Blunders in Automatic Qualitative Analysis (Peak Identification) for Energy Dispersive X-ray Spectrometry: The Low Voltage Microanalysis Situation

Comprehensive analytical systems for energy dispersive spectrometry (EDS) x-ray microanalysis include an automatic qualitative analysis (peak identification) procedure. The results of these automatic peak identifications are routinely accepted by some analysts, especially those novices with little experience as well as analysts who make use of the "one button spectrum-to-final report" operation that is now possible with some systems. Previous work has shown that at least some automatic peak identification systems of recent installation are subject to occasional egregious mistakes when identifying the prominent peaks of major constituents, i.e., elements present at concentrations above 0.1 mass fraction (10 weight percent) [1]. A misidentification of a major component of a material, such as a member of a binary, stoichiometric compound like KBr, constitutes such a severe mistake that it qualifies as a "blunder", described in statistics literature as an "illegitimate error" Blunders are mistakes that damage the credibility of the analyst. Such blunders occur for approximately 3% to 5% of peak identifications for major elements selected across the Periodic Table . In general, automatic peak identification appears to be quite robust for elements identified with K-shell x-rays. When blunders are encountered, they are usually associated with peak identifications that involve L-shell and M-shell photons below 4 keV. Automatic peak identification is also more robust when the procedure incorporates redundant information available when K-L or L-M x-ray combinations occur for an element, e.g., ZnK-ZnL or TaL-TaM. Many systems provide manual software tools to "ground truth" the automatic solution, but such tools often rely on a good understanding of x-ray spectrometry and may be ignored under full automation.