(Received 14 July 2003; accepted 5 March 2004)
Published Online: 07 April 2004
| ||Format||Pages||Price|| |
|PDF (276K)||23||$25||  ADD TO CART|
Cite this document
Classical least squares regression is commonly used for straight line fitting in field calibration of coal analyzers. The assumption in using ordinary least squares regression that the independent variable is not subject to error, when in fact both the independent and dependent variables are subject to error, leads to calibration biases. Also, Grubbs' estimators are often used in tests of analyzer measurement precision. The inherent assumption with use of Grubbs methodology for estimating analyzer precision that the analyzer is perfectly calibrated can result in test acceptance of an analyzer that is not measuring anything. This paper proposes use of a latent variables statistical model for both calibration and precision testing. Use of the latent variables model will result in better calibrations and more reliable assessments of analyzer performance. Application is demonstrated using data from a recent analyzer test.
SGS Minerals Services, Addison, TX
Stock #: JAI12072