DTSA-II and EDS analysis of the first NAB (C63000) powder: SEM/EDS of Oct 27, 2026 (agent-staging 3c25afa) #27

Activity

1. claude commented on Oct 28, 2026

   

   claudeBot

   on Oct 28, 2026 – with Claude

   Last edited by claude

   More actions

   Claude finished @sgbaird's task in 50m 2s —— View job • claude/issue-27-20261028-0415 • [Create PR ➔](https://github.com/vertical-cloud-lab/caliber/compare/main...claude/issue-27-20261028-0415?quick_pull=1&title=Issue%20%2327%3A%20Changes%20from%20Claude&body=This%20PR%20addresses%20issue%20%2327%0A%0AGenerated%20with%20%5BClaude%20Code%5D(https%3A%2F%2Fclaude.ai%2Fcode)

   ---

   NAB (C63000) powder: DTSA-II and EDS analysis

   I quantified all three spectra in DTSA-II and simulated them with its Monte Carlo. At 20 kV the powder is inside the C63000 limits for Al, Fe, Ni and Mn, and no tungsten from the plate is detected. Everything is in analysis/2026-10-28-nab-c63000-eds/ on PR #28 (not merged). All 20 staged files matched their sha256 in MANIFEST.md. I didn't commit the raw spectra or SEM images; fetch_data.sh re-fetches and checks them.

   Results (wt%, normalized to 100% metals, ±1σ):

   	Al	Fe	Ni	Mn	Cu
   C63000 limits	9.0–11.0	2.0–4.0	4.0–5.5	≤1.5	bal.
   Nominal (the Monte Carlo input)	10	3	5	–	82
   Measured 20 kV, DTSA-II	9.2 ± 2.4	3.42 ± 0.11	5.33 ± 0.17	0.14 ± 0.03	81.9 ± 2.2
   Measured 15 kV, DTSA-II	8.2 ± 1.5	3.33 ± 0.08	5.10 ± 0.13	0.08 ± 0.04	83.3 ± 1.4
   Measured 10 kV, DTSA-II	6.0 ± 0.7	2.77 ± 0.13	4.43 ± 0.33	n.d.	86.8 ± 0.7
   Measured 20 kV, exspy + Monte Carlo as standard	9.7–10.1	3.55–3.81	4.91–4.97	–	81.1–81.8
   Simulated flat bulk, 20 kV, run back through DTSA-II	10.03	2.91	4.97	–	82.1
   Simulated raster over a 50 µm sphere on C, 20 / 15 / 10 kV	9.65 / 8.71 / 7.44	2.72 / 2.76 / 2.83	4.87 / 4.93 / 4.86	–	82.8 / 83.6 / 84.9

   

   Peak identification

   • No tungsten: W Mα fits to zero, so W is below about 0.2 wt% (3σ), and the clean W Lβ1 line at 9.67 keV is absent.
   • The bump next to Al K is a pile-up peak, not W or Si. It sits at 1.82 keV (two Cu L-region photons) and grows with the square of the count rate. A Cu Kα + Cu Kα pile-up peak also shows at 16.1 keV.
   • Trace and light elements: Mn is present at about 0.1 wt%. Si (below 0.1 wt%) and Zn (below 0.25 wt%) were not detected. O is present (surface oxide), and the large C peak is the tape. Escape peaks are negligible, at most 0.05 wt% on Fe.
   • Energy scale: it was about 10 eV high at Cu Kα, which I corrected per spectrum. Before that fix, a first pass read 2.3 wt% Mn.

   What the loose spheres on carbon do to accuracy

   • On a particle top it is effectively bulk. The 50 and 100 µm sphere simulations match flat bulk within a few percent.
   • An area scan over particles is not. Shadowed and tilted surfaces absorb Al K and the L lines much more than the K lines. In the simulation Al reads 0.4, 1.1 and 1.8 wt% low at 20, 15 and 10 kV, and the measured Al K / Cu Kα ratio follows that trend within 3–6% at 15–20 kV. That is the most likely reason Al drops to 6 wt% at 10 kV.
   • Analytical totals: the measured spectra give 30–39%, against 65–70% for the simulated raster and about 100% for flat bulk.
   • The L lines don't agree with the K lines (Cu L / Cu Kα comes out 1.27× the simulation, Ni L / Ni Kα 0.6–0.8×), so I didn't use them for this alloy.
   • Accuracy without standards: Newbury (1998) found 95% of standards-based analyses on flat samples within ±5% relative, against ±25–50% without standards.

   Acquisition advice (in CALIBER's parameters: beam energy, current, acquisition time)

   • Sample: use polished cross-sections for quantification, and keep loose-powder stubs for morphology only.
   • Beam energy: 20 kV, with 15 kV as a second setting. Overvoltage for Cu K is 2.2 at 20 kV but only 1.1 at 10 kV, so drop 5 and 10 kV for quantification.
   • Current: 1.3–2.0 nA at 20 kV, for 20–30% dead time. The 3.2 nA used here gave 42 kcps out (about 43% dead time) and visible pile-up. Measure the current with a Faraday cup and save the real time.
   • Live time: 100–200 s. Counting statistics are not the limit: even the existing 24 s at 20 kV gives Fe ±0.04, Ni ±0.07 and Al ±0.09 wt%. Measure at least 10 locations per lot.
   • ±1 wt% on Fe and Ni is already met.
   • ±1 wt% on Al needs measured standards (Cu, Al, Ni, Fe, Mn, plus a polished coupon of the feedstock rod) on a flat sample. Without standards, DTSA-II's own Al uncertainty is ±1.5–2.4 wt%.
   • Reference: send the rod and each lot for ICP-OES/MS; that is the (EDX − ICP-MS)² objective.
   • The full protocol is in §4 of the README.

   Limits

   • I assumed the detector area (30 mm²) and window (Moxtek AP3.3 polymer); the .spc header has neither. They affect Al and the L lines, not the K-line ratios.
   • The analysed area isn't recorded. The SEM frames are 5 kV images, and the large C peak suggests the spectra were area scans that included tape.
   • I used HyperSpy/exspy for the second tool and did not try NIST NeXL.

   Commands (all exited 0)

   ./fetch_data.sh                                  # 20/20 sha256 OK
   python eds_peaks.py /tmp/nab-staging/files/eds   # calibration, peak checks
   ./run_dtsa2.sh dtsa2_quant.py                    # DTSA-II Polaris, Xvfb + Java 26 + --enable-preview
   ./run_dtsa2.sh dtsa2_mc.py                       # 12 MC spectra, about 12 min
   ./run_dtsa2.sh dtsa2_quant.py                    # measured + MC spectra
   python exspy_crosscheck.py ...; python compare_mc.py ...
   

   · Branch · PR #28
2. sgbaird commented on Oct 28, 2026

   

   sgbaird

   on Oct 28, 2026

   MemberAuthor

   More actions

   Correction to the staged MANIFEST.md (for the record; no action needed): sem/3.2nA.tif is a 4,379x close-up (HFW 94.6 µm, 0.0308 µm/px, 5 kV, 3.2 nA, 18:05), not a 150x field. sem/3.2nA_002.tif is the 150x field (HFW 2.76 mm, 0.899 µm/px). The spectra were acquired at 18:19–18:24, after both. Per-image metadata for all 16 images of the session is in the DAL drop, for/vertical-cloud-lab/digital-alloy-lab/2026-10-28-nab-c63000-sem-psd/MANIFEST.md (agent-staging 485dd7b).