I think if you are having too much "Throw Out" to review each single part you are going to have to sample the "Throw Out" bin population. There are many ways to determine how many or what % you need to review.

From there, and getting back to the point made earlier in this discussion pareto or otherwise understand what is "wrong" with the parts in an analytical way. You will find that a control chart may or may not apply to these failures directly as they appear on the part. In other words, each failure may not be called out on the drawing as x +/-y.

But

You will hopefully find that you can assign one or two or more process variables (example only! air pressure (I have no idea how to make a bolt)) which are creating these "Throw Out" parts. Capture the data for those process variables. Understand what variable value makes good parts, what variable data makes bad parts. Set up control limits for those process variables and then we have process monitoring.

You can not collect data 24/7. The operator or workers need to collect (now), understand (soon) and respond (one day off in the future) to this data.

I hope that I understood what you were asking well enough to give you an answer.

I hope that I explained myself well enough to be helpful. If this is not helpful at all then I am sorry I wasted your time reading it.

By the way, none of what I described above will be easy...

It should be kind of fun though.

 

[Bev D]

if I understand your question correctly...
you are dealign with two types of data: continuous (that wich can be measured - such as the length, width or weight) and categorical (that which can only be counted - such as your picture).

For counted data an appropriate control chart is a "p chart", a chart that plots the proportion defective (or thrown out for visual reasons).

 

[Your Soap]

jasonb defect was on the top of the bolt as visible from the attachent in my previouns post. I searched a lot in our manufactory similar bad bolt so I ask(ed) my question based on it, even if I could ask without it. Defect is just one example, similar defects, that cannot be found through measuring, are houndreds of others. Whats wrong with it? In the center line, on the left side, the material isn't correctly cutted (ignore the brightness error please - its because of my cell phone) and goes a way too far towards the top's hole of bolt. Right side is good.
In the the same producing range (same day / same bolt type) the worker always gives to the box with bad pieces (throw out) entire group of bolts from latest quality that has been checked. Its because of a high level of possibility that many bolts in the range will have similar defect. But on my (and not worker's) side, it is also high level of possibility that I will take the samples, located in throw out box, of bolts with defects, definited by ''free-eye'' (without dimension errors that can be measured). So I had to ask that question. If you asked about different range (which doesn't neccessary mean different type) of bolts, there can be many other defects such as too high or too low angle (if planned) at the end of rod (the most lowest location of bolt).
What is wrong with dimensional parts? They can be out of tolerance that is still allowed on the drawing (specification limits).

There are many ways to determine how many or what % you need to review.

Could you please suggest me any? In whatever calculation I try to do, I always get a way way too high, almost impossible to do, amount of samples to check. Some articles, found through google, ''mentoin'' that I have to check 1% of entire package to the customer. So lets say the customer ask for 2,5 million bolts. I work 8 hours per working day. Since I will take samples twice per day, there is absolutelly no way that I could analyze 1% out of 2,5 million for 10+ specifications + 9 * 1% of other bolt types (analyzing 10 machines with the most bad pieces) + 10 * 1% of rolling processes (different location of manufactory) in maximum 8 hours. I haven't calculated the sample size yet. The static number 50 was just a guess based on nothing. In few articles I have also seen that some parameter which depends on sample size calculation is a matter of biggest and lowest limit of articles found. But this makes no sense to me. Since we are determinating a sample size how can we know what is the lowest and what the highest dimension? For knowing this, we need to take the samples. But since we are trying to determinate how many samples, its logical that nothing can be taken (yet).
This is totally different topic comparing to what I asked in my previouns post but since you typed the sentence that I quoted, I had to add this part of text also.
jasonb you said to ''capture the data for those process variables''. Data and Variables? Which data? Actually there is nothing to ''capture'' in those kind of defects. If I take some amount of bad pieces but measuring dimensions show that they are good means that similar ''not able to be measured'' defect appeared. I should NOT try to find the bad pieces for dimensional errors so I can get out of tolerance values just to type something in the tables for control charts. I must care for the samples that I take and because of this I asked the question in my previouns post but I don't know what to type and where to. Nothing can be measured in this situation but I still need to get some analyzing results.


Bev D this was exactly what I wanted to ask yes. I was asking how (and where to archive them) to determiante the results (no obvious numerical datas given). Thank you for suggestion regarding the p control chart. Regarding this tutorial found through google for doing the test of P chart I need to know the entire number of all pieces from the range where defect that cannot be measured (as in my attachment of previouns post) appeared. This is definitely not possible. Because it can happen that the bad pieces from specific part of range end up in the box mixed together with previouns bad pieces. Also counting everything isn't possible because the worker can also put in the box several thousands of bolts if one similar defect appears.
Would be pareto diagrams useful anyhow?

 

[Marc]

Any comments on Post 24?

 

[jasonb067]

The only question I can answer is yes, pareto what you are finding. Acutal bad parts by what is wrong with them.

Start to understand what in the man, method, material or machine can have an impact on the highest one or two (it seems that you are over loaded so just take on the number you can manage). Once you know what can have an impact assign controls to them one by one and see which one or combination of any to all work. Repeat on the next one or two top issues until it becomes more managable.

I know that this is not the best advice, but, it seems that you are way overloaded. Sample what you can from the "throw out" and use that information to create your pareto.

I hope this helps.

 

[jasonb067]

One other thought. I may not be fully understanding. But from what I think I understand it seems that the process is too far out of control to worry about control charts.