material identification – the good, bad and ugly · material identification – the good, bad and...
TRANSCRIPT
Material Identification – the Good, Bad and Ugly
T.G. Fawcett, J. R. Blanton
International Centre for Diffraction Data,
Newtown Square, PA, USA 19073
Instructors
ICDD, Executive Director 2001-2017
The Dow Chemical Company1979-2001
ICDD, Manager Engineering and Design
• Decades of analyzing problems by diffraction methods
• Design databases forMaterial identification
• Develop software forMaterial identification
• Develop and create brochures,video’s and pubications
Outline
Introduction 15 min
What’s good, bad and ugly – and why ?
Good (Oyster Shell, Bauxite) 15 min
Automated Methods – How to handle good data 30 min
Bad (Portland Cement) 15 min
Break
Very Bad (Drill Core, Metal Die) 10 min
Methods for better material identification – How handle ugly data 20 min
Ugly (Allegra, Pristiq, Roman Coins) 20 min
Tools for better material identification – and how they work ? 15 min
Applic-o-meter
Theory Application
n
Week long instruction – XRD I
• Theory• Applications• Hands-on-workshops
Resources
at www.icdd.com
Tutorials and videos
Technical Bulletins
softwareapplications
Publications
Powder DiffractionAdv in X-ray Analysis
Introduction to material identification
Phase Identification is called a fingerprint technique
Match the experimentwith a reference
Unknown specimenCompared to a reference
8
X-ray Diffraction - Bragg’s Law
θθd
When this extra distance is equal to one wavelength, the x-rays scattered from the 2nd plane are in phase with x-rays scattered from the 1st plane (as are those from any successive plane). This “constructive interference” produces a
diffraction peak maximum at this angle. This is the basis of Bragg’s Law: λ= 2d·sinθ
X-rays scattered from the second plane travel the extra distance depicted by the yellow lines. Simple geometry shows this distance = 2d·sinθ
Incident coherent X-ray beam
d = interplanar spacing of parallel planes of atoms
XRPD Pattern for NaCl
(1 1 1) d = 3.256 Å2θ = 27.37°
°
λ = 1.5406 Å (Cu Kα1)a = 5.6404 Å
Bragg’s law prescribes the 2θ angular position for each peak based on the interplanar distance for the planes from which it arises. 2θ = 2 asin(λ/2d)
(2 0 0) d = 2.820 Å2θ = 31.70°
(2 2 0) d = 1.994 Å2θ = 45.45°
(3 1 1) d = 1.701 Å2θ = 53.87°
(2 2 2) d = 1.628 Å2θ = 56.47°
Each material that has a unique arrangement of atoms will produce peaks that correspond to interplanar distances between atom planes. The intensity of these peaks are proportion to the number of contributing atoms and the type of atom.
If the atoms are in a crystalline periodic array, coherent diffraction will result based on Bragg’s law.
If the atoms aren’t in a periodic array, diffuse scattering will result based on Debye’s scattering theory that considers interactions between adjacent atoms. The scattering intensity and distribution be a function of number of atoms and atom type
Why It Works
• Each crystalline component phase of an unknown specimen produces its own X-ray powder diffraction (XRPD) pattern.
• These patterns arise from the crystal structures of the component phases and are, at least in principle, unique.
• The XRPD pattern for a multi-component mixture consists of the weighted sum of the individual XRPD patterns for each component in the mixture.
CaSO4•2H2O CaO Ca(OH)2 CaCO3
D-spacings and Intensities
Simple – Control wavelength and angleand you can accurately determine d-spacingsto a part per thousand (ppm with calibration)
Complex (very powerful) – Many variables typically leads to more uncertainty in thevalues. The structure factor F is also multi-variant.
Perfect specimen preparation, great data………wrong answers
• How do you know the answers are wrong• How to get the right answers
Most powerful material identification tool
ever created
What your brain knows that your computer doesn’t
Where the sample came from
How the sample was made (or was it dug up or purchased)
The samples composition
Additional analytical data (not diffraction) that is relevant to the analysis
Is the sample a powder, gel, a solid piece, tar, goo or gunk !
What does the sample look like
Color
Crystal habit
What do the data look like
Sharp peaks, very broad features
Peaks of the right intensity in the right location
Overlapping peaks – how crowded is your data
Peak shape and symmetry
What does the background look like –what is it telling you
How difficult it was to prepare to a powder
Soft
Hard
Brittle
Waxy/deforms
All photo’s taken on a cell phone
What are these data sets telling you ?
What are these data sets telling you ?All data sets taken under routinelaboratory conditions
85,000 counts !Very sharp peaks
Orientation
• Very crowded pattern • Data scan started at 20 degrees• Intense peak at high angles
Missing data & granularity
• Weak signal 800 counts• Unusual background
Absorption (Pb) and surface roughness
Where is the baseline ?
A mix of crystalline andnon-crystalline compounds
What are these data sets telling you ?
Orientation• Regrind the sample• Apply an orientation function to the data
Missing data & granularity
Start data collection at 5 degreesRegrind the sample
A mix of crystalline andnon-crystalline compounds
Use a pattern fitting method to identify crystalline and non crystalline materialsAbsorption (Pb) and surface roughness
• Improve counting statistics• - multiple scans, longer count times• Polish surface (if possible)• Use a different radiation (reduce absorption and
increase depth penetration)
What are the data sets telling you ?
In all of the previous data sets you can identify most of the materials in the specimen using the data provided– if you identify the problem and take it’s effects into consideration
Orientation can be mathematically corrected, granularity (large grains) cannot
Orientation and granularity effect intensities, d-spacing can still be used to identify the materials
In one case the surface roughness and absorption explain the unusual background. In the other case the unusual background is being caused by the material you need to identify – they need to be treated differently to get the right results.
Know your InstrumentThese data sets were all taken on the samedesktop diffractometer with ~ 0.02 step size and~1 hour scans. Hundreds of data sets collected on this instrument produce a maximum intensity from5,000-30,000 counts depending on the materialand number of phases.
Sample A – 80,000 counts, 1hr - too highSample B – 800 counts, 1hr - too lowSample C – 35,000 counts, 2 hrs, about right
A
B
C
Too sharp ? – What is the peak width of myInstrument standard (NIST SRM - Si, LaB6)
The Good
The Click of a Mouse gets you the answer you desire
EVA, HighScore, Jade, Match, PDXL, Sieve (PDF-2/PDF-4+)
Steps in the treatment of diffraction data
From “Introduction to X-ray Powder Diffractometry “ by Jenkins & Synder (John Wiley & Sons, Inc)
Oyster Calcium
Reference PDF 04-077-4388Calcium CarbonateMineral Calcite
X-ray Powder DiffractionIdentification of oyster shell
Experiment Reference
26
Note the match to 5 significantfigures in d-spacing
Why is this good data
14,000 cts peak Intensity – easily see 1% peaks
Peaks are well resolved with good peak shape
Background is very low, easily defined, and has large regions of low intensity
0.4, 1…….. 2, and 1 Intensity
Nisinovici et al 1987-1993
42,020Minerals
Where aren’t there peaks
Identification Process
33 experimental peaks found
Compared 383,535 References
61 Have at least 2 peaks that match the experiment
1 Material, calcite, matches 30 of 33 peaks accurately !
(Note other 3 peaks match quartz, SiO2 and dolomite)
Simple “Easy” Phase ID
This data sets has no instrumental or specimen errors
(i.e. displacement, transparency or molecular orientation)
Peaks are well separated and do not require deconvolution
Large areas with no signal eliminates many possible candidates
Synthetic Bauxite
Good Signal to Noise
Good resolution not much peak overlap
Very low background, easily defined
148 Peaks are identified
Identification Process
33 experimental peaks found149 experimental peaks found
Compared 383,535 References
61 have at least 2 peaks that match the experiment529 have at least 2 peaks that match the
experiment
1 Material, calcite, matches 30 of 33 accurately !7 Materials match all 149 peaks !
(Note other 3 peaks match quartz, SiO2)
All 148 peaks accounted for !
10 X scale
Justin
How can the good go wrong !
Don’t identify all the possible peaks
Don’t have the right reference patterns
Synthetic BauxiteIUCr RR sample
Most automated systems do not find 137 peaks in automated mode !
Perfect sample – perfect data – DON’T GET THE RIGHT ANSWER !
Automated Analyses
Software A = 44 peaks all peaks ht.> 3% 4 phases Software B = 105 peaks all peaks ht.> 1% 7 phasesSoftware C = 69 peaks all peaks area > 7% 6 phasesSoftware D = 98 peaks all peaks areas > 2% 5 phases*
ICDD has found in phase identification round robin testing that the number of peaks in a pattern had to be specified, prior to testing, so that the tests were not biased by the analysts/software ability to identify peaks !
* Additional phases are top candidates
Material Identification Round Robin Testing by ICDD and IUCr
If specimens were provided the testing was biased relative to the analysts ability to prepare a finely ground random powder.
If data sets were provided, the results were biased by the analysts ability to identify peaks
If data sets were provided, results were effected by the ability of the analysts to use appropriate reference standards (both quality and coverage)
Human error dominates round robin results !
The Bad
The Bad
Is there a baseline ?
Extensive peak overlap
What’s a peak, a shoulder, just noise ?
Poor signal to noise – why with 1-2 hour scans ?
Bad and Ugly Attributes
USP Method <941> X-ray Powder Diffraction
A Bad Case –Portland Cement
Overlappingpeaks
Portland Cement
Portand Cement
8 phases, ~ 100-200 peaks
Many, many shoulders
“Automated” phase ID gets 2-5 phases
Databases make a difference as not all databases have high
quality references or coverage for all phases
Most databases do not have a cement subfile – requiring more user expertise (what makes sense)
Portland CementNo filters
CementsA B3 X5Ca3 Si O5
Mineral Filter
Major phasesC3S, C2S, Anhydrite, BrownmilleriteOffset Plot
EVA, HighScore, Jade, Match, PDXL, Sieve (PDF-2/PDF-4+)
Portlandite (4.92)
Black = references added and scaled
2nd C2S phase
Peak asymmetry and overlapExplained by C3S and a second C3S phases
9 phases !
HaturiteAllite
Portland Cement
191 Peaks identified
5 phases > 5 weight %4 phases 1-3 weight %
Portland Cement
Technique
10 hour scan* – 105,000 counts
Small step size 0.02 or less
Large sample in a cavity mount
Well ground, random powder
Method
Look for shoulders, small peaks
Analyze peak assymmetry
Use subfiles to reduce candidates (use known cement phases and chemistry)
Use summation plots, difference plots and other graphics tools
Use databases with the appropriate references !
• Difficult problems require best methodsto improve signal to noise and resolution
7 Phase Bauxite
9 Phase Portland Cement
Perfect specimen preparation, great data………wrong answers
• How do you know the answers are wrongUse plots, graphs, subfilesIs the chemistry appropriate, does the answer look rightDo your answers match the appearance of the sample
• How to get the right answersUse good technique, right methods (subfiles and filters)Check a summation plot – do all peaks and intensities matchCheck peak profiles – does your answer explain shoulders and
asymmetryUse the right database having the appropriate references
More bad cases -
Too many phases – bad peak overlap
Die - Sample courtesy of Jannaz Tavadi, ArcelorMittal, Chicago
Core Drill Sample – courtesy of Prof. Christie Rowe, McGill University & USGS
Missing Data - Collected
Identified CaF2
The low angle data confirmed pseudo wollastonite and eliminated several other possible phases
Combined simulated pattern (black) contains
• CaF2
• Pseudo-Wollastonite
• Dolomite
• Wollastonite
• Quartz
2017-0639
Now we add Cuspidene but still several unidentified peaks with significant peaks at 4.19 and 2.66
A (100) orientation on the cuspidinehelps the intensity match better than shown but doesn’t account for missing peaks.
Casting Die
Could identify seven phases
Low angle data clarified cuspidene identification and ruled out other phases
Couldn’t resolve all customer questions (low concentration phases)
Still have granularity issues
Need more data on well ground samples – may need to separate phases either physically or through thermal analysis to identify all low concentration phases – or get a high resolution data set.
Assymmetric Crystals and GranularityPharmaceutical Round Robin
Acetaminophen & Si
Mannitol & Si
Courtesy Andy Fitch, ESRF
GranularityGranularity
Scattering power falls as a function of angle (top right graph) so we shouldn’t have strongpeaks at high angles
Comparing Good to Bad
Good
33 experimental peaks found
Compared 383,535 References
61 Have at least 2 peaks that match the experiment
1 Material, calcite, matches 30 of 33 accurately !
Bad
101 experimental peaks found
~20 peaks as shoulders
Compared 383,535 References
8,491 match at least 2 peaks
705 match when strong unmatched lines considered
Best matches only match 6-8 peaks
HELP !
Experiment
Collect low angle data
Use better counting statistics
(collect longer each data point)
Regrind the sample
Compare to similar data
Separate the phases
Problem
What is the sample history
Do you know any composition
Is it a mineral, synthetic, cement, pigment etc. etc.
US Geological Survey Drill Core Samples -Palmdale, California
Courtesy of Prof. Christie RoweEarth and Planetary SciencesMcGill University, Montreal, Canada
Los AngelosAquaduct
Use Mineral classifications
Quartz, Feldspar, Zeolite, Chlorite
Areas of peak overlap
and peak symmetry
(NaCa)AlSi3O8(NaK)AlSi3O8Montmorillonite
Clinochlore
Summation Plot
7 Phase Bauxite
9 Phase Portland Cement
8 Phase Drill Core
7 Phase Metal Die
Garth
XRD Garth
The Ugly
The Ugly
Where is the baseline ? Is there a baseline ? Extensive peak overlap What’s a peak, a shoulder, just noise ? Poor signal to noise – why with 1-2 hour scans ? Amorphous materials and/or nanomaterial –
problems for background and peak finding algorithms problems
Ugly
Where is the baseline ?
The Ugly
Where is the baseline ?
Is there a baseline ?
Peak overlap is extensive
Poor signal to noise – why with 1-2 hour scans ?
Amorphous materials and/or nanomaterial give peak finding algorithms problems
Baseline – which one is right ?
Peak finding – Which one is right
Forces a fit to an incorrect answer
Did not use an ICDD database – appropriate materials not in this database
Why this data set is ugly !The data are from a mixture of crystalline (sharp peaks) and amorphous materials (broad peaks).Data analyses have difficulty finding both an appropriate background and peak positions. Users canset to algorithms to handle sharp or broad peak in most commercial software – but not both.
This requires user input to correctly identify/locate both sharp and broad peaks
Use a similarity index
that examines every point in a point by point comparison to a reference
Justin
Lots of peaksLots of peak overlap and shouldersUnstable baselineOverall weak intensity – poor signal to noisePeak overlap means that not all peaks at are theirexpected positions
The Ugly Case of Allegra
Whole tablet pattern
Fines concentrate Fenofexadine HCLSynchrotron
Separate out the componentsby particle size
Take multiple data sets
Fines concentrate Fenofexadine HCLLaboratory data
Fenofexadine HCLActive ingredient
Shell
• Anatase• Glucose
• Cellulose Iβ• Amorphous
cellulose
Anatase and Glucose
Experimental digital patters for Nanocrystalline and non-crystalline materials
Microcrystalline celluloseGelatinized StarchPovidone
Microcrystalline celluloseand Povidone
Microcrystalline cellulose
Starch
Allegra
D-MannitolD-Mannitol hydrateFexofenadine HClD-Mannitol (polymorph)
FinesFexofenadine HCL
ShellAnataseSucroseCellulose IβAmorphous CellulosePovidoneStarch
Whole TabletD-MannitolD-Mannitol HydrateD-Mannitol
Fines
Shell
Ground Tablet
10 phase Pharmaceutical Tablet- Allegra, Uses all Tools
6 Crystalline phases2 nanocrystalline phases2 amorphous phases Note the incredible number
of reference peaks
Best Methods
Technique
Case 2 - Really Ugly, Pristiq
Pharmaceuticals and Excipients Search All Phases
Pristiq – second specimen
Polymorph I
• Hydroxypropyl cellulose• rac-desvenlafaxine succinate monohydrate (I); Pristiq• Talc
Ugly Case 3 – Roman Coins
Not much signalTerrible backgroundVery broad peaks
Compare to similar data
Data collected on 24 Roman Coins
Ugly coins Good coins
14,000 counts1700 counts
Tin Oxide- Casserite
Lead Carbonate
Don’t see much Cu or Cu2Oin these bronze coin –dominant phases in the other Roman coins
Pb and Sn !
Mass attenuation coefficients
IF we have Ag, Sn or Pb – attenuation will be ~>4X that of copper when using Cu radiationbecause the x-ray absorption is effected by attenuation and density
IF we compare XRD (Cu) to XRF (Ag, Rh) data there is a 3-10 X Factor in sampling volume
If we calculate a half-depth of penetration, our data is coming from the top micron of the coinIn coins with high Ag, Sn and/or Pb. The surface features of the coin vary by ~50 microns
If we calculate a half-depth of penetration, our data is coming from the top micron of the coin
In coins with high Ag, Sn and/or Pb. The surface features of the coin vary by ~50 microns
Ugly Case 3
Instability in the background partially due to variable surfaceCorrosion products are on the surface – poorly crystallineNot much diffraction because not much depth penetration – greatlyreduced sampling volume using Cu radiation
Cu Ag Pb Sn Cl39 3.9 23.6 8.7 0.4
Si Ca P Al Fe10.4 7.3 2.0 1.7 1.0
XRF Analysis
XRD Analysis
• 37 % SiO2• 27 % CaAlFeSilicate• 20 % CaCO3• 6 % SnO2• 4 % PbCO3• 2 % Cu2O• 3 % Ag
This is a nano phase and identification is supported by XRF analysis
Best Methods
ComplimentaryTechniques
• Absorption by heavy metals limits diffraction to the top surface of the coins
• Uneven surface (coins images), small crystallite size both contribute to and uneven background
• Used XRD combined with XRF to determine phases and their relative concentrations
Justin
Data attributesGood Data
Good signal >10,000 cts
Low noise < 200 cts
Peaks are symmetrical
Areas with no peaks
All peaks are sharp (i.e. crystalline)
Bad Data Areas with overlapping
peaks
Lots of shoulders
High noise
Few or no areas without peaks
Assymmetric peak profiles
Ugly Data Where is the baseline ?
Do you have a baseline ?
Combinations of sharp and broad peaks ?
Amorphous or nanocrystalline phases in a complex matrix (many phases)
Peak profiles show asymmetry, shoulders etc
Peaks are at specified reference positions
Not all peaks at reference positionsdue to peak overlap and merging
Peaks not in proper position due tooverlap/merging or low crystallinity ornon-crystallinity – probably requireswhole pattern fitting methods
Solutions
Good Data but bad results
You don’t have the right database and quality reference materials
Did you find all the peaks –
LOOK AGAIN
With good data, a modern diffractometer and agood database – you shouldn’t have a problem !
Solutions
Bad Data
Increase your resolution reducing peak overlap (smaller slits, more monochromatic radiation, smaller steps)
Pay attention to peak asymmetry – an indicator of multiple overlapping phases
Make sure you identify all peaks including shoulders and asymmetry – this probably requires manual inspection and adjusting program parameters
Ugly Data
Try to separate the materials physically or thermally – take multiple data sets
Consider similarity indexes or other whole pattern recognition programs for non-crystalline materials.
Improve your statistics (longer count times) and resolution – do both if possible
Use subfiles or input known chemistry into your search match – eliminating inappropriate phase candidates
Take into consideration peak symmetry and breadth
Hard problems usually require complimentary analytical dataHint: If you search/match does not produce common
sense results – regrind the sample, prepare it again and runa second data set. The original data probably have a transparency,displacement, orientation or granularity problem.
Synthetic BauxiteIUCr RR sample
Easy
Gibbsite (Al(OH)3) 54.90 %
Hematite (Fe2O3) 10.00%
Boehmite (Al(OH3) 14.93%
Goethite 9.98%
Moderate
Quartz 5.16%
Kaolin 3.02%
Anatase 2.00%
Key to the analysis- How many peaks did you
find ?
If you find all 148 (or >110) – you will find all phases
If you find only the strongest peaksyou find 5 phases
Most automated systems do not find 148 peaks in default mode.
Data from threeVendors using CPD data