Lisa M. Gill
Statistical Engineering Division, CAML
Precision Engineering Division, MEL
This reference material consists of 1cm X 2. 2cm X 3cm steel block that are manufactured outside of NIST. Each block is coated bye the electroless nickel deposition process. Then the specimens, approximately 25 at a time, are put onto a large drum. While the drum is turning, a single-point diamond tool machines a sinusoidal roughness profile on the top surface of each specimen using a numerically controlled lathe.
Each drum of machined specimens constitutes a batch; for this SRM, there are 4 batches. The specimens are then sent to NIST to be measured. Every specimen is measured individually for roughness, R, which is the average absolute deviation of the surface peaks and valleys about the mean line, and surface wavelength, D, which is the average period of the sinusoidal surface profile. Profiles for the certified values for R and D are based on measurements taken at 9 different positions on each specimen.
Ninety specimens were certified. SRM 1104 was designated as the check standard for this SRM. A uncertainty component for instrumental imprecision was estimated from this data. In addition, a between-occasion uncertainty component was determined using the check standard.
Roughness was certified individually for each specimen as the average roughness over the 9 positions. The uncertainty for the roughness measurements included a variance component for the between-occasion variability and the measurement error. The estimate of within- occasion variability was estimated from the individual specimens and the check standard and then pooled.
Wavelength was certified collectively for the specimens as the grand average of all the data across all specimens. There was a slight quadratic curvature over the rows, and a slight linear trend over the columns. This structure was negligible compared to other sources of variability. Although the position differences were not practically significant, they were statistically significant; therefore, to properly estimate the uncertainty, a model was fit that included variance components due to position, an interaction between batch and position and measurement error (within-occasion).
A 95% prediction interval was calculated for each specimen for roughness, and collectively
Figure 5: This figure is a plot of wavelength versus position versus specimen identification number. It reveals 3 specimens that contain wavelength outliers.
Date created: 7/20/2001