Forensic Audit Report: Joint Optimization of Layout and Electrolyzer Capacity for Decentralized Offshore Wind-Powered Hydrogen Production (DOI: 10.1109/TII.2025.3538067)

Academic fraud report · Geng Detector

Summary

This forensic audit examined the 2025 IEEE Transactions on Industrial Informatics paper by Yunfei Du, Xinwei Shen, Boan Lu, and Xiaochi Ding on joint optimization of layout and electrolyzer capacity for offshore wind hydrogen production. The verdict is CLEAN (no evidence of academic misconduct). The paper belongs to the mathematical modeling and optimization domain, so image-reuse and image-splicing checks (Western blots, microscopy, flow cytometry) were not applicable. Instead, the audit focused on data self-consistency. Reverse calculations of the Levelized Cost of Hydrogen (LCOH) using Equation (16) confirmed exact agreement: Case 4 in Table II (EAC=43.23 M€, AHP=11436) reproduces LCOH=3.78 €/kg, and Case 1 in Table IV (EAC=58.39) reproduces 4.70 €/kg. Computational runtimes scale plausibly with problem size. Timeline checks are consistent: Matlab R2019a and Gurobi are reasonable legacy choices, and the Intel i5-12500 CPU (released 2022) predates the October 2024 submission. The unusually fast 3-month review (Oct 2024 → Jan 2025) is noted but does not constitute misconduct. Limitations: audit is based on the PDF only, with no access to raw data or solver logs.

Verdict

CLEAN — No evidence of academic fraud or data fabrication was detected. The paper's numerical results are internally self-consistent and the methodological timeline is plausible.

Key findings

Image-based fraud checks (reuse, splicing) are not applicable because the paper contains no biomedical images; it is a mathematical optimization study.
Reverse arithmetic verification of the Levelized Cost of Hydrogen (LCOH) using Equation (16) reproduced the reported values to two decimal places, effectively ruling out casual fabrication.
Algorithm runtime statistics show normal scaling behavior between the 16-platform and 72-platform test cases; no signs of artificial rounding or uniformity.
Software (Matlab R2019a, Gurobi) and hardware (Intel i5-12500, released early 2022) versions are temporally consistent with the October 2024 submission date.
Publication timeline (Received October 2024 → Accepted January 2025 → Published May 2025) is unusually fast for an IEEE Transactions journal but does not, by itself, indicate misconduct.
All cited datasets (Global Wind Atlas, NREL 5-MW turbine model) are recognized industry-standard sources.

Evidence highlights

Table II, Case 4 (Page 4165): Reported EAC = 43.23 (M€), AHP = 11436; recomputed LCOH = 43.23 × 10⁶ € / (11436 × 10³ kg) ≈ 3.78 €/kg, exactly matching the tabulated value.
Table IV, Case 1 (Page 4166): Reported EAC = 58.39; recomputed LCOH = 58.39 × 1000 / 12423 ≈ 4.70 €/kg, exactly matching the tabulated value.
Table III vs Table V: BLDDM runtime grows from 1.25 h (16 platforms) to 2.13 h (72 platforms), consistent with expected complexity scaling.
Hardware timeline: Intel i5-12500 release (2022 Q1) precedes submission (October 2024) by roughly two years — no temporal anomaly.

Notes

This audit is limited to the contents of the published PDF; raw optimization solver logs, code, and underlying wind-resource data were not independently verified.
The 3-month review-to-acceptance interval is unusually short for IEEE TII and warrants routine monitoring of the author group's future submissions, although it is not, in itself, evidence of misconduct.
All findings are based on forensic logic checks rather than institutional investigation. Any final determination of academic integrity rests with the relevant publisher or research-integrity body.