Exactly when a process would be unstable? The measurement system analysis is said to be ineffective if total gauge R&R is more than how much? Is it...80%, 50%, 95% or 90%?

These seem to be two different questions.

1. MSA is ineffective if total gage R&R is more than 10% of total process variation. In some cases we may tolerate up to 30%.
   
2. A process is unstable if special causes are found to occur too often. A run chart or a control chart may be used to assess stability.
   

Process stability can be judged only when you are sure you have a reliable measurement system in place. Hope this helps.

how does affect the data that in a gage R&R you try with 5 operator with another gage in what you only work with 3 operators?

To state it simply, the more assessors you have, the harder it is to achieve an acceptable measurement system. 

Hello Team,

I am executing a GB project on Productivity improvement. All the required data are generated by the application thorough query to mine from application itself. in this case, do we need to do a Gage R&R Study? if Yes, what results should be considered as Operator and which result should be considered as Part to Part?

Hi Giritharaprasad, 

 

You are right in saying that Gage R&R is not needed. You do need to carry out logical validation though. Basically you need to verify that the system generated data -

 

1. Uses a validated method of data capture (error free and timely)
2. The data being recorded matches the definition of the CTQ that you have selected in your project. 

 

Thanks much VK.

 

Hello Team,

 

[Measurement system analysis-AIAG]

In Bias, we just compare measured value with reference value to see significant different by hypothesis test, but I think bias must be compared to product specification for better evaluation (like GRR value). In some case, the bias is large (hypothesis fail) but the specification is too high (Ex: bias/specification < 10%) so can we accept bias in this case ? 

 

Regards, Duy  

Hi Duy

 

There are two parts to your question. Let me try answering it part wise.

 

Part 1 - whether bias should be compared against the reference value or the specification? 

During a Gage study, bias is observed in the complete range of the gauge with repeated observations. Simply stated it gives you an indication of the central tendency of the observations. This central tendency of the observations should be compared with another central tendency which in this case becomes the reference value (you may also understand the reference value as the target value). The problem with comparing it with specification is that it will not be like to like comparison. Let us take an example - suppose you are doing a gage R&R for measuring the inner diameter of a bolt. The reference value (or target) is 2.5 mm and specifications are +/- 0.1 mm. You picked a sample of 5 such bolts and did 12 trails with the operators and the study suggests that there is a bias of 0.05 mm i.e. the observed value is greater than the actual by 0.05 mm. Now if you compare it with the target, you are still within specification, however if you compare it with the specification, you might end up rejecting good bolts as well. Hence, the bias should always be compared to the reference or the target value and not with the specification.

 

Part 2 - In cases where bias is large (>10%) and bias/specification < 10%, can the bias be accepted?

Ideally speaking you should NOT accept the bias if is greater than 10%. You should first fix the measurement system. However, practically, in some projects, you may choose to accept the bias under the following conditions

1. There is only 1 sided specification limit (because in two sided specification limits, if you do not address the bias, it will start impacting the process capability as your reference value is improved)

2. Specification is too high even for the revised reference value (or the revised target) + the bias

E.g. Let us assume, the moisture in a powder compound cannot exceed 13%. Your current reference value is 8% and the measurement system has a bias of 1%. If your target is 10%, you may choose to ignore the bias as you will not breach the specification of 13% with 1% bias. However, if you target is 12.5%, then you will definitely need to fix the bias before improving the process

 

Hope this helps.

 

Thank you very much for your information. I really appreciate that you spend your time for my problem. I'd like to explain my real case for example:

We use measurement system 1 to evaluate our product. After some years, we upgrade our measurement system  with new software (reduce cycle time - measurement system 2)  and we want to compare the different between measurement system 1 and 2. Hear is the  method:

1. we measure golden sample by measurement system 1 to find reference value , the reference value is 98

2. We measure golden sample by measurement system 2 to find bias (MSA method from AIAG), the reference value is 98.7, standard deviation is 0.65. 

After using hypothesis test, we see that hypothesis was rejected, that mean two measurement system has bias and need to calibrate. But the specification of product is from 93 to 103, so we see that is, with respect to the specification limits, the potential to make the
wrong decision about the part exists only when the measurement system error intersects the specification limits, and this is not in this case (we measure 100 sample to see distribution and calculate Cpk >1.3--> the distribution capability is far away from specification limit). So we continue using new measurement system although fail hypothesis test. Is it correct ? If not, can you give me better solution in this case ?