Pith. sign in

REVIEW 4 major objections 5 minor 24 references

Evolutionary computing-based image segmentation method to detect defects and features in Additive Friction Stir Deposition Process

T0 review · 4 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Attention-based segmentation with PSO-chosen thresholds identifies incomplete bonding and inhomogeneities in AFSD joints.

desk verdict Routine PSO-thresholding applied to AFSD micrographs; the paper oversells it as validated defect detection without any ground truth. read the letter →

arxiv 2507.00046 v1 pith:2V6FUDHZ submitted 2025-06-24 cs.CV cs.CE

Akshansh Mishra , Eyob Mesele Sefene , Shivraman Thapliyal This is my paper · ORCID
classification cs.CVcs.CE
keywords AdditiveFrictionStirDepositionimagesegmentationParticleSwarmOptimizationdefectdetectionself-attentionmapsmulti-channelvisualizationinterfacequality
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper proposes an image-segmentation pipeline for detecting bonding defects in Additive Friction Stir Deposition (AFSD), a solid-state additive manufacturing process where incomplete bonding and unfilled regions between layers can weaken parts. The pipeline uses Particle Swarm Optimization to choose a segmentation threshold for each optical micrograph, then builds attention-weighted and multi-channel visualizations that make interface zones visible. The authors apply the method to five AFSD samples and report that it identifies regions of incomplete bonding and inhomogeneities that conventional imaging does not make obvious. If the method works as claimed, it gives manufacturers a non-destructive, quantitative way to assess interface quality in AFSD builds.

What carries the argument

The load-bearing machinery is a two-stage image pipeline. Stage one is Particle Swarm Optimization, a population-based optimizer in which candidate particles move toward better threshold values; here it searches for the grayscale threshold that best separates deposited material from defects, using a fitness function based on edge count with a penalty for over- or under-segmentation. Stage two is an attention-weighted visualization built from gradient magnitude and distance transforms, which assigns each pixel a relevance weight based on global image context, making high-gradient interface regions stand out. A multi-channel composite then encodes boundaries, spatial relationships, and density in separate color channels. The PSO threshold is what makes the segmentation adaptive across samples, and the attention weighting is what renders subtle interface features visible.

What would settle it

Section the five analyzed AFSD samples and compare the high-attention regions from the maps against metallographic evidence of voids, porosity, or unbonded interfaces; if the flagged regions frequently contain no corresponding defect, or if known defects are missed, the central claim fails.

Watch

Extended reading notes

Core claim

The central claim is that attention-based analysis, driven by PSO-optimized thresholds, successfully identifies regions of incomplete bonding and inhomogeneities in AFSD joints. The PSO search, run with threshold bounds 50-200, a swarm of 30 particles, and up to 100 iterations, selects a threshold for each sample by minimizing the total detected edges after binary segmentation while penalizing outputs that are almost entirely black or white. The resulting thresholds fall between 156 and 173 across the five samples, which the authors read as evidence that the method adapts to each sample while staying consistent. The multi-channel visualizations combine boundary information in the red channel, spatial relationships in the green channel, and material density information in the blue channel, and the self-attention maps place bright emphasis on U-shaped interface boundaries where material transitions are sharpest. On the paper's own terms, those bright regions are the critical interface and potential defect sites.

Load-bearing premise

The load-bearing premise is that the sharp brightness gradients and dense edges detected in the optical micrographs correspond to actual material defects such as lack of bonding and unfilled regions, rather than to surface texture, polishing marks, or illumination artifacts.

Editorial extensions

If this is right

  • If the central claim holds, AFSD interface quality can be scored from optical micrographs alone, without destructively sectioning the part.
  • The narrow threshold range of 156 to 173 across different deposition conditions suggests the method can be applied to new AFSD samples with little or no retuning.
  • Because the pipeline is purely image-based, it could be integrated into a production line for near-real-time assessment of layer bonding during deposition.
  • The multi-channel visualizations give process engineers a common visual format for comparing bonding quality across samples made with different parameters.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural validation step, which the paper lists as future work, would be to correlate the attention-highlighted regions with tensile strength, fatigue resistance, or hardness maps of the same samples.
  • The same threshold-plus-attention recipe could transfer to other solid-state joining or deposition processes, such as friction stir welding, where interface bonding is the dominant quality question.
  • If the PSO search can be made fast enough, the method becomes a candidate for online monitoring, since it needs only optical micrographs and no specimen preparation beyond polishing.
  • The attention maps should be read as candidate defect indicators until ground-truth validation is done, since the paper does not yet tie specific highlighted regions to measured mechanical or microstructural defects.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes a Particle Swarm Optimization (PSO) based threshold selection method for segmenting optical micrographs of multilayer Additive Friction Stir Deposition (AFSD) builds. It combines gradient magnitude analysis, distance transforms, and composite 'self-attention' visualizations to highlight interface regions that the authors interpret as defects such as incomplete bonding and unfilled areas. Five AFSD samples from a public dataset are analyzed, and PSO-determined thresholds (156–173) are used to produce segmented images, multi-channel visualizations, and attention overlays. The abstract and conclusion claim that the method provides quantitative, non-destructive defect detection and establishes correlations between processing parameters and bond integrity.

Significance. If the central claim were validated, the method could offer a low-cost image-based screening tool for AFSD interface quality, complementing destructive testing. The use of a public dataset [24] and the idea of automating threshold selection are constructive, and the authors correctly identify a real inspection need in solid-state additive manufacturing. However, the paper provides no ground-truth validation, no comparison to baseline segmentation methods, no error or sensitivity analysis, and no quantitative link to processing parameters or mechanical properties. The claimed defect-detection capability is therefore not established. The manuscript also ships no code or reproducible algorithm details for the attention maps, further limiting its scientific contribution as submitted.

major comments (4)
  1. [Results and Discussion, Figures 1–4] The central claim that the method 'successfully identifies regions of incomplete bonding and inhomogeneities' (Abstract) is not tested against any ground truth: there are no expert-labeled defect masks, no comparisons with the Vickers hardness data available in the source dataset [24], no mechanical test correlation, and no baseline segmentation comparison. The color-coded regions in Figures 2–4 are interpreted as 'critical interface zones' purely by visual inspection, so the paper does not demonstrate that the detected edges correspond to defects rather than to ordinary image contrast at layer or substrate boundaries.
  2. [Methodology (PSO fitness function)] The PSO fitness function explicitly minimizes the sum of edges after binary thresholding, with a penalty for degenerate black/white images; the same edge map is then presented as evidence of lack of bonding and unfilled regions. Because the threshold is chosen per sample to optimize edge statistics, the subsequent visualizations are constrained to highlight exactly the high-gradient pixels the optimizer selected, making the defect interpretation circular and preventing any external confirmation from image statistics alone.
  3. [Conclusion, fourth bullet] The conclusion states that 'quantitative correlations between processing parameters (temperature, deposition speed) and bond integrity were established, yielding predictive metrics (transition sharpness, defect density),' but the Results and Discussion never report the processing parameters for the five samples, any correlation coefficients, confidence intervals, or predictive metrics. This claim is load-bearing for the process-optimization contribution and is unsupported by the presented results.
  4. [Results and Discussion, attention-based analysis / Appendix I] The attention map construction is described only by reference to 'Appendix I,' which is not present in the manuscript; no equations or algorithm details are given for the 'self-attention' or multi-channel composite, so the method is not reproducible and the term 'self-attention' is used without the standard transformer formulation.
minor comments (5)
  1. [Introduction] The phrase 'process optimization and optimization' in the first paragraph repeats 'optimization'; please correct.
  2. [Methodology] The sentence 'PSO which is a is a population-based optimization technique' contains a duplicated 'is a'; please fix.
  3. [Figure captions] The captions for Figures 1 and 2 are long and contain interpretive claims; please shorten them and move the interpretation into the body text.
  4. [References] The reference list mixes citation styles (e.g., reference [17] uses a proceedings format inconsistent with the others); please standardize all references.
  5. [Results and Discussion] Please state explicitly how the five samples differ in processing conditions and include the deposition parameters, since the paper claims these as explanatory variables.

Circularity Check

2 steps flagged · score 6.0 of 10

PSO-thresholded edges and gradient maps are relabeled as 'incomplete bonding' and 'attention' without independent ground truth, so the claimed defect detection partially reduces to the segmentation criterion itself.

  1. fitted input called prediction [Methodology, PSO fitness function paragraph; Conclusion]
    "A fitness function was defined to evaluate each threshold, which involved applying a binary threshold to the image and then measuring the effectiveness of segmentation based on edge detection. ... The PSO algorithm iteratively refined the threshold values over multiple generations ... searching for the value that minimized the sum of edges."

    The only image-content signal in the PSO fitness function is edge detection; the threshold is fitted per image to minimize the sum of edges. The same edge/gradient information is then presented as the detected defects: the Conclusion states that 'attention-based analysis successfully identifies regions of incomplete bonding and inhomogeneities in AFSD joints.' No ground-truth defect labels, mechanical data, or independent segmentation enters the loop. The output is therefore a deterministic rendering of the optimization criterion, not an independent prediction of defects; finding edges in the result is guaranteed by construction.

  2. self definitional [Results and Discussion, Figure 3 caption and MLD-M3 paragraph]
    "Yellow/bright boundaries highlight regions of maximum gradient magnitude where significant material property transitions occur ... MLD-M3 ... The U-shaped depression shows sharp yellow boundaries at the bottom curve, suggesting high attention regions where material stress concentration likely occurs."

    The paper equates 'regions of maximum gradient magnitude' with 'significant material property transitions' and then interprets the same yellow boundaries as 'intermittent bonding' or 'material stress concentration.' The defect regions are defined as the high-gradient/attention regions, and then the method is credited with successfully identifying those defects. Without any external validation that these gradient features correspond to actual incomplete bonding or unfilled regions, the claimed detection reduces to relabeling the computed gradient map as a defect map.

full rationale

There is no substantial self-citation chain here: the paper uses an external optical-micrograph dataset [24] and generic image-processing tools, so the circularity is not citation-based. The main issue is that the central claim of defect detection is not validated against any ground truth; the PSO thresholds are fitted to an edge-minimization objective, and the resulting edge/attention maps are then interpreted as material defects and inhomogeneities. This makes the claimed 'successful identification' partially circular: the outputs are constructed from the very edge/gradient information that the optimizer was asked to produce, and the physical meaning of those outputs is asserted rather than demonstrated. Separately, the paper states in the Conclusion that 'quantitative correlations between processing parameters and bond integrity were established,' but no correlation coefficients, uncertainty estimates, or comparisons with defect locations are reported; this is an unsupported claim rather than a circular one. The missing Appendix I also prevents verification of the attention equations, but absence of detail is not itself circularity. Overall, the defect-detection claim is not a fully independent empirical finding; it is, at least in part, a restatement of the optimization criterion and the gradient-map construction.

Assumptions & free parameters 2 free parameters · 4 assumptions · 1 invented entities

The paper's central claim rests on the assumption that image gradients equal defects, and on a fitness function engineered to find those gradients. The five fitted thresholds are free parameters. No independent validation or comparison to baselines is provided.

free parameters (2)
  • PSO segmentation threshold per sample = 156, 166, 167, 170, 173
    Each threshold is optimized by PSO against an edge-based fitness function on the same image, then used to produce the segmentation that is interpreted as defect detection.
  • PSO fitness function penalty weights = not specified
    The penalty for predominantly black or white images is mentioned but not quantified; it shapes the threshold selection.
assumptions (4)
  • domain assumption Gradient magnitude and edge density in optical micrographs correspond to material defects such as lack of bonding and unfilled regions.
    This is the central interpretive link invoked in the Results and Discussion when edge maps and attention overlays are labeled as defects; it is never independently validated.
  • ad hoc to paper The PSO fitness function (minimizing edge sum with a penalty for degenerate segments) selects thresholds that produce meaningful material interface segmentation.
    The fitness function is defined by the authors for this study; its adequacy for defect segmentation is assumed, not established.
  • domain assumption The images from Chou et al. [24] are representative of AFSD builds and contain visible defect signatures.
    The analysis relies entirely on this public dataset, whose images may not include validated defect labels or sufficient contrast for the method.
  • standard math Standard image processing operations (grayscale conversion, edge detection, distance transforms) faithfully represent material property transitions.
    The visualizations are built from these operations; the assumption that optical intensity gradients reflect compositional or bonding changes is unstated.
invented entities (1)
  • Self-attention map (gradient/distance composite)
    purpose: A visualization that weights image regions by gradient magnitude and distance transform, intended to highlight material transition zones.
    The paper calls these 'self-attention maps' but they are deterministic image transforms, not outputs of a trained attention model. No external handle distinguishes them from ordinary edge maps.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Evolutionary computing-based image segmentation method to detect defects and features in Additive Friction Stir Deposition Process." pith.science (2026). https://pith.science/paper/2V6FUDHZ

@misc{pith2026250700046,
  author       = {Pith},
  title        = {Pith review of: Evolutionary computing-based image segmentation method to detect defects and features in Additive Friction Stir Deposition Process},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2V6FUDHZ}},
  note         = {Machine review of arXiv:2507.00046}
}
read the original abstract

This work proposes an evolutionary computing-based image segmentation approach for analyzing soundness in Additive Friction Stir Deposition (AFSD) processes. Particle Swarm Optimization (PSO) was employed to determine optimal segmentation thresholds for detecting defects and features in multilayer AFSD builds. The methodology integrates gradient magnitude analysis with distance transforms to create novel attention-weighted visualizations that highlight critical interface regions. Five AFSD samples processed under different conditions were analyzed using multiple visualization techniques i.e. self-attention maps, and multi-channel visualization. These complementary approaches reveal subtle material transition zones and potential defect regions which were not readily observable through conventional imaging. The PSO algorithm automatically identified optimal threshold values (ranging from 156-173) for each sample, enabling precise segmentation of material interfaces. The multi-channel visualization technique effectively combines boundary information (red channel), spatial relationships (green channel), and material density data (blue channel) into cohesive representations that quantify interface quality. The results demonstrate that attention-based analysis successfully identifies regions of incomplete bonding and inhomogeneities in AFSD joints, providing quantitative metrics for process optimization and quality assessment of additively manufactured components.

Figures

Figures reproduced from arXiv: 2507.00046 by the authors.

Figure 2
Figure 2. Multi-channel visualizations of AFSD interfaces across multiple samples (a-e), highlighting material transition zones. Color mapping represents a multi-channel composite where cyan/light blue regions indicate high-gradient areas associated with mechanical property Attention-Based Analysis of AFSD Material Interfaces The attention-based analysis of AFSD samples employs a multi-stage mathematical framework that highli… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 21 canonical work pages

  1. [24]

    and Carroll, J., 2022

    Chou, K., Eff, M., Cox, C.D., Saukas, C. and Carroll, J., 2022. Optical image and Vickers hardness dataset for repair of 1080 steel using additive friction stir deposition of Aermet 100. Data in Brief, 41, p.107862

    work page 2022

  2. [1]

    and Gao, M., 2023

    Yu, Y ., Wang, C., Fu, Q., Kou, R., Huang, F., Yang, B., Yang, T. and Gao, M., 2023. Techniques and challenges of image segmentation: A review. Electronics, 12(5), p.1199

    work page 2023

  3. [2]

    and Zhang, Q., 2023

    Xiao, H., Li, L., Liu, Q., Zhu, X. and Zhang, Q., 2023. Transformers in medical image segmentation: A review. Biomedical Signal Processing and Control, 84, p.104791

    work page 2023

  4. [3]

    and Szczuko, P., 2023

    Qureshi, I., Yan, J., Abbas, Q., Shaheed, K., Riaz, A.B., Wahid, A., Khan, M.W.J. and Szczuko, P., 2023. Medical image se gmentation using deep semantic-based methods: A review of techniques, applications and emerging trends. Information Fusion, 90, pp.316-352

    work page 2023

  5. [4]

    and Jin, Y ., 2023

    Wu, J., Ji, W., Liu, Y ., Fu, H., Xu, M., Xu, Y . and Jin, Y ., 2023. Medical sam adapter: Adapting segment anything model for medical image segmentation. arXiv preprint arXiv:2304.12620

    arXiv 2023

  6. [5]

    and Wang, B., 2024

    Ma, J., Li, F. and Wang, B., 2024. U-mamba: Enhancing long -range dependency for biomedical image segmentation. arXiv preprint arXiv:2401.04722

    arXiv 2024

  7. [6]

    and Chen, D.Z., 2023, October

    Zhang, Y ., Zhou, T., Wang, S., Liang, P., Zhang, Y . and Chen, D.Z., 2023, October. Input augmentation with sam: Boosting medical image segmentation with segmentation foundation model. In International Conference on Medical Image Computing and Computer -Assisted Intervention (pp. 129-139). Cham: Springer Nature Switzerland

    work page 2023

  8. [7]

    and Liu, D., 2023

    Zhang, K. and Liu, D., 2023. Customized segment anything model for medical image segmentation. arXiv preprint arXiv:2304.13785

    arXiv 2023

Show all 24 references
  1. [8]

    and Fang, Z., 2023

    Yuan, F., Zhang, Z. and Fang, Z., 2023. An effective CNN and Transformer complementary network for medical image segmenta tion. Pattern Recognition, 136, p.109228

    2023

  2. [9]

    and Pandey, D., 2023

    Boopathi, S., Pandey, B.K. and Pandey, D., 2023. Advances in artificial intelligence for image processing: techniques, ap plications, and optimization. In Handbook of research on thrust technologies’ effect on image processing (pp. 73-95). IGI Global

    2023

  3. [10]

    and Guo, M.F., 2023

    Hong, C., Qiu, H.Y ., Gao, J.H., Lin, S. and Guo, M.F., 2023. Semantic segmentation -based intelligent thresholdfree feeder detection method for single-phase ground fault in distribution networks. IEEE Transactions on Instrumentation and Measurement

    2023

  4. [11]

    and Chen, D.Y ., 2023

    Gao, J.H., Guo, M.F., Lin, S. and Chen, D.Y ., 2023. Application of semantic segmentation in high-impedance fault diagnosis combined signal envelope and Hilbert marginal spectrum for resonant distribution networks. Expert Systems with Applications, 231, p.120631

    2023

  5. [12]

    and Laddha, S., 2023

    Babu, D.V ., Jyothi, K., Mishra, D.K., Dwivedi, A.K., Raj, E.F.I. and Laddha, S., 2023. Visual exploration of fault detection using machine learning and image processing. International Journal of Engineering Systems Modelling and Simulation, 14(1), pp.8-15

    2023

  6. [13]

    and Alkhazraji, A., 2023

    Altaie, A.S., Majeed, A.A., Abderrahim, M. and Alkhazraji, A., 2023. Fault detection on power transmission line based on wavelet transform and scalogram image analysis. Energies, 16(23), p.7914

    2023

  7. [14]

    and du Plessis, A., 2025

    Ledwaba, T., Steenkamp, C., Chmielewska-Wysocka, A., Wysocki, B. and du Plessis, A., 2025. Development of AI crack segmentation models for additive manufacturing. Tomography of Materials and Structures, p.100053

    2025

  8. [15]

    and Reisch, R.T., 2025

    Franke, J., Heinrich, F. and Reisch, R.T., 2025. Vision based process monitoring in wire arc additive manufacturing (WAAM). Journal of Intelligent Manufacturing, 36(3), pp.1711-1721

    2025

  9. [16]

    and Leung, C.L.A., 2024

    Li, W., Lambert-Garcia, R., Getley, A.C., Kim, K., Bhagavath, S., Majkut, M., Rack, A., Lee, P .D. and Leung, C.L.A., 2024. AM-SegNet for additive manufacturing in situ X-ray image segmentation and feature quantification. Virtual and Physical Prototyping, 19(1), p.e2325572

    2024

  10. [17]

    Segmentation of Additive Manufacturing Defects Using U -Net

    Wong, VWH, Ferguson, M, Law, KH, Lee, YT, & Witherell, P. "Segmentation of Additive Manufacturing Defects Using U -Net." Proceedings of the ASME 2021 International Design Engineering Technical Conferences and Computers and Information in Engineer ing Conference. V olume 2: 41s...

    2021 doi

  11. [18]

    Evaluation of Image Segmentation Methods for In Situ Quality Assessment in Additive Manufacturing

    Saini, Tushar, Panos S. Shiakolas, and Christopher McMurrough. 2024. "Evaluation of Image Segmentation Methods for In Situ Quality Assessment in Additive Manufacturing" Metrology 4, no. 4: 598-618. https://doi.org/10.3390/metrology4040037

    2024 doi

  12. [19]

    and Mishra, R.S., 2024

    Sharma, A., Jain, R., Agrawal, P ., Mukherjee, S., Gumaste, A., Davis, D.F., Haridas, R.S. and Mishra, R.S., 2024. Enhanc ed thermal stability in additive friction stir deposited ODS IN9052 Al alloy. Acta Materialia, 279, p.120284

    2024

  13. [20]

    and Mishra, R.S., 2024

    Agrawal, P., Shiau, C.H., Sharma, A., Hu, Z., Dubey, M., Lu, Y ., Shao, L., Prabhakaran, R., Wu, Y . and Mishra, R.S., 2024. Ion irradiation and examination of Additive friction stir deposited 316 stainless steel. Materials & Design, 238, p.112730

    2024

  14. [21]

    and Bhattacharjee, P.P ., 2024

    Gor, M., Barnett, M., Fabijanic, D. and Bhattacharjee, P.P ., 2024. Additive friction stir deposition of super duplex sta inless steel: Microstructure and mechanical properties. Additive Manufacturing Letters, 9, p.100204

    2024

  15. [22]

    and Paradowska, A.M., 2024

    Yakubov, V ., Ostergaard, H., Hughes, J., Yasa, E., Karpenko, M., Proust, G. and Paradowska, A.M., 2024. Evolution of material properties and residual stress with increasing number of passes in aluminium structure printed via additive friction stir dep osition. Materials, 17(1...

    2024

  16. [23]

    and Allison, P.G., 2024

    Roper, B.E.T., Roper, C.M., Rao, H.M., Brewer, L.N., Vasudevan, V .K., Zhu, N., Jordon, J.B. and Allison, P.G., 2024. Microstructure Evolution in 304L Stainless Steel Cladding Produced by Additive Friction Stir Deposition. Metallography, Microstructure, and Analysis, 13(1), pp.174-180

    2024

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.