Automatic qualitative analysis for peak identification is a standard
feature of virtually all modern computer-aided analysis software for
energy dispersive X-ray spectrometry with electron excitation. Testing of
recently installed systems from four different manufacturers has revealed
the occasional occurrence of misidentification of peaks of major
constituents whose concentrations exceeded 0.1 mass fraction (10 wt%).
Test materials where peak identification failures were observed included
ZnS, KBr, FeS2, tantalum-niobium alloy, NIST Standard Reference
Material 482 (copper–gold alloy), Bi2Te3,
uranium–rhodium alloys, platinum–chromium alloy, GaAs, and
GaP. These misidentifications of major constituents were exacerbated when
the incident beam energy was 10 keV or lower, which restricted or excluded
the excitation of the high photon energy K- and L-shell X-rays where
multiple peaks, for example, Kα (K-L2,3)–Kβ
(K-M2,3); Lα (L3-M4,5)–Lβ
(L2-M4)–Lγ (L2-N4),
are well resolved and amenable to identification with high confidence.
These misidentifications are so severe as to properly qualify as blunders
that present a serious challenge to the credibility of this critical
analytical technique. Systematic testing of a peak identification system
with a suite of diverse materials can reveal the specific elements and
X-ray peaks where failures are likely to occur.
Topological Fragment Spectra (TFS) Peak Identification System for Chemical Structure Data Mining