How to measure process quality? Is DPPM the correct metric?

[juan_mexia]

Hello all,
I've been challenged by my boss to review if what we are doing to measure our process in correct.
Currently we are using DPPM. The problem I have with the way we do it right now is that we calculate DPPM = total number of defects / total number of produced units * 1 000 000. This way we can have more defects than produced units so the DPPM number will be higher than 1 000 000.

For example, if I want to build 10 cables with a terminal and the machine that crimps a terminal into a cable throws away 10 terminals during the process then I will have 10 defects and 10 produced units so the DPPM will be 1M.

Is this correct? It seems to me that this should be trated in a different way.

What do yo think about this? Any help will be appreciated

Thanks,
JP

 

[Steve Prevette]

Yes, if you attempt to produce 10 items, and all 10 are rejected, then yes, the failure rate is 1 million per 1 million.

However, DPPM is not a very high quality indicator of process quality. It is a go - no go piece of data. An item just shy of the specification is "good" and an item just on the other side is "bad". DPPM offers the ability to lump a bunch of data together, but I would refer proponents to the following road sign (https://en.wikipedia.org/wiki/File:New_cuyama.jpg)

Much better are control charts measuring the actual data quantities related to the specification to see how much variation there is. It is also possible to accumulate across processes by using techniques such as chi square or Hotelling's T square (https://en.wikipedia.org/wiki/Hotelling's_T-squared_distribution)

 

[Steve Prevette]

P.S. Also - there are significant items that affect process quality that should be measured up stream, and also should understand the downstream effects of process quality - such as the part may be "in specification" but the customer found it to be useless.

See also https://www.in2in.org/insights/Bellows-ReflectionsontheFabricofTPS.pdf

 

[juan_mexia]

Thanks for the reply Steve.

So if I have 10 defects produced during the process and I at the end have 10 good units, how the DPPM should be calculated? I understand that I had 10 defects but shouldn't the denominator of the DPPM formula need to be changed?

I don't quite understand the logic of why the process can have more defects than the total of produced units.

In other hand, I do not believe we are going to change from DPPM to other quality tool to monitor our processes (at least not soon), so I am looking a more reasonable way to calculate the DPPM level, if there is.

Regards,
JP

 

[Steve Prevette]

Generally the DPPM model is used for - I attempted to make a product, and I could not send the product to the customer for a gross failure.

So, if I attempted to make 20 items, and only 10 could be shipped to the customer, this is a DPPM of 500,000 (1 of 2).

When we talk defects though, not all defects are the same. Perhaps you made an item and there was a scratch in the paint. If either you shipped it anyway figuring it would not affect the customer, or you buffed out the scratch and repainted the item, this is not a complete failure of the product.

Now, it is well worth while keeping track of minor defects, and the rework costs to fix the minor defects and ship them on to the customer.

 

[Bev D]

It actually sounds like the metric you are looking for is Rolled Throughput Yield (RTY). this is used to track defects internally to the manufacturing operation while DPPM is used for defects found post shipping. It is especially well suited to products that take more than a day to produce and can have multiple defects at multiple steps of the process...

RTY includes defects that are reworked and those that result in scrap. RTY will vary from 0 to 100%.

The two best articles on how to calculate this are:

Tom Pyzdek's Article "Yield the Right Way" in the March 2000 issue of Quality Digest

Spencer Grave's article "Six Sigma Rolled Throughput Yield" in the December 2001 issue of Quality Engineering. this is free for ASQ members who subscribe to QE. If you don't it's $10 - much cheaper than buying ad hoc from QE...