Como você mede e melhora a capacidade do seu processo usando o índice Cpk?

If your reading this and thinking 🤔 Cpk 🤷♂️? That is alright! Sometimes people think they need to be … fill in the blank - mathematicians, statisticians, or data scientists - to be able to understand and apply Lean Six Sigma to improve processes. The good news … you don’t need to be any of them. Additionally, if you are working in the service industry, you can be applying what you learn in Lean Six Sigma and drive real value without knowing about or leveraging knowledge about Cpk. There are varying levels of expertise and proficiency in the world of Lean Six Sigma & Cpk is not something we typically teach at the introductory certification levels such as Yellow Belt. Matt Rebeiro - Challenge the Status Quo #LeanSixSigma #lss #cpk

At the real electronics manufacturing with high-accuracy and high-volume, there are 2 standards in fact, once for machine buyoff Cpk 1.33 (4sigma) with the specs limit of +/-25um and another one for production criterias Cpk 1.33 (4sigma) with the specs limit of +/-50um. The hardship here is machine buyoff criteria, in order to achieve this standard, we have to remove all posible tolerance from materials, carrier/tray mechanical shift, measuring method, measuring operator, operation tolerance, etc. Once we take away the external factors which are able to affect the final accuracy, we can focus on our main system's accuracy.

Another area to also consider is the limits specified for the critical characteristic we are measuring. A goal of CpK=2 is great, but if the process is not giving that result then review your limits. In leading APQPs I have come across this situation of not getting a minimum CpK of 1.33 and one of the things the team reviewed were the limits.

Not just "realistic," but if your Cpk is greater than 2 are you not over-processing and producing muda? Perhaps your customer requirements are being adequately met, but when your Cpk is too high then you deliver excess quality. Such a thing is possible and represents one of the classic "eight wastes."

When Cpk is very large, you run the risk of Passing Outliers. Q: Which process would you prefer? One that always meets the spec and each unit is indistinguishable OR a process that always meets the spec though parts can vary widely? Clearly either the spec (most likely) or the precision of the measurement system is the issue here!

When comparing the Cpk value for product quality, also check the additional charts to see where the product nominal specification is in relation to the actually produced part nominal. The data doesn't have to always be centered and a high Cpk can still be obtained. If the data isn't centered and it shows a wide spread, the Cpk is low & a tiny spread Cpk should be high.

The general theory at the left side almost covers all factors which contribute to the final Cpk index. Let me share some practical situations as followings 1. The best you only measure with one golden sample (this is not actual product), then you can eliminate the tolerance of samples. 2. Ideally, only one operator does this measurement as there is a tolerance between human to human for sure if we measure with a manual measuring equipment. 3. The number of data should be 33 at least, if less than 33, then you have to measure many samples to achieve a desired Cpk 4. In the worst case, you can measure with the actual products. Then firstly, please check the tolerance of all samples and categorize them into some common type to select good ones

The larger the sample size one has will grant a more accurate picture of the Cpk value. Lower sample sizes (like 30) are a good way to gauge where variation may be coming from. Variation also comes from part nature like a torsion spring versus a compression spring or when tested for a Gauge R&R. One big factor is corporate culture and how honest the inspector is recording data. In that event it is wise to mandate a comment on why the data is trending the way it is and normally on a tighter pattern such as 4-5 data points opposed to the advised 7 or 8. Additionally, state based changes can affect the Cpk such as a material change during the study that derived the original Cpk value.

When the value of Cpk is less than one it is advisable to create a good quick process to eliminate larger cost potentials such as the machine. Typically the findings of a low Cpk are from poor training of proper measurement methods.

Please be careful that Cpk index only doesn't mean anything, we have to see it with the specification limits. For example, Cpk is very good with the production accuracy criteria of +/-50um. But if we see it with the machine buyoff criteria of +/-25um, maybe if it's not acceptable at all.

This is a great article but a perfect example of why some folks just put their blinders on when the higher-ups bring up the Lean push. This is a discussion for those Green belt and above and then the Green and Yellow belt folks need to get it to the point where those below and to take part in the process. When people have zero bandwidth in their day-to-day work cycle and they are told to take part in the process, they most likely won’t. All they hear are new terms that they have no time for but hook them with an idea of what’s possible and they will make time to work on it in the background. It’s not about simplifying or dumbing down an end-goal, it’s about showing them a gold nugget and make them want to find the mother-load!

Along with the Cpk value, is the Ppk value which further helps guide how the results gained tell how well a process works. Ppk is similar to Cpk and looks at the process slightly differently. The gap between these two values further indicate the stability and capability of the process. It is advisable to not just solely rely on Cpk alone.