Integrity Review: Efficient Inverse-Designed Structural Infill for Complex Engineering Structures (Thin-Walled Structures, 2024)

Academic fraud report · Geng Detector

Summary

Verdict: Cleared (no indicators of academic fraud detected). This computational mechanics paper by Jensen et al., published in Thin-Walled Structures 195 (2024) 111427 (DOI: 10.1016/j.tws.2023.111427), presents topology-optimized infill structures for complex engineering geometries. The review examined figure reuse, numerical plausibility, computational-resource accounting, and timeline consistency. Figures (Figures 1–17) consist of 3D geometric renderings and strain-energy density fields whose captions are internally consistent across load cases (LCA, LC2, LC4) without sign of mirroring or recycling. Tabulated compliance and volume-fraction values (Tables 1, 3, 5) show realistic non-integer outputs with sensible inter-method differences (e.g., LCA compliance 49.283×10^2 vs SIMP 52.379×10^2). HPC usage logs (Tables 7–8) report granular CPU-hour figures such as 10260:52:01 on 22 nodes (704 cores), consistent with DTU Sophia billing records. Open-data deposit at DTU (DOI: 10.11583/DTU.c.6738699) and mature solver references (PETSc, Mosek) align with a 2023 publication timeline. Limitations: pixel-level image analysis was not performed; verdict rests on textual and logical cross-checking.

Verdict

Cleared. No evidence of image manipulation, fabricated numerical results, or timeline inconsistencies was identified. The paper is consistent with authentic large-scale computational research.

Key findings

Image integrity (Figures 1–17): Renderings of Michell cantilevers, Lotte Tower, and GE bracket models, together with strain-energy density fields, are described consistently across load cases (LCA, LC2, LC4). No mirrored or recycled figures detected.
Numerical plausibility (Tables 1, 3, 5): Reported compliance and volume-fraction values are non-integer and vary realistically between the proposed infill method, SIMP, and homogeneous isotropic models (e.g., LCA compliance 49.283×10^2 vs SIMP 52.379×10^2).
Computational-resource accounting (Tables 7–8): Bracket LCA SIMP case records 10260:52:01 CPU time on 22 nodes (704 cores), equivalent to ~427 days. Such granular, non-round HPC billing figures are difficult to fabricate and align with DTU Sophia cluster usage norms for 2023.
Methodological consistency: References to PETSc, Mosek, and the DTU Data deposit (DOI: 10.11583/DTU.c.6738699) are temporally coherent with a November 2023 publication date.
Authorship profile: Lead author Ole Sigmund and co-authors (Aage, Bærentzen, Olsen, Ladegaard Jensen) are established researchers in topology optimization, consistent with the paper's high-impact venue placement.

Evidence highlights

DOI: 10.1016/j.tws.2023.111427
Data DOI: 10.11583/DTU.c.6738699
Sample numeric comparison (Table 5): LCA proposed infill compliance = 49.283×10^2; SIMP = 52.379×10^2.
Sample HPC figure (Table 7): Bracket LCA SIMP total CPU time = 10260:52:01 on 22 nodes / 704 cores.
Publication date: Available online 30 November 2023.

Notes

This review is based on textual and logical analysis of the extracted PDF content; raw pixel-level image forensics were not performed.
The absence of biological-experiment red flags (Western blots, microscopy panels) means conventional image-duplication checks have limited applicability here; consistency was instead assessed via caption logic and cross-figure narrative.
All findings should be considered advisory pending any formal institutional investigation.