Data Mining Minor Phase Analysis

1
Data Mining Minor Phase Analysis
This tutorial was created from a presentation by
Dr. James Kaduk, Senior Research Associate, INEOS
Technologies. The presentation was given at an
ICDD workshop held during the 2008 International
Union of Crystallography Meeting in Osaka, Japan.
The tutorial includes three case histories of
industrial problems solved using the PDF-4
database and some creative thinking! The ICDD is
grateful to both Dr. Kaduk and INEOS Technologies
for allowing the ICDD to use their data for this
tutorial.
2
Examples of Data Mining Applications of the
Powder Diffraction File in Industrial Problem
Solving
James A. Kaduk Senior Research Associate Analytica
l Sciences Research Services INEOS
Technologies James.Kaduk_at_ineos.com
3
A Vanadium Phosphate Catalyst for the Oxidation
of Butane to Maleic Anhydride
4
From the top, raw data scan, background
subtracted scan and then the phase identification
match to phases in the Powder Diffraction
File. The reference phases are represented as
stick figures. The identification accounts for
some of the peaks, but not all peaks in the
pattern.
5
Locate peaks by interactive deconvolution, and
create GOED80.PEAK
6
GOED80.PEAK
7.2107 10 6.3038 15 5.6645 7 4.8107
12 4.4577 1 4.2699 1 4.0957 2 3.9854
4 3.8799 70 3.5823 9 3.2904 1 3.1447
100 3.0760 7 3.0487 4 3.0027 24 2.9864
28 2.6625 14
2.6141 1 2.4649 2 2.4415 15 2.3997
2 2.3665 5 2.2550 1 2.2123 1 2.0946
19 2.0780 5 1.9916 1 1.9730 1 1.9377
7 1.9026 2 1.8420 9 1.8293 2 1.7939
1 1.7503 1
1.7147 1 1.6488 3 1.6373 5 1.6257
1 1.6007 3 1.5781 10 1.5604 1 1.5232
1 1.5073 2 1.4925 1 1.4757 5 1.4611
5 1.4458 1 1.4210 3 1.3920 1 1.3840
4 1.3525 2
7
Import into SIeve
Note SIeve is the Search and Identification
program, from ICDD, that is used with PDF-4
databases. It can utilize a d,I file, such as the
one in this example, or use an experimental data
file.
8
(No Transcript)
9
Only weak peaks left. Redo Hanawalt/Fink
search, or use other capabilities
10
There is a peak at 7.2107 Å. Limit our search
to phases containing just V, P, O, and H, and
which have a strong peak 7.16 lt d lt 7.26 Å.
Note In this real example, supplementary XRF
data, taken at INEOS, limited the number of
elements in the specimen. This information can be
used in the search process in combination with
the location of a single d-spacing.
11
(No Transcript)
12
At the time of this analysis,no structure had
been reported forH4V3P3O16.5(H2O)x. Search the
ICSD for phases containing onlyV, P, O, and H.
Found ICSD entry 92847 This subsequently was
entered into the PDF as PDF 01-074-2749!
13
Boolean search on just H,V, P and O with a
strong line at 7.21Å.
14
There are still peaks at 3.5823 and 3.0760Å.
Look for phases which have strong peaks
3.55-3.61 and 3.05-3.11 Å.
15
(No Transcript)
16
Carry out a Rietveld Refinement
Note In this example, the identified phases were
sourced from the ICSD data. Using the cross
references in the PDF database, the atomic
coordinates for a Rietveld refinement can be
extracted from the ICSD database or from the
original references. Alternatively, in PDF-4
for Rietveld refinement PDF 04-008-8054 replaces
01-070-0265 PDF 04-011-5579 replaces
00-050-0380 The replacements are identified
through PDFs cross references located in both
entries.
17
Rietveld Refinements
In all the examples, Jim Kaduk follows phase
identification with Rietveld refinement for
quantitative analysis. This requires that each
entry has a set of atomic coordinates. There are
several ways to get this information
1. Directly from PDF-4 in the Structure tab,
as shown above. PDF-4 contains 114,630 data
sets with atomic coordinates.
2. From the literature reference in the
Experimental tab.
3. From the cross reference collection code found
in the comments tab, or a cross reference
structure found in the Miscellaneous tab.
18
(No Transcript)
19
(No Transcript)
20
There is still a peak at 3.985Å.
21
(No Transcript)
22
Add to the refinement
23
(No Transcript)
24
The quantitative analysis is
25
A Deactivated Pd/C Hydrogenation Catalyst
26
Automated identification finds the Pd catalyst,
but there are clearly additional phases in the
pattern. There is a characteristic unidentified
long line, shown by the arrow, at 4.05Å.
27
Pd and long line 4.05?0.02Å
The search finds a number of Pd alloys, most have
the formula XPd3.
28
The prior search suggested that an XRF analysis
would be appropriate (to find X) and Pb was found
in high concentration. PbPd3 was easily
identified (above). The source of Pb in the plant
was found and removed, solving the deactivation
problem.
29
Rust from a bag filter in a refinery unit
30
(No Transcript)
31
Locate the peaks by interactive deconvolution,
and create WINT190.PEAK
32
WINT190.PEAK
1.8431 8 1.7118 5 1.6968 12 1.6554
3 1.6300 6 1.6149 23 1.5982 1 1.5638
1 1.5244 2 1.4839 32 1.4546 6
6.2904 12 4.8517 12 4.1888 1 3.6883
7 3.4385 2 3.2950 11 3.1251 15 2.9682
30 2.8207 2 2.7040 22 2.6365 2
2.5819 3 2.5309 100 2.5196 22 2.4723
9 2.4233 7 2.3596 4 2.2371 5 2.2097
5 2.0978 20 1.9350 5 1.9118 6
33
Import into SIeve
34
Data screen for SIeve, candidate phases are
shown at the top, with matched lines in red.
Identified phases are shown at the bottom left.
Matched lines for each identified phase are shown
on the bottom right, with matches in blue, and
peaks to be matched in black. The candidate list
strongly suggests ZnS as a match for the
unmatched lines!
35
Magnetite, hematite, and lepidocrocite were easy
to identify (and expected componentsof rust),
but now we have to think
36
ZnS would be very strange here,so look far down
(gt100 hits) the list.We start seeing F-cubic
things with a 5.407Å. Do an authors cell
search for compounds withlattice parameter
around this value
37
Selected criteria for a search using Cubic
structures with a 5.40 cell edge
38
Could this be CeO2? Bulk chemical analysis
shows 2.1 wt Ce, so, yes!And this is what the
customer wanted to know!
Note The customer suspected that this was a
contaminant, but did not tell the analyst until
after the analysis was conducted!
39
All the results, including the CeO2, are
identified. SIeve also provides integrated
intensities and I/Ic values enabling a
quantitative analysis by the Reference Intensity
Ratio method.
40
Simulation of the reference pattern
identification as compared to the raw data all
peaks identified and accounted for.
41
Compare the RIR concentrations to those from a
Rietveld refinement
Note SIeve provides a scaled simulation, not a
refinement. Improved RIR results would be
expected from a refinement. Refined results are
not provided by SIeve, but are provided with
many OEM data analysis programs.
42
(No Transcript)
43
Conclusions
These examples demonstrate how different pieces
of knowledge about a sample can be combined, with
the aid of data mining, to solve complex
problems. In the three examples, the unknown was
always a minor phase with a small number of
diffraction peaks identified through a residual
peak analysis. The use of XRF data, and/or a
knowledge of the specimen history, was cross
referenced with the diffraction peaks to
greatly reduce the number of candidate materials
that fit all the known observations.
44
Thank you for viewing our tutorial. Additional
tutorials are available at the ICDD web site
(www.icdd.com).
International Centre for Diffraction Data 12
Campus Boulevard Newtown Square, PA 19073 Phone
610.325.9814 Fax [phone]