Pith. sign in

REVIEW 5 major objections 6 minor 38 references

Predicting Mechanical Properties of Lignin-Containing Polyurethane Rigid Foams from Microstructure Using Convolutional Neural Networks

T0 review · 5 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A dual-head CNN predicts density and three mechanical properties of lignin-containing polyurethane foams from SEM images, with reported R² between 0.850 and 0.91.

desk verdict Reported R2/MAPE are almost certainly fit statistics on the full dataset, so the paper's central generalization claim is not established; the materials data and architecture are reasonable, but the evaluation must be redone. read the letter →

arxiv 2608.11447 v1 pith:ABMSTBUS submitted 2026-08-11 cs.CE cond-mat.mtrl-sci

Ilige S. Hage , Charbel Y. Seif , Jose Enrico Q. Quinsaat , Daniel J. van de Pas , Richard Vendamme , Walter Eeversd , Karolien Vanbroekhovend , Elias Feghalid This is my paper · ORCID
classification cs.CEcond-mat.mtrl-sci
keywords PolyurethanefoamligninmechanicalpropertypredictionconvolutionalneuralnetworkssustainablematerialsGrad-CAMvisualizationSEMimagingmulti-outputregression
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 argues that the load-bearing mechanical properties of lignin-containing rigid polyurethane foams can be read directly from scanning electron microscopy images, without destructive testing. A dual-head convolutional neural network is trained on 107 SEM images to predict density, specific compression modulus, specific yield strength, and specific compression strength, reporting $R^2$ values between 0.850 and 0.91, Pearson correlations above 0.92, and mean absolute percentage errors below 9%. The two-head design reflects the measured correlation structure: density is nearly independent of the three mechanical properties, while those three are strongly coupled. Grad-CAM heatmaps show the network attends to cell walls, struts, and junctions rather than pore voids, which the authors take as evidence that it learned physically meaningful morphology-property relationships. If correct, the approach offers a fast, non-destructive route to screening bio-based foam formulations.

What carries the argument

The load-bearing mechanism is a dual-head convolutional neural network. A shared backbone — three convolutional stages with 3×3 kernels, batch normalization, LeakyReLU activation, 2×2 max pooling, global average pooling, and a shared fully connected layer — produces a compact latent embedding. One fully connected head outputs density, while a second head outputs the three strongly correlated mechanical properties. The split is justified by the measured correlation structure: density correlates only weakly with the mechanical properties ($r$ between 0.29 and 0.36), whereas the three mechanical properties correlate strongly with each other ($r$ between 0.61 and 0.93), so sharing features helps the mechanical head without forcing density into an irrelevant correlation. Training uses a weighted multi-output loss, an adaptive three-phase schedule, gradient clipping, and early stopping; Gradient-weighted Class Activation Mapping (Grad-CAM), a standard technique for highlighting the image regions that drive a convolutional prediction, is then applied to each head to produce the attribution heatmaps.

What would settle it

Evaluate the trained network on SEM images of lignin-containing foam formulations whose images were held out of all training phases, and compare the resulting $R^2$ and MAPE with Table 2; if the held-out values fall well below the reported 0.850–0.91, the generalization claim is unsupported. Recomputing the metrics on the untouched 30% test subset alone, as the paper's split description allows, would provide this check directly.

Watch

Extended reading notes

Core claim

The paper's central claim is that a custom dual-head convolutional neural network can predict four properties of lignin-containing rigid polyurethane foams — density, specific compression modulus, specific yield strength, and specific compression strength, each of the latter normalized by density in MPa/(g·cm⁻³) — directly from 40× SEM images of the cellular microstructure. On its dataset of 107 images covering several lignin-based polyols and loadings, the network achieves reported $R^2$ values of 0.8505 for density, 0.897 for specific compression modulus, 0.888 for specific yield strength, and 0.914 for specific compression strength, with Pearson correlations between 0.922 and 0.956 and MAPE between 5.57% and 9.13%. The claim is not only numerical: Grad-CAM visualizations show that the density head activates around thicker struts and reduced pore volume, while the mechanical-property heads activate across interconnected junction networks and aligned pore chains, matching the load-bearing skeleton expected to govern foam stiffness and strength. The authors present this as an interpretable, image-only surrogate for mechanical testing of bio-based foams.

Load-bearing premise

The load-bearing premise is that the 107 images behave like independent samples and that the reported accuracy numbers come from images the network was not trained on, so they describe prediction of new foams rather than recall of images it had already seen.

Editorial extensions

If this is right

  • One SEM image could replace a compression test for routine screening of foam formulations, removing specimen cutting, conditioning, and destruction from the workflow.
  • Because the model preserves the ranking of materials by stiffness and strength across diverse lignin polyols, it can serve as a fast comparative tool for selecting promising formulations before full mechanical characterization.
  • The two-head design shows that weakly correlated property groups can be predicted from a single shared image representation, a pattern that transfers to other multi-output microstructure-property tasks.
  • The Grad-CAM maps give a per-image sanity check that a prediction is driven by load-bearing morphology, so the approach can be applied to new compositions whose microstructure resembles the training set.
  • The small per-image errors, mostly between 1% and 5% on the formulations shown in detail, suggest the model is precise enough to flag formulation-level differences rather than only global trends.

Reading between the lines

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

  • A testable extension would hold out entire formulations or specimens during training and then measure whether the reported $R^2$ survives; the paper's description of evaluating the entire dataset leaves open whether some of the accuracy comes from images already seen during training.
  • Because the three mechanical properties are tightly correlated, a natural baseline is a single-output CNN predicting only compression strength; comparing the dual-head model against such a baseline would reveal how much the extra outputs and shared backbone actually contribute.
  • The Grad-CAM foci suggest a simpler structural measurement — for example, strut-junction density or cell-wall thickness extracted from the same SEM images — might already predict the mechanical targets; testing that feature-based regression would clarify what the convolutional features add beyond a hand-engineered descriptor.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 6 minor

Summary. The manuscript proposes a dual-head convolutional neural network (CNN) that takes SEM images of lignin-containing rigid polyurethane foams as input and predicts four target quantities: density, specific compression modulus, specific yield strength, and specific compression strength. The authors report R2 values between 0.850 and 0.914, Pearson correlation coefficients above 0.92, and MAPE below 9.12% (Abstract; Section 4.1, Table 2). They further use Grad-CAM to argue that the network focuses on physically meaningful microstructural features such as cell walls and strut junctions, and they position the method as a non-destructive, interpretable framework for predicting foam mechanical properties (Section 4.2). The work is based on 107 SEM images with corresponding compression-test data from an earlier study.

Significance. If the reported predictive performance were established on held-out data, the paper would offer a useful non-destructive screening tool for lignin-based PU foam formulations and a reasonable demonstration of image-to-property learning for a small, heterogeneous materials dataset. The dual-head architecture motivated by target correlation is a sensible design choice, and the inclusion of Grad-CAM visualization is a positive step toward interpretability in materials machine learning. The paper also documents many training details (Section 3.4) that aid reproducibility. However, the central claim of generalization is currently unsupported because the key performance metrics appear to be computed over the entire dataset, including training samples, and because data-splitting and leakage issues are not addressed. The significance of the work therefore hinges on a re-evaluation that is not provided in the manuscript.

major comments (5)
  1. [Section 4.1, Table 2, Eqs. (31)-(33)] The reported R2, MAPE, and Pearson r are described as 'averaged for all samples' and the text states 'The entire dataset was assessed using several quantitative performance metrics.' Since Section 3.4 states the data were split into 40% training, 30% validation, and 30% testing, the metrics in Table 2 appear to include training and validation samples as well as test samples. If that is the case, the results quantify training fit, not predictive generalization, and the Abstract's claim of a network that 'predicts' material properties is not established. Please report metrics computed only on the held-out 30% test split, and additionally provide repeated random-split or cross-validated results with means and standard deviations.
  2. [Section 3.2 and Section 3.4] The data-partitioning description only says that the three subsets have matched means and standard deviations; it does not specify whether images from the same foam specimen or formulation are grouped together. If multiple SEM images come from the same foam panel and images from that panel appear in both training and test sets, the reported accuracy may be optimistically biased due to image-level leakage. The authors should describe how images relate to specimens and perform a specimen-level or formulation-level split (e.g., group by foam panel) before recomputing all metrics.
  3. [Section 3.4.1] The text states that Phase 1 'prevents degradation of pretrained representations' and refers to 'optimal utilization of pre-trained backbone features,' but nowhere does the manuscript specify what pretrained backbone is used, how it was pretrained, or on what data. The architecture in Table 1 describes layers initialized with He initialization, which suggests random initialization. If pretraining is used, the source, architecture, and pretraining dataset must be identified; if no pretraining is used, the claim should be removed. This is a load-bearing detail because the training protocol and its effect on the reported performance cannot be evaluated without it.
  4. [Section 4.2] The Grad-CAM analysis is used to conclude that 'the network focuses its predictions on physically meaningful microstructural regions' and that the model 'does not rely on superficial image statistics but rather captures true morphological determinants of mechanical behavior.' This reasoning is circular: the model's own saliency maps are used to confirm the model's physical meaningfulness without any external validation, such as comparison to known structure-property relationships on independent data, ablation of highlighted regions, or quantitative attribution metrics. The interpretability claim exceeds what Grad-CAM alone can support and should be substantially softened or supplemented with an external validation.
  5. [Section 4.1 and Discussion] No baseline models are compared, and no uncertainty quantification is provided. With only 107 images, a simple baseline (e.g., linear regression on handcrafted morphological features, a single-head CNN, or random forest on histogram features) is necessary to interpret whether the dual-head CNN's performance is actually strong. Additionally, the absence of confidence intervals or repeated-split statistics means the reader cannot assess the stability of the reported R2 and MAPE values. Please include baselines and repeated-evaluation statistics in the revised manuscript.
minor comments (6)
  1. [Section 3.1] Equation numbering is inconsistent: equations are numbered 1-10, then jump to 21 onwards; the intermediate equations (11-20) are missing. Please renumber sequentially.
  2. [Section 3.1] The text says the three mechanical parameters 'demonstrate strong independence with r=0.61-0.93'; given the context and the correlation heat map, this should read 'strong correlation' or 'strong interdependence,' not 'independence.'
  3. [Section 4.1, Eq. (31)] MAPE as defined uses the raw yi denominator. For small-magnitude targets such as density (~0.05-0.07 g/cm3), MAPE can be sensitive to small absolute errors; reporting MAE and RMSE alongside MAPE would give a more complete picture.
  4. [Table 3] For the seven formulations listed, the manuscript does not state whether these images were part of the training, validation, or test split. This information is essential for interpreting the 'predicted vs. actual' comparison in that table.
  5. [Section 2] The formulation text contains an unresolved placeholder '(ref)' in the description of the foam ingredients; please provide the intended citation.
  6. [Throughout] There are minor typographical issues, e.g., 'state-of-art' should be 'state-of-the-art,' 'diver lignin foams' in the Conclusion, and 'weaklyy' in Section 3.1. Please proofread the manuscript.

Circularity Check

1 steps flagged · score 6.0 of 10

Reported R2/MAPE/r are computed on the entire dataset, not a held-out test split, so the central prediction numbers may be fit statistics.

  1. fitted input called prediction [Section 4.1 (Eqs. 31-33, Table 2); Section 4.2.2]
    "The best-performing model corresponding to the lowest validation loss was stored after training for evaluation. The entire dataset was assessed using several quantitative performance metrics... Table 2 represents the metrics performance in terms of R2, MAPE, and Pearson correlation, for the predicted versus true values... averaged for all samples."

    If 'the entire dataset' in Section 4.1 is read literally, Table 2's R2, MAPE, and Pearson r are computed over the same 107 images used for training and validation (Section 3.4 defines a 40/30/30 train/test/val split and applies augmentation only to training batches). These quantities then measure how well the fitted model reproduces its own training labels, not how well it predicts unseen foam microstructures; calling them 'strong generalization capability' converts fit statistics into a prediction claim. Table 3 lists predictions for seven formulations but never states whether those images fell in the training split, so it cannot establish holdout generalization.

full rationale

The dataset split is described in Section 3.4: 40% training, 30% testing, and 30% validation, with augmentation applied only to training batches and early stopping based on validation loss. The evaluation section then says 'The entire dataset was assessed,' and Table 2 is labeled 'averaged for all samples.' If those words mean all images, then the R2, MAPE, and Pearson r in Table 2 are fit statistics, not held-out predictions. The abstract and Section 4.1 explicitly frame these numbers as predictive performance and generalization capability, so the central claim reduces by construction to fitting accuracy unless a separate test-only evaluation is supplied. This is the one significant circular step. No other circularity is present: the dual-head architecture is motivated by the paper's own correlation analysis rather than by an imported uniqueness theorem; the experimental data come from the authors' prior paper (Quinsaat et al. [20]), but using one's own experimentally measured data is legitimate and not circular; and the Grad-CAM analysis is an interpretability tool internal to the trained model, not an independent derivation. The finding is therefore partial circularity, not complete self-derivation.

Assumptions & free parameters 6 free parameters · 5 assumptions · 0 invented entities

The central claim rests on standard supervised learning assumptions plus several domain assumptions about image-label pairing and representativeness. The main unexamined item is that Table 2 metrics are computed on a held-out test set; the text suggests otherwise. Hyperparameters are numerous and hand-chosen, and no pretraining source is described for the claimed pretrained backbone.

free parameters (6)
  • Task loss weights w_density, w_others = 1.4, 0.4
    Chosen in Section 3.4 to balance density and mechanical heads; no sensitivity analysis is reported.
  • Initial learning rates = backbone 2e-4, heads 0.75x, warmup 1e-6
    Selected by hand in Sections 3.4 and 3.4.1; no experimental justification is provided.
  • Regularization strengths = L2 lambda=1e-5; AdamW weight decay 1e-4; dropout 30% and 20%
    Set ad hoc in Section 3.4; no grid search or ablation is reported.
  • Augmentation ranges = rotation +/-5 deg, scale +/-3%, translation +/-3 px
    Chosen to preserve microstructure in Section 3.2; no systematic study of their effect is given.
  • EMA smoothing tau = 0.9
    Used for dynamic task weights and loss smoothing in Section 3.4; no ablation is provided.
  • Early stopping patience and threshold = p=20 epochs, delta_min=1e-3
    Arbitrary thresholds in Section 3.4.2 that affect which model is selected and reported.
assumptions (5)
  • domain assumption SEM images at 40X magnification are representative of the bulk foam microstructure determining compressive properties
    Mechanical testing uses 15x10x10 mm specimens but imaging samples only a 5x5 mm surface; no multiple images per specimen or statistical representativeness check is reported (Sections 2 and 3.1).
  • domain assumption Each SEM image can be paired with mechanical ground truth from the same formulation without specimen-level leakage
    The dataset is built from prior compression testing in Quinsaat et al. [20] and new microscopy in this paper; the pairing procedure is not described in detail (Section 3.1).
  • domain assumption The 40/30/30 split separates independent foam samples
    The paper states each split has identical mean and standard deviation, but does not state whether images from the same foam block or formulation appear in multiple splits (Section 3.2).
  • ad hoc to paper Pearson correlation among targets justifies the dual-head architecture
    The architecture decision is based on label correlations (Section 3.1, Figure 2), but no ablation or comparison to a single-head network is provided.
  • ad hoc to paper A pretrained backbone exists and its features are the starting point for Phase 1
    Section 3.4.1 refers to 'pre-trained backbone features' and 'pretrained representations', but no pretraining dataset or source is described anywhere in the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Predicting Mechanical Properties of Lignin-Containing Polyurethane Rigid Foams from Microstructure Using Convolutional Neural Networks." pith.science (2026). https://pith.science/paper/ABMSTBUS

@misc{pith2026260811447,
  author       = {Pith},
  title        = {Pith review of: Predicting Mechanical Properties of Lignin-Containing Polyurethane Rigid Foams from Microstructure Using Convolutional Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ABMSTBUS}},
  note         = {Machine review of arXiv:2608.11447}
}
read the original abstract

Bio-based alternatives for conventional rigid foams have proven to be good substituents owing to their enhanced sustainability and competitive performance. However, because their manufacturing processes are complex and destructive testing is often impractical, this study investigates whether microstructural features can be correlated with mechanical properties in lignin-containing rigid polyurethane (PU) foams using machine-learning approaches. Various types and percentages of lignin-based polyols were investigated as a partial polyol replacement. Scanning electron microscopy (SEM) images and corresponding mechanical compression data were used to train a custom state-of-art dual-head convolutional neural network (CNN) targeting the specific prediction of density, specific compression modulus, specific yield strength, and specific compression strength. The CNN was optimized with a weighted multi-output loss function, achieving strong predictive performance R2 values ranging from 0.850 to 0.91 and correlation coefficients above 0.92 while maintaining mean absolute error percentages below 9%. This proves the trained network capability to predict and capture morphological features governing load bearing responses. On the other hand, Grad-CAM visualization revealed that the network focused its predictions on physically meaningful microstructural regions such as cell walls and strut junctions, which confirm that the proposed network can be classified as an interpretable, non-destructive and data-driven framework for predicting and understanding bio-based PU foams mechanical behavior, hence reducing the inconvenience caused by time consuming manufacturing and destructive testing.

Figures

Figures reproduced from arXiv: 2608.11447 by the authors.

Figure 1
Figure 1. Images of PU foams prepared using the different lignin [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. color coded heat map for the correlation matrix of the density and the mechanical [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Pairwise Scatter plots correlating parameters of density, compression modulus, [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Dual head network architecture [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: Predicted versus True values of : (top) the density in g.cm-3 , (bottom) Specific compression modulus (MPa/g.cm³), Specific yield stress (MPa/g.cm³) and Specific compression strength (MPa/g.cm-3 ) According to [PITH_FULL_IMAGE:figures/full_fig_p017_5.png]
Figure 6
Figure 6. Figure 6: Heat maps for the selected images in figure 1 ( [PITH_FULL_IMAGE:figures/full_fig_p022_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 28 canonical work pages

  1. [1]

    Polyurethane Composite Foams in High-Performance Applications: A Review,

    A. Kausar, “Polyurethane Composite Foams in High-Performance Applications: A Review,” Polym Plast Technol Eng, vol. 57, no. 4, pp. 346–369, Mar. 2018, doi: 10.1080/03602559.2017.1329433

    work page arXiv 2018

  2. [2]

    Redefining Construction: An In-Depth Review of Sustainable Polyurethane Applications,

    M. Mistry, V . Prajapati, and B. Z. Dholakiya, “Redefining Construction: An In-Depth Review of Sustainable Polyurethane Applications,” J Polym Environ, vol. 32, no. 8, pp. 3448–3489, 2024, doi: 10.1007/s10924-023-03161-w

    work page doi:10.1007/s10924-023-03161-w 2024

  3. [3]

    Production of polyols and polyurethane from biomass: a review,

    M. H. Tran and E. Y . Lee, “Production of polyols and polyurethane from biomass: a review,” Environ Chem Lett, vol. 21, no. 4, pp. 2199–2223, 2023, doi: 10.1007/s10311- 023-01592-4

    work page doi:10.1007/s10311- 2023

  4. [4]

    Lignin‐based polyurethane: recent advances and future perspectives,

    X. Ma, J. Chen, J. Zhu, and N. Yan, “Lignin‐based polyurethane: recent advances and future perspectives,” Macromol Rapid Commun, vol. 42, no. 3, p. 2000492, 2021

    work page 2021

  5. [5]

    Improved mechanical property, thermal performance, flame retardancy and fire behavior of lignin-based rigid polyurethane foam nanocomposite,

    L. Gao, G. Zheng, Y . Zhou, L. Hu, and G. Feng, “Improved mechanical property, thermal performance, flame retardancy and fire behavior of lignin-based rigid polyurethane foam nanocomposite,” J Therm Anal Calorim, vol. 120, no. 2, pp. 1311–1325, 2015, doi: 10.1007/s10973-015-4434-2

    work page doi:10.1007/s10973-015-4434-2 2015

  6. [6]

    Lignin-derived bio-based flame retardants toward high-performance sustainable polymeric materials,

    H. Yang, B. Yu, X. Xu, S. Bourbigot, H. Wang, and P. Song, “Lignin-derived bio-based flame retardants toward high-performance sustainable polymeric materials,” Green Chemistry, vol. 22, no. 7, pp. 2129–2161, 2020, doi: 10.1039/D0GC00449A

    work page doi:10.1039/d0gc00449a 2020

  7. [7]

    Cell Morphology and Mechanical Properties of Rigid Polyurethane Foam,

    Michelle Cameron Hawkins, Brendan O’Toole, and Dacia Jackovich, “Cell Morphology and Mechanical Properties of Rigid Polyurethane Foam,” Journal of Cellular Plastics, vol. 41, no. 3, pp. 267–285, May 2005, doi: 10.1177/0021955X05053525

    work page doi:10.1177/0021955x05053525 2005

  8. [8]

    Density, Microstructure, and Strain-Rate Effects on the Compressive Response of Polyurethane Foams,

    K. B. Bhagavathula, C. S. Meredith, S. Ouellet, S. S. Satapathy, D. L. Romanyk, and J. D. Hogan, “Density, Microstructure, and Strain-Rate Effects on the Compressive Response of Polyurethane Foams,” Exp Mech, vol. 62, no. 3, pp. 505–519, 2022, doi: 10.1007/s11340- 021-00772-z

    work page doi:10.1007/s11340- 2022

Show all 38 references
  1. [9]

    Machine learning‐based model for predicting the material properties of nanostructured aerogels,

    O. Aghababaei Tafreshi, Z. Saadatnia, S. Ghaffari‐Mosanenzadeh, S. Okhovatian, C. B. Park, and H. Naguib, “Machine learning‐based model for predicting the material properties of nanostructured aerogels,” SPE Polymers, vol. 4, Dec. 2022, doi: 10.1002/pls2.10082

    2022 doi

  2. [10]

    Applying machine learning for predicting thermal conductivity coefficient of polymeric aerogels,

    B. Valipour Goodarzi and A. Bahramian, “Applying machine learning for predicting thermal conductivity coefficient of polymeric aerogels,” J Therm Anal Calorim, vol. 147, pp. 1–12, Jul. 2021, doi: 10.1007/s10973-021-10960-7

    2021 doi

  3. [11]

    A developed convolutional neural network model for accurately and stably predicting effective thermal conductivity of gradient porous ceramic materials,

    P. Liu, Z. Han, W. Wu, Y . Zhao, Y . Song, and M. Chai, “A developed convolutional neural network model for accurately and stably predicting effective thermal conductivity of gradient porous ceramic materials,” Int J Heat Mass Transf, vol. 225, p. 125428, 2024, doi: https://do...

    2024

  4. [12]

    Research on multi-source microstructure image recognition of foam ceramics using convolutional network combine with frequency domain,

    Y . Yin, J. Pan, F. Wang, P. Li, Z. Cai, and X. Xu, “Research on multi-source microstructure image recognition of foam ceramics using convolutional network combine with frequency domain,” Sci Rep, vol. 15, no. 1, p. 3032, 2025, doi: 10.1038/s41598-025- 87305-z

    2025 doi

  5. [13]

    Predicting 3D particles shapes based on 2D images by using convolutional neural network,

    K. Giannis, C. Thon, G. Yang, A. Kwade, and C. Schilde, “Predicting 3D particles shapes based on 2D images by using convolutional neural network,” Powder Technol, vol. 432, p. 119122, 2024, doi: https://doi.org/10.1016/j.powtec.2023.119122

    2024

  6. [14]

    Predicting Mechanical Properties from Microstructure Images in Fiber-Reinforced Polymers Using Convolutional Neural Networks,

    Y . Sun, I. Hanhan, M. D. Sangid, and G. Lin, “Predicting Mechanical Properties from Microstructure Images in Fiber-Reinforced Polymers Using Convolutional Neural Networks,” 2024. doi: 10.3390/jcs8100387

    2024 doi

  7. [15]

    Adaptive residual convolutional neural network for compressive strength prediction of energetic materials using SEM images,

    L. Song, A. Yin, S. Chen, and Z. Long, “Adaptive residual convolutional neural network for compressive strength prediction of energetic materials using SEM images,” Phys Scr, vol. 99, no. 5, p. 56011, 2024

    2024

  8. [16]

    Data-driven prediction of the mechanical behavior of nanocrystalline graphene using a deep convolutional neural network with PCA,

    W. Shin, S. Jang, Y . Hwang, and J. Han, “Data-driven prediction of the mechanical behavior of nanocrystalline graphene using a deep convolutional neural network with PCA,” Eng Comput, vol. 41, no. 2, pp. 1085–1102, 2025, doi: 10.1007/s00366-024- 02074-z

    2025 doi

  9. [17]

    Neural network-optimized imaging for classifying lignin-based polyurethane foams: Linking molecular composition to cellular microstructure using advanced machine learning,

    I. S. Hage et al., “Neural network-optimized imaging for classifying lignin-based polyurethane foams: Linking molecular composition to cellular microstructure using advanced machine learning,” Polymer (Guildf), vol. 324, p. 128235, 2025, doi: https://doi.org/10.1016/j.polymer....

    2025

  10. [18]

    Exploring the microstructure–property relationship in polymer foams using advanced statistical methods, machine learning and deep learning: A review,

    D. Walicki, P. Zawistowski, and J. Ryszkowska, “Exploring the microstructure–property relationship in polymer foams using advanced statistical methods, machine learning and deep learning: A review,” Comput Mater Sci, vol. 256, p. 113909, 2025, doi: https://doi.org/10.1016/j.co...

    2025

  11. [19]

    Solid-State Structures and Properties of Lignin Hydrogenolysis Oil Compounds: Shedding a Unique Light on Lignin Valorization,

    O. J. Driscoll et al., “Solid-State Structures and Properties of Lignin Hydrogenolysis Oil Compounds: Shedding a Unique Light on Lignin Valorization,” 2024. doi: 10.3390/ijms251910810

    2024 doi

  12. [20]

    Preparation of Mechanically Robust Bio-Based Polyurethane Foams Using Depolymerized Native Lignin,

    J. E. Q. Quinsaat, E. Feghali, D. J. van de Pas, R. Vendamme, and K. M. Torr, “Preparation of Mechanically Robust Bio-Based Polyurethane Foams Using Depolymerized Native Lignin,” ACS Appl Polym Mater, vol. 3, no. 11, pp. 5845–5856, Nov. 2021, doi: 10.1021/acsapm.1c01081

    2021 doi

  13. [21]

    Toward bio-based epoxy thermoset polymers from depolymerized native lignins produced at the pilot scale,

    E. Feghali, D. J. van de Pas, and K. M. Torr, “Toward bio-based epoxy thermoset polymers from depolymerized native lignins produced at the pilot scale,” Biomacromolecules, vol. 21, no. 4, pp. 1548–1559, 2020

    2020

  14. [22]

    Cohen, Statistical power analysis for the behavioral sciences

    J. Cohen, Statistical power analysis for the behavioral sciences. routledge, 2013

    2013

  15. [23]

    James, D

    G. James, D. Witten, T. Hastie, and R. Tibshirani, An introduction to statistical learning: with applications in R, vol. 103. Springer, 2013

    2013

  16. [24]

    A survey on image data augmentation for deep learning,

    C. Shorten and T. M. Khoshgoftaar, “A survey on image data augmentation for deep learning,” J Big Data, vol. 6, no. 1, pp. 1–48, 2019

    2019

  17. [25]

    The effectiveness of data augmentation in image classification using deep learning,

    J. Wang and L. Perez, “The effectiveness of data augmentation in image classification using deep learning,” Convolutional Neural Networks Vis. Recognit, vol. 11, no. 2017, pp. 1–8, 2017

    2017

  18. [26]

    An overview of gradient descent optimization algorithms,

    S. Ruder, “An overview of gradient descent optimization algorithms,” arXiv preprint arXiv:1609.04747, 2016

    2016 arXiv

  19. [27]

    A method for stochastic optimization,

    K. D. P. B. J. Adam, “A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, vol. 1412, no. 6, 2014

    2014 arXiv

  20. [28]

    Decoupled weight decay regularization,

    I. Loshchilov and F. Hutter, “Decoupled weight decay regularization,” arXiv preprint arXiv:1711.05101, 2017

    2017 arXiv

  21. [29]

    Practical recommendations for gradient-based training of deep architectures,

    Y . Bengio, “Practical recommendations for gradient-based training of deep architectures,” in Neural networks: Tricks of the trade: Second edition, Springer, 2012, pp. 437–478

    2012

  22. [30]

    Sgdr: Stochastic gradient descent with warm restarts,

    I. Loshchilov and F. Hutter, “Sgdr: Stochastic gradient descent with warm restarts,” arXiv preprint arXiv:1608.03983, 2016

    2016 arXiv

  23. [31]

    Gradnorm: Gradient normalization for adaptive loss balancing in deep multitask networks,

    Z. Chen, V . Badrinarayanan, C.-Y . Lee, and A. Rabinovich, “Gradnorm: Gradient normalization for adaptive loss balancing in deep multitask networks,” in International conference on machine learning, PMLR, 2018, pp. 794–803

    2018

  24. [32]

    Pearson correlation coefficient,

    J. Benesty, J. Chen, Y . Huang, and I. Cohen, “Pearson correlation coefficient,” in Noise reduction in speech processing, Springer, 2009, pp. 1–4

    2009

  25. [33]

    Early stopping-but when?,

    L. Prechelt, “Early stopping-but when?,” in Neural Networks: Tricks of the trade, Springer, 2002, pp. 55–69

    2002

  26. [34]

    Multi-task learning using uncertainty to weigh losses for scene geometry and semantics,

    A. Kendall, Y . Gal, and R. Cipolla, “Multi-task learning using uncertainty to weigh losses for scene geometry and semantics,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7482–7491

    2018

  27. [35]

    The coefficient of determination R-squared is more informative than SMAPE, MAE, MAPE, MSE and RMSE in regression analysis evaluation,

    D. Chicco, M. J. Warrens, and G. Jurman, “The coefficient of determination R-squared is more informative than SMAPE, MAE, MAPE, MSE and RMSE in regression analysis evaluation,” PeerJ Comput Sci, vol. 7, p. e623, 2021

    2021

  28. [36]

    Grad-cam: Visual explanations from deep networks via gradient-based localization,

    R. R. Selvaraju, M. Cogswell, A. Das, R. Vedantam, D. Parikh, and D. Batra, “Grad-cam: Visual explanations from deep networks via gradient-based localization,” in Proceedings of the IEEE international conference on computer vision, 2017, pp. 618–626

    2017

  29. [37]

    A computationally efficient hybrid neural network architecture for porous media: Integrating convolutional and graph neural networks for improved property predictions,

    Q. Zhao, X. Han, R. Guo, and C. Chen, “A computationally efficient hybrid neural network architecture for porous media: Integrating convolutional and graph neural networks for improved property predictions,” Adv Water Resour, vol. 195, p. 104881, 2025

    2025

  30. [38]

    A computationally efficient hybrid neural network architecture for porous media: Integrating cnns and gnns for improved permeability prediction,

    Q. Zhao, X. Han, R. Guo, and C. Chen, “A computationally efficient hybrid neural network architecture for porous media: Integrating cnns and gnns for improved permeability prediction,” arXiv preprint arXiv:2311.06418, 2023

    2023 arXiv

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.