Research proposal · v1.6 preview · English supplement

Not an official page of Kyushu University or I²CNER. The Japanese proposal is the full document; this English page is a supplement. Proposed host: I²CNER, Kyushu University (under discussion). The research stay is not confirmed.

Proposal (English supplement)

1. Question and success criterion

When can an outlet clock offset be told apart from a real hydrogen loss in a dispenser-line mass balance? At an equal budget of unnecessary shutdowns, identify the conditions under which a leak at or above a minimum significant leak is detected within a pre-chosen deadline. The deadline and the minimum significant leak are agreed with the proposed host (if the stay goes ahead) and frozen before evaluation.

2. Method (one system, one report)

• Line-inventory mass balance over a 5 s window between inlet and outlet meters; Abel–Noble density checked against NIST (Leachman EOS); first-order meter lags; mass-balance leak monitoring as in API RP 1130 and the NRC Canada HRS mass-balance report.
• Timing faults kept separate: clock offset and drift, jitter (±0.1 s per sample), late packets handled as hold-last-value, and physical meter lag. Ordinary faults: frozen meter, frozen pressure gauge, meter-bias step.
• Calibration protocol: every threshold is set in advance on clean (no-leak) calibration fills with another seed and other stations, for an equal nominal target of 1% per fill in the worst of 7 declared clock-offset strata (±0.3 s). False-alarm rates are then measured on separate test fills. Confidence intervals: station-cluster bootstrap (400 fills per condition, 50 stations).
• Detectors: a first-order timing allowance u·|Δq| (not a bound), CUSUM (Page 1954), sequential testing (Wald 1945) and GLR with a nuisance offset (Basseville & Nikiforov 1993). EKF / profile-likelihood identifiability (Särkkä 2013; Raue et al. 2009) are extensions only.

3. Pre-registration and rejection criteria

Code and thresholds are frozen, and their hash recorded, before a blind test whose conditions would be set by the proposed host (if agreed). The deadline and the minimum significant leak are fixed before evaluation. If no detector meets the success criterion inside the declared range, that is reported as the result.

4. Plan

8 weeks (report in week 8) or 12 weeks: confirm assumptions with experts at the proposed host (if agreed) → adapt the model → freeze code and thresholds (hash recorded) → blind test → performance map, value of added measurements, failure cases → one technical report (code, data, uncertainty, failures). No connection to real equipment; no operator-confidential data.

5. Limitations

• All results are from a simplified simulation; numbers are illustrative.
• Independent pressure/temperature confirmation is not evaluated (in this model a leak does not affect pressure).
• Meter error per OIML R 139-1: 1.5% for the meter (class 2), 2% / 3% for the whole measuring system; applying 2% / 3% per meter is a stress case only.

6. Sources (verified 20–21 Oct 2026)

API RP 1130, 2nd ed. (2022) · L. Zhang et al., NRC-EME-56557 (2023), doi:10.4224/40003216 · E. S. Page, Biometrika 41 (1954) 100–115 · A. Wald, Ann. Math. Stat. 16 (1945) 117–186 · M. Basseville, I. V. Nikiforov, Detection of Abrupt Changes (1993) · S. Särkkä, Bayesian Filtering and Smoothing (2013) · A. Raue et al., Bioinformatics 25 (2009) 1923–1929 · OIML R 139-1:2018 · J. W. Leachman et al., J. Phys. Chem. Ref. Data 38 (2009) 721–748 · NIST Chemistry WebBook SRD 69 · SAE J2601. Full list in the Japanese document.