< back to main site


Classification techniques and the application to metrology.

Byng, M*; Langdell, S* (2004) Classification techniques and the application to metrology. NPL Report. CMSC 37/04

[img] Text

Download (359kB)


In this report we consider techniques for analyzing and classifying data. The report is divided into two sections: methodology and case studies. The methodology section provides a basic overview of some common statistical techniques in the fields of data reduction, cluster analysis and classification methods. The second section provides two case studies showing how these techniques can be applied to practical problems in the field of metrology.
The first case study reports on the calculation and use of a visual representation, called a dendrogram, for groupings of spectroscopy measurements of different samples of the same substance. The analysis identified those samples which had been distressed in some manner. The utility of the dendrogram in comparison to standard techniques for visualizing measurement groupings is demonstrated. We conclude that the methods used in this case study not only help the analyst better understand the data but could, with additional training data, form the basis for an automated classification procedure. This procedure could aid in grouping spectra of the same substances and identifying possibly contaminated samples.
The second case study reports on a probabilistic method for identifying out-lying values in measurement data where the outlier is due to the presence of an external factor, in this case the presence of dust particles on the sur-face being measured. The techniques used in this second case study are highly exible and allow the incorporation of prior information in a rigorous manner.

Item Type: Report/Guide (NPL Report)
NPL Report No.: CMSC 37/04
Keywords: Data fusion, cluster analysis, classification, circular dichroism, spectroscopy
Subjects: Mathematics and Scientific Computing
Mathematics and Scientific Computing > Software Support for Metrology
Last Modified: 02 Feb 2018 13:16
URI: http://eprintspublications.npl.co.uk/id/eprint/2895

Actions (login required)

View Item View Item