InsightFab
Knowledge Base/Common Errors and Prevention in MSA Data Collection
MSA6 min read

Common Errors and Prevention in MSA Data Collection

This article explores the critical impact of distrust in factory data, where 'lying data' can be more detrimental than no data at all. It uses personal experience to illustrate why excellent CPK values are meaningless without accurate measurement tools, underscoring the crucial role of Measurement System Analysis (MSA) in ensuring data reliability for sound decision-making.

That Day, The CPK Report Came Out, And The Room Went Silent For Three Seconds

I still remember many years ago, our factory launched a new product. At that time, the production line just started running, and everyone was on edge, watching the data. When the CPK report came out, the CPK value was surprisingly only 1.08, and DPMO directly soared to 6210. The entire conference room instantly fell silent; you could hear a pin drop. The boss's face turned ashen, and he directly asked, "Is there something wrong with your data?" It was then that I truly understood what "data lying" means, and how it's more terrifying than "having no data." Upon later investigation, we discovered that there was a major issue with MSA data collection.

What Was The Problem?

To put it simply, data collection is to allow us to trust those numbers. Think about it: if the measurement instrument itself is inaccurate, or if different people measure vastly different results, then using this data to calculate CPK or DPMO is essentially garbage. It's like using an inaccurate ruler to measure the dimensions of a wafer; the measured numbers certainly cannot be trusted! The purpose of MSA (Measurement System Analysis) is to help you check whether this "ruler" is accurate and stable. So, the key point is, before analyzing product yield, you must first ensure that your measurement data is reliable.

How Is It Done In Practice?

The most common MSA is GR&R (Gage Repeatability & Reproducibility). Simply put, it involves having several people measure the same sample multiple times, to see how much variation there is in their measured data.

  1. Repeatability: The same person, using the same instrument, measures the same product multiple times, to see if the results are very close each time. If there's a significant difference, it indicates that your measurement instrument might be unstable, or your measurement technique is not precise enough.
  2. Reproducibility: Different people, using the same instrument, measure the same product, to see if their results are very close. If the results measured by different people vary significantly, it indicates that your operating SOP is not clear enough, or the operators lack sufficient training.

Generally, the acceptance criteria for GR&R look at "% Contribution" or "% Study Variation." If your GR&R %Study Variation is below 10%, then congratulations, your measurement system is excellent; if it's between 10% and 30%, it's still acceptable, but requires improvement; if it exceeds 30%, then your data is simply unreliable, and you must quickly rectify the measurement system.

The Most Common Pitfalls

That time I encountered the unfortunate incident of CPK 1.08, I later realized we had fallen into a major GR&R pitfall.

  1. Unrepresentative Sampling: At that time, to save effort, measurement personnel randomly took a few "seemingly normal" products to run GR&R. What was the result? The variation in these products themselves was very small, so the GR&R results naturally looked good. However, the product variation on the production line was much larger! Therefore, the key is that you must select products for GR&R that can represent "actual production conditions," ideally covering the upper limit, lower limit, and mid-range of the product specifications.
  2. Instrument Calibration "Just" Expired: Engineers often forget this point. Before running GR&R, the instrument must be within its valid calibration period. In our case, the instrument calibration had just expired, which led to systematic data deviation. Although the GR&R data looked acceptable, the absolute values of the entire batch of products were off. To put it plainly, even if your GR&R shows high precision, if the zero point of the entire "ruler" is off, it's still useless.
  3. Vague SOP, Operators Improvise: This is a common problem with Reproducibility. Previously, in our factory, there was a veteran operator whose measurement technique differed from what was written in the SOP; he believed his method was faster. As a result, when new personnel came to measure, the data started to drift. So, to put it simply, SOPs must be written very clearly, and regular training must be conducted to ensure everyone follows the SOP.

One Thing You Can Do Today

Check the calibration report dates of all your measurement equipment to ensure they are all within their validity period!

Want to try it yourself?

Every tool mentioned in this article is available on InsightFab — just upload a CSV to analyze.

Go to Tools

More in this category

When Gage R&R Exceeds 30%, Your Cpk Is a Lie

After two weeks of process adjustment, Cpk reportedly improved from 0.98 to 1.35. However, if a Measurement System Analysis reveals %GRR to be 38%, then half of the perceived "improvement" might be an illusion caused by measurement error.

Gage R&R: How Truthful Are Your Measurement Tools?

A new engineer measures a part multiple times with slightly different results, while a colleague also gets a different reading. With a specification of ±0.05mm, this might seem acceptable, but is your calculated Cpk of 0.95 truly a process capability issue or primarily due to measurement system error?

NDC: Number of Distinct Categories – Can Your Gage Really Tell the Difference?

Only evaluating %GRR after completing Gage R&R is insufficient. The Number of Distinct Categories (NDC) indicates how many distinct levels a gage can classify products into, critically determining the usability of your measurement system.