SAF Lab

VUV-AI automated isomer ID

Bell-method spectral matching → isomer prior → tier-α property prediction.

How it works

  1. VUV spectrum per GC×GC peak (125-240 nm absorption fingerprint)
  2. Match against NIST spectral library using Bell 2022 seven-moment statistic
  3. Threshold ≥ 5 of 7 moments pass → confident ID; ambiguous matches kept as constrained alternates
  4. Identified isomers replace uniform random draw in tier-α Monte Carlo sampling

Live: H125 (AC-Fuels ATJ) Bell-accepted IDs

Top 25 Bell-accepted IDs by R²; production pipeline output, ≥ 5/7 moments pass.

RT (min)C#ClassIdentified isomerFormulaScoreAlts
56.3C10iso-Paraffins3,3-DimethyloctaneC10H225/70.5695
58.17C?unknown3-MethylheptaneC8H185/70.5435
77.98C11Monocycloparaftert-ButylcyclohexaneC10H205/70.5225
104.44C13iso-ParaffinsDodecaneC12H265/70.5013
120.59C14iso-ParaffinsTetradecaneC14H305/70.4953
68.29C8AlkylbenzenesEthylbenzeneC8H105/70.4825
70.06C10Monocycloparaftert-ButylcyclohexaneC10H205/70.4755
36.73C?unknown5-MethylnonaneC10H225/70.4375
59.06C9DicycloparaffiTrimethylcyclohexane isomerC9H185/70.4145
73.93C?unknown1-HexeneC6H125/70.4115
72.29C12iso-ParaffinsDodecaneC12H265/70.4013
102.93C14iso-ParaffinsTridecaneC13H285/70.3513
26.22C7iso-Paraffins2,3,3-TrimethylpentaneC8H185/70.3325
87.31C12Monocycloparaft-1-Methyl-2-(4MP)cyclopentaneC12H245/70.3072
57.24C10iso-Paraffins3,3-DiethylpentaneC9H205/70.2955
165.23C16Monocycloparafalpha-BergamoteneC15H245/70.2745
138.33C16iso-ParaffinsPentadecaneC15H325/70.2633
84.87C11Monocycloparafcis-DecalinC10H185/70.2595
110.32C14iso-ParaffinsPentadecaneC15H325/70.2533
163.62C17iso-ParaffinsOctadecaneC18H385/70.2533
110.37C12Dicycloparaffit-1-Methyl-2-(4MP)cyclopentaneC12H245/70.2492
67.46C11Monocycloparaftert-ButylcyclohexaneC10H205/70.2325
146.22C16Monocycloparafalpha-CedreneC15H245/70.235
143.67C15Monocycloparafalpha-CedreneC15H245/70.2265
46.0C?unknownalpha-Copaene, EEC15H245/70.2195

Across H125: 302 Bell-accepted IDs out of 1,126 peaks (27 %). Across H125-H128 ATJ samples: ~250-300 accepted IDs each.

WJFS-11 prior sensitivity (preview of what Bell IDs unlock)

WJFS-11 is the worst density over-predictor in the cohort (+9.89 kg/m³ baseline bias). Constraining iso-paraffin sampling to just the lower-quartile NIST densities — the kind of constraint a real Bell ID would impose — closes the bias by 54 %:

Combining iso-paraffin and alkylbenzene priors closes 63 % of the bias. Aromatics-only prior gives a smaller density benefit (~10 %) because the alkylbenzene NIST pool is already well-behaved.

Coming next