In Our MSA stsudy we are performing Linearity and Bias measurment. We are currently controlling at Linearity and Bias to be below 5%. We actually generate the following report for Linearity and BIas. However in AIAG handbook the criteria for the gauge to pass bias criteria has the following criteria in the picture attahcment
I have the following questions
1. We have set the linearity and Bias guidelines as less than 5%, however no one can identify how we derive the 5% criteria. Any guidelines on explanation why its 5%?
2. In the linearity and bias script there is dialog box where we enter a process sigma value- what should be this value? We are currently keying in the tolerance (USL-LSL) is this correct?(See powerpoint slides)
3. In the powerpoint slide case1 and case 2 - Case 2 the zero and bias lines are not within the confidence interval but its meeting the Bias% and Linearity% of less than 5% is this correct and accepatable?
Joanne Wendelberger's article about measure uncertainty in her article "Uncertainty in Designed Experiments" , Quality Engineering (2010), Vol 22: 88-100, where she gave some historical context and formula derivations. Another source is W.J. Youden's Experimentation and Measurement, (1997, originally published in 1962), NIST Special Publication 672, U.S. Department of Commerce.
I don't have a good answer for your first item.
For item two - the value you should enter in the dialog should not be the tolerance or spec limits. It should be a historical process standard deviation - perhaps from a control chart on the process. This number will be used in the calculations of the figures of merit.
For item 3 - Case number 1 the bias slope is not significant, so you can say there is no bias. For Case 2 - the slope of the line is significant - but it is is also not recognizing the non linearity in the data. The first 2 standards at 5 and 10 are close to 0 bias, but the third standard at 25 is high, and the last standard at 100 is low. This line is not a good fit to this data - there is a quadratic effect.
Case 1 looks okay - case 2 not so much.