|
3.2.10 Performance Evaluation for Lead-in-Paint Measuring Devices Under Simulated Field Conditions
Eric S. Lagergren
Susannah B. Schiller Statistical Engineering Division, ITL
Mary E. McKnight Building Materials Division, BFRL In January 1995, USA Today reported: ``Federal Housing officials have ordered retests to detect toxic lead paint in 85 public housing projects where hundreds of millions of dollars in tests may have been flawed. At issue is whether tests by portable X-ray machines are reliable." As part of an effort to improve the reliability of lead-in-paint measuring devices, the U.S. Department of Housing and Urban Development asked NIST to identify and quantify factors affecting the field performance of these portable X-ray fluorescent (XRF) devices. The ultimate objective of this study is to develop a protocol for assessing field precision and bias of XRF instruments that measure lead concentration in painted surfaces. The protocol would consist of taking XRF measurements on lead-in-paint standards at a specified set of noise conditions known to cause variability in XRF field measurements. The noise conditions are combinations of settings of ``noise" factors known to cause variability in field measurements. A candidate list of noise factors can be generated. However, the order of importance of these candidate noise factors, i.e., which cause greatest measurement variability, is currently unknown. A laboratory experiment was conducted to identify the most important noise factors. In this experiment, the noise factors were systematically varied according to a statistical experimental design to study their effect on XRF measurements. Of course, the conclusions from this lab experiment must be validated in the field to ensure that all important sources of variability have been captured. In this experiment, eight noise factors were studied at each of two settings using a 16-run (out of a possible 128 = 28) fractional factorial design. This well-chosen subset of 16 runs permits free and clear estimation of the primary effects of all factors and limited information on two-way interactions between factors. The 16 noise factor conditions were studied for each of four XRF instruments and two lead concentrations.
The figure shows that x6, the distance of the instrument from the
surface, is the dominant noise factor, followed by x3, the
underlying substrate (wood or steel). Substrate has a large
effect for only two of the instruments, indicating that the other
two invoke a substrate correction. The ``distance-from-surface"
noise factor was included to simulate non-flat surfaces such as
wood molding, metal pipe, and stucco. A second experiment is
being conducted to assess whether ``distance-from-surface" is an
adequate surrogate for non-flat surfaces, and if so, which
distance best reflects the induced variability. The experiment
examines non-flat surfaces for several substrates using a full
factorial design.
The results from these experiments will be used to develop a test
protocol for assessing field precision and bias of portable XRF
instruments.
Figure 20: Plots of mean XRF response versus factors for instruments from four manufacturers (columns) and two lead levels (rows).
Date created: 7/20/2001 |