How is that Cpk calculated? There must be a problem with the measurement system!
Remember last month, when the yield of our new machine line just wouldn't improve? Process engineer Xiao Chen came to me with a grim face, holding a Cpk report that read 1.08. He asked, "Brother A-yao, how is this Cpk calculated? I really don't think our machine is that bad! Is there a problem with the measurement system?" I glanced at it and thought to myself, this kid finally gets it; he knows not to just look at Cpk, and that the stability of the measurement system is key. That's when I knew it was time to talk to him about the latest version of the AIAG MSA manual.
Where's the problem? Your measurement system might be a liability
To be honest, many people only stare at numbers like Cpk 1.33 or 1.67, forgetting how these numbers are derived. Think about it, if you use an inaccurate gauge to measure products, what good is beautiful data if it's based on flawed measurements? That's simply self-deception. If your measurement system fluctuates too much, or if there's too much variation between measurement personnel, the data collected is just garbage. Can any improvements made to the process, based on this garbage data, be effective? Therefore, the key is to first ensure your measurement system is reliable before evaluating process capability. MSA (Measurement System Analysis) is the tool used to assess this reliability.
How is it actually done? The new manual gives you more relaxed standards
Frankly speaking, after the update to the MSA 4th Edition, the most significant change is the more flexible acceptance criteria for GR&R (Gauge Repeatability & Reproducibility). Previously, everyone was quite stressed by the 10% threshold, but now it's much better.
- Adjusted Acceptance Criteria:
* Now: GR&R < 10% is certainly ideal; if it falls between 10% and 30%, it indicates "may be acceptable," but you must have sufficient justification, such as high cost or the measurement process itself being prone to variation.
* If GR&R > 30%, it is still "unacceptable," and your measurement system definitely has a problem that needs immediate improvement.
- More Explicit Evaluation of Bias: The new edition places greater emphasis on evaluating bias, not just repeatability and reproducibility. Bias is the difference between your measured value and the true value. Think about it: if your gauge consistently measures an extra 0.01mm, even if its readings are always very close to each other, it's still not accurate! Therefore, bias analysis is now given more importance.
- Distinction between Continuous and Attribute Data: The manual more clearly distinguishes the analysis methods for these two types of data and provides more detailed guidance. For instance, when performing MSA for attribute data, it suggests more frequent use of Kappa values to assess the consistency of judgments among different appraisers.
In other words, the new MSA gives you more flexibility to determine if your measurement system is suitable, but at the same time, it reminds you not to only focus on variation, as bias is also very important.
The most common pitfall: Unclear sample preparation and evaluation items
The most outrageous incident I encountered was during the implementation of a new process. When running MSA, the GR&R result was astonishingly high at 45%! Everyone was terrified, thinking the machine was completely unusable. Later, upon closer inspection, it was discovered that there was a major problem with sample preparation. To save effort, they had only prepared 5 samples, and the difference between these 5 samples was very small, almost identical. Think about it: if the samples themselves have little variation, no matter how precise the measurement system is, its "ability" to distinguish differences will be difficult to highlight, and naturally, the GR&R will be very high.
Simply put, the purpose of conducting MSA is to evaluate whether the measurement system can distinguish between differences in products. If the samples you provide have no inherent differences, then naturally, it cannot "distinguish" anything, and in this situation, the measurement system will appear to be very poor. Therefore, the samples prepared must encompass the entire range of variation that might occur in your process, and they must be selected randomly.
Another common pitfall is focusing solely on GR&R while neglecting bias and linearity. As mentioned earlier, if your gauge consistently measures a little extra, even if the GR&R is good, your data will still deviate from the true value, which is unacceptable. Therefore, when evaluating MSA, bias, linearity, and stability must all be considered together; none can be neglected.
One thing you can do today
Re-examine the MSA reports for commonly used gauges in your facility to see if they align with the spirit of the new manual.