Need help in quality control
February 19, 2003 at 4:42 pm #31508
ParameswaranParticipant@Krish Include @Krish in your post and this person will
be notified via email.
Here is the problem:We had given a contract to a vendor to make CDROM sets (1 set consists of 2 CDs) of our training module. He delivered 7000 sets (total 14,000 CDs). These came in small boxes of 25 sets (50 CDs); eight such boxes are placed inside a bigger box (total 400 CDs). Therefore in all there are 35 large boxes and 35 X 8 = 280 small boxes. When we started to examine the CDs we found the several CDs did not work. Each CD takes 2hrs and a set takes 4 hrs to examine for errors and the errors appear at random. So it is impossible to verify each CD or set for errors. How should I determine a method to reject a lot of either the small box or large box whichever is more appropriate? Should I go in for a simple percentage of bad CDs out os small sample (how large should that be?) and extrapolate to the entire lot.Please remember that even a sample size of 100 sets (200 CDs) would take 400 hrs for verification. Also I concede that there is no way I can segregate good and bad CDs and I can only refuse payment for rejects and ask for free replacement whenever a bad CD is detected.Other details: I have no idea of population mean or std deviation of bad CDs.Please help ASAPKrish0February 20, 2003 at 4:28 am #83118
Michael SchlueterParticipant@Michael-Schlueter Include @Michael-Schlueter in your post and this person will
be notified via email.
[Moderator’s Note:This forum post contains formulas that may not appear. Mr. Schlueter was kind enough to attach a Microsoft Word file — see below.]Dear Krish, What will happen, when you pass the training CDs to your customer without checking them? I suppose the answer depends:· When your customer can bear waiting for a replacement, his/her monetary loss will be low (transport + materials).· When he/she has to send trainees home, his/her monetary loss can be as high as several k$. Can you estimate the average loss to your customer? That is that amount of money, 50% of your customers would loose when you ship defective CDs. (Ill give an example below). Reviewing the errors youve found, specifying error-tolerances:What do you think, how many errors will your customer accept on average?(tolerance D: 50% can bear this amount of errors, while 50% cant.) Lets have a look at a low-loss scenario: Bearable error-tolerance D = 10 errors (just my wild guess) Cost of a defective CD (its price) A = 3$ (just my wild guess)Cost of measurement B = 2h * 40$ = 80$Cost of adjustment (=replacement) C = 8$ (price + shipment) Current measurement interval n0 = 1 (you measure each CD)Current control limit D0 = 1 (1 error => CD rejected)Current adjustment interval u0 = 10 (just my wild guess) Background:I suppose that you found both critical and minor errors. Try estimating a 50% tolerance value: 50% trainings will be ok, 50% will not be ok at that error level.All costs are average cost per CD.My assumptions: so far you measured each CD, rejected it when there was at least 1 error. You probably had time to investigate 10 CDs so far.The underlying production model (you receive and ship CDs) assumes:You measure every n0th CDThis imposes an average measurement cost BYou readjust your process on target (=ask for replacement), when your measurement result exceeds the control limit D0This imposes an average adjustment cost CYou adjust (=replace) after every u0th adjustment (=replacement), when necessary (control limit D exceeded)On average A$ are incurred by a defective CD. The average monetary loss L you pass with every CD to your customer is in this process-adjustment model: (looks more complicated than it is) In the low-loss scenario this evaluates to: or = 80.98 $ As you can see, in this scenario the cost of measurement dominates, while the loss incurred to customer is very low. Can we optimize it? Can we minimize this total monetary loss? Better measurement interval n: Better control limit D: Better adjustment interval u: Which yields for the low-loss scenario: n=230D=2.9 ~ 3U=29 ~ 30 It is economically justified in the low-loss scenario to:· Measure each 230th CD· To raise the control limit to 3 and pass it with 0,1,2 or 3 errors· To ship all the other 229 CDs without measurement (low loss to customer from defective CD`s)· To adjust (=ask for replacement when D is exceeded) every 30th CD(which in fact means coupling replacement to the measurement rate) Average monetary loss will reduce to: or = 1.75 $ per CD We can expect the total monetary loss to decrease by: Lold = 80.93 $(Lnew = 1.75 $ )gain 79.18 $ per CD Amazing, isnt it? Just by tuning your measurement and replacement process. Here is a high-loss scenario: Bearable error-tolerance D = 10 errors (just my wild guess) Cost of a defective CD A = 20000 $ (training canceled)Cost of measurement B = 2h * 40$ = 80$Cost of adjustment (=replacement) C = 8 price + shipment) Current measurement interval n0 = 1 (you measure each CD)Current control limit D0 = 1 (1 error => CD rejected)Current adjustment interval u0 = 10 (just my wild guess) (looks more complicated than it is) or = 167.40 $ per CD Adapting the intervals:Better measurement interval n: Better control limit D: Better adjustment interval u: Which yields for the high-loss scenario: n=2.8 ~ 3D=0.58 ~ 1u=10 It is economically justified in the low-loss scenario to:· Measure each 3rd CD· To leave the control limit at 1 and pass it with 0 or 1 errors· To ship all the other 2 CDs without measurement (take a small risk)· To adjust (=ask for replacement when D is exceeded) every 10th CD(which in fact means potential replacement by your supplier every 3rd measurement) Average monetary loss will reduce to: or = 114.07 $ per CD We can expect the total monetary loss to decrease by: Lold = 167.40 $(Lnew = 114.07 $ )gain 53.33 $ per CD which origins completely from savings in measurement in this case. There is some risk of passing defective CDs to customers, who are urgently waiting for training material. However, it is not justified to overdo measurements in this situation (shared risk). I think you can adapt it now to your situation. This approach is heavily used by Mazda in manufacturing. I took the formulas from: Taguchi on Robust Technology Development Bringing Quality Engineering Upstream, G. Taguchi, ASME Press, New York, 1993 Hope this helps,Michael Schlueter (A western member of the Quality Engineering Society, Tokyo) Download: Letter from Michael Schlueter [Word file]Viewing Tip: Usually, you can click on a link to view the document — it may open within your browser using the application (in this case Microsoft Word). If you are having difficulty, try right clicking the link and selecting “Save Target As…” or “Save As…” to save it to your computer harddrive.0February 20, 2003 at 1:11 pm #83132
MstewartParticipant@Mstewart Include @Mstewart in your post and this person will
be notified via email.
Send them all back and let your supplier deal with it !!!!0February 20, 2003 at 2:32 pm #83144
Anne Ladenson, Disc MakersParticipant@Anne-Ladenson,-Disc-Makers Include @Anne-Ladenson,-Disc-Makers in your post and this person will
be notified via email.
This might seem simplistic, but I would recommend returning to manufacturer to do this whole evaluation.0February 20, 2003 at 4:29 pm #83149
Theoritically speaking, it is an excellent problem that you are facing. Practically speaking, I could imagine the nightmare! You acutally have a number of solutions paths. My first suggestion is for you and whomever is helping you with this problem, to sit down and put together a decision tree. You can start by the two extereme decisions: Accept as is OR Reject the whole batch. You can then perform the cost/benefit (rough calculations) for each decision. One of the respondents gave you a very good model to follow. In reality you are asking yourself: “what is the risk with this solution path.”
You can also consider the inbetween decisions; i.e., as you suggested, perform some “quality” checks yourself. For each decision, you will have the cost, resources, etc – then you can make the appropriate decision. One suggestion: look at your contract – you might be stuck with a certain decision due to verbage in your contract. That is, your contract might be written such that you may only return the “defective” CDs and not reject the batch. Using the constraints you have (contract) and the potential risk/benefit, make your decision.
Now in regards to your sampling for “defects” in the CDs. I don’t know how much you know about your supplier’s process; mainly, is there a logical sequence to their packaging (first CDs in the first boxes, etc.). That might help with your sampling plan. If the answer is no, then you might want to consider the “Acceptance Sampling” method. In the acceptance sampling, you will accept or reject the lot based on limited observations. What you can not get is the actual quality level of the lot. It is fast, but as anything that is quick, it provides you with little knowledge about your product. Again, this works if you can reject the lot and send it back to the supplier.
The other method is to perform sampling to determine the quality level of your product with a certain confidence interval. There have been numerous postings on calculating the confidence intervals.
I wish you success.0February 21, 2003 at 4:45 am #83174
John H.Participant@John-H. Include @John-H. in your post and this person will
be notified via email.
As a first step, you should narrow the scope of the problem and the vendor must work with you on this. Some sample questions: Is there any marking traceability(CD Lot code, dates, times) that traces the CD’s packaging etc.. to the vendor’s production runs, disk burners, operators..?Was the Master ok or was it damaged?What type of Quality Specifications were listed in the contract and did the vendor follow Best Industry Practices? What is the acceptable defect level in the industry? Is there a better* way of checking the CD’s ? A superior supplier could guarantee CD critical defects in the PPM range and a rejection based on what you found would be obvious in this case.
In theory,I can appreciate the benefits of a risk cost benefits analysis. However, in the real world, elegant risk reduction mathematical models can often fail with disasterous financial consequences(example: Derivatives Trading). In lieu of the bad press that American Industry is getting with repect to financial reporting honesty, I don’t think that it is a good idea to pass off a quality problem to a customer because it is profitable to do so.
This is a difficult problem.
*Perhaps the NIST, DOD or other Government Technical Agency could assist with this.0February 21, 2003 at 6:18 am #83177
Dr. Steve W.Participant@Dr.-Steve-W. Include @Dr.-Steve-W. in your post and this person will
be notified via email.
of defective rate (1%, 5% etc.) is acceptable first. With that established, you can compare it with your confidence interval for the proportion of defetive parts based on the out come of your test. The confident level dictates how many you need to sample. This is a traditional inspection issue and any well trained statistician can help you.0
The forum ‘General’ is closed to new topics and replies.