Integrity Assessment Report: "Optimizing flapping foil dynamics: A data-driven framework for motion optimization" (Ocean Engineering, 2025)

Academic fraud report · Geng Detector

Summary

This integrity assessment report evaluates the paper "Optimizing flapping foil dynamics: A data-driven framework for motion optimization" by Jinyu Li, Ruipeng Li, Jiaye Gong, Weicheng Cui, and Dixia Fan, published in Ocean Engineering (DOI: 10.1016/j.oceaneng.2025.122990). The overall verdict is that the paper appears clean, with no substantive indicators of academic misconduct. Four checks were performed: (1) image reuse analysis could not be conducted at the pixel level because high-resolution figure files were not available, only captions; (2) numerical sanity checks on Table 1 showed physically plausible nonlinear trade-offs between thrust coefficient Ct and efficiency eta, with no signs of fabricated digits; (3) the Gaussian Process Regression kernel configuration and PPO hyperparameters (actor learning rate 5e-5, critic 5e-3) are consistent with standard reinforcement learning practice, with no anomalous super-convergence; (4) the publication timeline (received June 3, 2025; accepted September 24, 2025) is reasonable, and cited open-source tools (LilyPad solver, Weymouth 2015) are verifiable. The paper is a computational fluid dynamics plus machine learning study, with no wet-lab imagery susceptible to manipulation. Limitations: the absence of source figures prevents deep image forensics; final determination requires institutional review.

Verdict

Cleared. No substantive evidence of academic fraud was identified. The paper is a computational fluid dynamics (CFD) and machine learning study with no wet-lab imagery. All verifiable aspects — numerical data, methodology, hyperparameters, and publication timeline — appear consistent and credible. Deep image forensics could not be performed because high-resolution figure files were not supplied to the detector.

Key findings

Image reuse (Finding 1): Not evaluable. Only figure captions were available; pixel-level, noise-pattern, and splicing analyses could not be executed for Figures 1–7 and Figures A.8–A.10.
Numerical sanity (Finding 2): Table 1 representative points (Ct = 2.88, 7.94, 18.56, 30.08; η = 0.22, 0.15, 0.098, 0.04) display a physically plausible nonlinear trade-off between thrust and efficiency, with no pathological digit uniformity suggesting fabrication.
ML/CFD methodology consistency (Finding 3): GPR kernel configuration (constant + RBF) and PPO hyperparameters (actor learning rate 5×10⁻⁵, critic 5×10⁻³), together with imitation-learning loss values around 10⁻⁵, conform to standard reinforcement-learning practice. No anomalous super-convergence or data spikes were observed.
Timeline and citation check (Finding 4): Submission received June 3, 2025; accepted September 24, 2025 — a reasonable turnaround. The cited open-source CFD solver LilyPad (Weymouth, 2015) is verifiable. No fabricated reagents, non-existent equipment, or ethics-approval conflicts were detected.

Evidence highlights

DOI: 10.1016/j.oceaneng.2025.122990
Authors: Jinyu Li, Ruipeng Li, Jiaye Gong, Weicheng Cui, Dixia Fan
Table 1 representative Ct values: 2.88, 7.94, 18.56, 30.08
Table 1 representative η values: 0.22, 0.15, 0.098, 0.04
PPO actor learning rate: 5×10⁻⁵
PPO critic learning rate: 5×10⁻³
Imitation-learning loss magnitude: ~10⁻⁵
Manuscript timeline: received 2025-06-03; accepted 2025-09-24

Notes

Scope limitation: This is a text-only analysis. Image-manipulation forensics were not feasible without the original figure files. A complete clearance would benefit from re-running the image checks once high-resolution figures and supplementary multimedia are obtained.
Suggested reader action: Authors could be contacted to release open-source training code and CFD initial-condition files to support full reproducibility, as noted in the original advisory.
Disclaimer: This report was AI-assisted and is intended for academic discussion only. Final determinations of academic misconduct rest with the responsible institutional bodies. Both false positives and false negatives remain possible.