Pith. sign in

REVIEW 4 major objections 8 minor 61 references

Physics-Informed Residual Deep Learning for Constitutive Modeling of Hot Deformation and Dynamic Recrystallization in a Mo-Rich $\alpha+\beta$ Titanium Alloy

T0 review · 4 major / 8 minor · reviewed 2026-08-02 · deepseek-v4-flash

Pith's one-line read R² of 0.985: physics-aware AI reproduces titanium hot-deformation flow stress

desk verdict Honest PINN extension for a titanium alloy; stress interpolation is solid, but the headline generalization claim rests on a point-level split and the DRX head is really a JMAK prior. read the letter →

arxiv 2607.13467 v1 pith:2L2JOHW5 submitted 2026-07-15 cond-mat.mtrl-sci

classification cond-mat.mtrl-sci
keywords physics-informedneuralnetworksconstitutivemodelinghotdeformationdynamicrecrystallizationtitaniumalloyflowstressJMAKkineticsZener-Hollomonparameter
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

The paper tries to establish that a neural network trained on hot-compression data for a molybdenum-rich α+β titanium alloy can reproduce the full flow-stress behavior—hardening, peak, dynamic recovery, and recrystallization softening—across 800–1050 °C and 0.01–10 s⁻¹, provided it is steered by physics-informed penalty terms computed with automatic differentiation. The authors claim that their residual-network architecture, with derivative-sign constraints plus DRX kinetic priors, reaches R² ≈ 0.985 (RMSE ≈ 11.7 MPa) on held-out test points, and that the same framework reproduces the Zener–Hollomon scaling of stress and plausible sigmoidal recrystallized-fraction evolution. A sympathetic reader would care because classical Arrhenius and empirical constitutive models cannot describe the peak–softening transition that dominates finite-element simulations of forging, rolling, and extrusion of titanium alloys. If correct, the approach would provide a physically consistent, data-driven constitutive surrogate that also reports its own predictive uncertainty.

What carries the argument

The Stacked Residual Physics-Informed Neural Network (STAR-PINN) is the carrier: a residual encoder with layer-normalized SiLU blocks whose outputs are differentiated with respect to inputs to evaluate physics penalties, with adaptive weighting λ(t). The DRX-Aware variant adds a second output head predicting the recrystallized fraction X_DRX, concatenates X_DRX into the latent features before stress decoding, and regularizes it with monotonicity, saturation, coupling (DRX increases ⇒ stress decreases), Arrhenius-consistency, and JMAK kinetic priors (k = 2.0, n = 2.0). Automatic differentiation is the mechanism that converts these physical trends into differentiable training signals, and Mont

What would settle it

Retrain the DRX-Aware STAR-PINN on 24 of the 30 deformation conditions and evaluate the remaining 6 held-out conditions as disjoint (T, strain-rate) cells: if the held-out RMSE rises well above the in-sample ~12 MPa while training-set RMSE stays low, the interpolation-versus-forecast gap is exposed. A complementary check is to measure recrystallized area fraction by EBSD at a few deformed specimens and compare it to the predicted X_DRX; a substantial mismatch would show the DRX head is duplicating the JMAK prior rather than learning real microstructure.

Watch

Extended reading notes

Core claim

The central claim is that stacking residual blocks and encoding hot-deformation physics as soft penalties—thermal softening (∂σ/∂T < 0), strain-rate sensitivity (∂σ/∂ln ε̇ > 0), pre-peak hardening (∂σ/∂ε > 0), post-peak softening (∂σ/∂ε < 0), plus curvature regularization, and, in the DRX-Aware version, JMAK–Avrami, DRX monotonicity, saturation, and DRX–softening coupling terms—is enough to make a neural network approximate the experimentally measured flow curves of Ti–6Al–4Mo–1V–0.1Si across 30 thermomechanical conditions with an RMSE of 11.69 MPa and R² of 0.9850. The DRX-Aware model further claims to reproduce the expected sigmoidal evolution of recrystallized fraction starting near εc ≈

Load-bearing premise

The load-bearing premise is that a random point-level 80/20 split measures the model's predictive value; if the test set were composed of entirely unseen temperature and strain-rate conditions, the claimed accuracy across the processing window could be substantially lower.

Editorial extensions

If this is right

  • If the model generalizes as claimed, flow stress at any (ε, T, strain-rate) combination inside the 800–1050 °C and 0.01–10 s⁻¹ window can be obtained without fitting an analytical constitutive equation.
  • The embedded derivative-sign constraints suppress unphysical predictions outside training points, which matters for robust finite-element process models that must maintain thermodynamic consistency.
  • MC-Dropout confidence intervals give per-condition uncertainty, allowing high-risk process windows to be flagged for additional experimental validation.
  • The DRX-Aware head yields a continuous recrystallized-fraction field that can be queried as a qualitative microstructure indicator at any strain or deformation condition.
  • Zener–Hollomon behavior emerges without being explicitly imposed, indicating that the physics penalties are compatible with thermally activated deformation theory and can recover established metallurgical scaling.

Reading between the lines

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

  • A decisive test the paper does not perform is holding out entire deformation conditions rather than random points: the 80/20 point-level split lets training and test points come from the same experimental flow curves, so the R² ≈ 0.985 should be read as interpolation accuracy, and generalization to new temperature/strain-rate cells remains unproven.
  • The near-identical accuracy of the two models (R² 0.9850 vs 0.9847) suggests the DRX head, at present, does not improve stress prediction; its value lies in the interpretability of X_DRX, not in raw predictive accuracy.
  • Because X_DRX is trained only on macroscopic stress data with a strong JMAK prior (k = 2, n = 2), the predicted recrystallization curves probably reflect that prior more than material-specific microstructure; EBSD or optical metallography on a few deformed specimens would be needed to give the DRX outputs quantitative meaning.
  • A natural extension the paper hints at is adding grain-size prediction as a third output head and using EBSD-derived recrystallized fractions as soft training targets—a testable upgrade that would turn the qualitative DRX predictions into calibrated microstructural forecasts.
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 / 8 minor

Summary. The paper presents two physics-informed neural networks, the Improved/Enhanced STAR-PINN and the DRX-Aware STAR-PINN, trained on isothermal hot-compression flow-stress data for a Mo-rich α+β titanium alloy (30 conditions spanning 800–1050 °C and 0.01–10 s⁻¹). The models impose soft constraints via automatic differentiation for thermal softening, strain-rate sensitivity, pre-peak hardening, post-peak softening, and DRX-related kinetics, with the DRX-Aware model adding a JMAK-based DRX-fraction branch. The authors report test-set RMSE ≈ 11.7–11.8 MPa, R² ≈ 0.985, and five-fold CV RMSE ≈ 12.8 MPa, and claim that the models generalize across a wide thermomechanical processing range and reproduce DRX kinetics, Zener–Hollomon behavior, and physically consistent constitutive response.

Significance. If the reported generalization were established under a condition-level evaluation, the paper would be a useful contribution to ML-based constitutive modeling: it provides a detailed architecture, explicit hyperparameters, MC-dropout uncertainty quantification, and a candid discussion of the qualitative nature of the DRX fraction in §4.8. However, the central generalization claim is not currently supported because the 80/20 split and, as described, the 5-fold CV operate on individual strain–temperature–strain-rate points rather than on independent deformation conditions. In addition, the DRX-fraction output is essentially a hand-parameterized JMAK prior with no microstructural validation, so the abstract's claim to have 'accurately reproduced ... DRX kinetics' is overstated. The paper also contains internal tension between the claimed post-critical-strain softening constraint and the plotted rising flow curves at 1000 °C/0.01 s⁻¹. These issues are addressable, but they require reframing or additional evaluation.

major comments (4)
  1. [§2.2.6, Table 3; §4.3, Table 4; Conclusion] The train/test split is described as '80% / 20% (random, seed = 42)' and the 5-fold CV as '5-fold' without condition-level grouping. Because each of the 30 thermomechanical conditions is a continuous flow curve, a random point-level split places held-out test points adjacent to training points from the same experimental curve. The reported RMSE/R² therefore certify in-curve interpolation, not prediction of unseen deformation conditions. The conclusion that the model 'generalizes successfully throughout a wide thermomechanical processing range' and the corresponding abstract claim are not supported by this split design. Please re-evaluate using a condition-blocked split (e.g., leave out whole deformation conditions) and report condition-level errors. If the CV folds were already grouped by condition, this must be stated explicitly; the current text is ambiguous at best.
  2. [§4.8, Fig. 8; Eq. (1.4); Abstract] The DRX-Aware model's X_DRX output is shown to nearly coincide with the JMAK reference curve with hand-set k = 2.0, n = 2.0. The text in §4.8 states that the JMAK regularization is 'effectively enforcing strain-rate-independent DRX kinetics' and that the predicted curves 'nearly overlap' the JMAK target. The monotonicity loss, saturation loss, and Avrami prior essentially determine the sigmoidal shape and saturation. Since no microstructural measurements (EBSD, optical fraction, etc.) were used as training or validation targets, the abstract's claim that the model 'accurately reproduced ... DRX kinetics' is unsupported; the model predicts its own prior. Please either add quantitative microstructural validation or explicitly reframe the X_DRX output as a qualitative, prior-regularized latent variable, consistent with the caveat already stated in §4.8.
  3. [§2.2.1, §4.4, Fig. 3] The physics constraints in §2.2.1 are introduced as enforcing ∂σ/∂ε < 0 after the critical strain εc = 0.0876, and §4.4 claims the penalty terms 'restrict ∂σ/∂ε to be ≤0 once εc is reached.' Yet Fig. 3(c) shows a continuously rising flow stress at 1000 °C/0.01 s⁻¹ past εc, and the text describes this positive slope as appropriate. Because all constraints are soft, the network may violate them when the data loss dominates. The manuscript should quantify actual constraint violations (e.g., the fraction of test points with ∂σ/∂ε > 0 for ε > εc) and reconcile the use of a single global εc = 0.6εp with the condition-dependent peak behavior admitted in §4.4. Without this, the claim of 'physically consistent constitutive behavior' is not verifiable.
  4. [§4.6, Table 5] The five-fold cross-validation summary is reported without specifying the fold construction. If the folds are random point-level partitions, the CV RMSE ≈ 12.8 MPa is consistent with the same interpolation concern as the main split and does not validate condition-level generalization. The paper should state explicitly whether folds were grouped by deformation condition; if not, condition-blocked CV should replace the current procedure. This point is load-bearing because the CV is presented as evidence that 'both models are well-generalized.'
minor comments (8)
  1. [Abstract, §4.3, Table 4] The abstract reports MAE = 4.83 MPa for the DRX-Aware STAR-PINN, but Table 4 lists MAE = 4.89 MPa for that model. Please correct the inconsistency.
  2. [General] The model is called 'Improved STAR-PINN' in §2.2.2 and Table 4, but 'Enhanced STAR-PINN' in §4.10 and the Conclusion. Use one consistent name.
  3. [§3–§4] Section numbering and order are corrupted: §4.6 appears twice, a section numbered 4.6 follows §4.8, and a section numbered 4.10 follows it. Figure 3's caption and a block of descriptive text are duplicated in §4.5. Please renumber and remove duplicate material.
  4. [Tables] Table 2 is used both for the DRX loss terms and later referenced as the CV summary; the CV summary should be Table 5. Check all cross-references to table numbers.
  5. [§2.2.3] The data loss is a heteroscedastic Gaussian NLL, but the reported metrics are point-level RMSE/MAE/R². Clarify how the predictive mean μ̂ relates to the NLL parameterization and how the variance output is used in the reported metrics.
  6. [General] No comparison is made to a plain residual network without physics losses. Such a baseline would directly support the claim that the physics-informed terms add value beyond the residual architecture.
  7. [References] References [51] and [52] have garbled author-order/formatting, and 'Zener–Hollomon' is occasionally misspelled 'Zener–Holloman.'
  8. [§4.4] The sentence stating that narrow CI bands at 1000–1050 °C 'verify that the models are not merely interpolating' is directly contradicted by the point-level split design and should be removed or replaced with a statement about the model's behavior within the training-condition domain.

Circularity Check

3 steps flagged · score 6.0 of 10

DRX-fraction 'prediction' is the JMAK prior by construction; the Zener-Hollomon trend is imposed by the physics losses; and the point-level 80/20 split makes the headline RMSE/R2 certify in-curve interpolation rather than prediction of unseen conditions.

  1. self definitional [Section 2.2.5 (Eq. 1.4, Table 2); Section 4.8, Fig. 8]
    "XJMAK = 1 − exp[−k·((ε−εc)/(εp−εc))n] (1.4) with k = 2.0, n = 2.0. ... This happens as the JMAK regularization loss acts as the main limitation on the DRX trajectory, effectively enforcing strain-rate-independent DRX kinetics. ... The nearly precise alignment between the neural network predictions and the JMAK target across all temperatures ... validates the effectiveness of the Avrami regularization as a physics prior."

    The DRX head is penalized with LAvrami weight 1.0, Lsat 0.5, and LDRX 1.0 to follow the hand-set JMAK sigmoid with k=2, n=2 and to saturate to 1. Section 4.8 also concedes there are no microstructural training targets ('no direct microstructural measurements ... were available'). The near-overlap between the predicted XDRX and the JMAK curve in Fig. 8 is the regularizer doing its job, not an independent prediction. Therefore the abstract's claim of accurately reproduced 'DRX kinetics' reduces to a consistency check of the prior.

  2. self definitional [Section 2.2.1/Tables 1-2; Section 4.10, Fig. 11]
    "Both frameworks show a monotonic increase in flow stress with ln Z, indicating that the constitutive response aligns with the thermally activated characteristics of hot deformation. ... The capacity of both models to replicate this Arrhenius-type constitutive behavior suggests that the integrated physics constraints effectively directed the learning process toward thermodynamically valid solutions, even without an explicitly defined constitutive equation during training."

    The training loss explicitly enforces ∂σ/∂T < 0 and ∂σ/∂ln(ε̇) > 0, and Model B adds LArr with purpose 'Kinetic consistency' / 'Arrhenius consistency.' Since ln Z = ln(ε̇) + Q/(RT), these two constraints mathematically imply ∂σ/∂ln Z > 0. The monotone σ-lnZ plot in Fig. 11 is therefore imposed by the physics losses, not discovered by the network. The claim that the model replicates Arrhenius behavior 'even without an explicitly defined constitutive equation' omits the explicit Arrhenius-consistency penalty.

1 more flagged steps
  1. fitted input called prediction [Section 2.2.6/Table 3; Section 4.3/Table 4; Conclusion, first bullet]
    "Train / Test split 80 % / 20 % (random, seed = 42) ... Consequently, 30 independent thermomechanical deformation conditions were investigated. ... These findings verify that the physics-constrained, residual-stacked architecture generalizes successfully throughout a wide thermomechanical processing range."

    The random 80/20 split is applied to individual recorded points, not to the 30 independent (T, ε̇) conditions. Held-out test points therefore come from the same experimental flow curves, at the same temperature and strain rate and at adjacent strain values, as training points; the network can interpolate along each measured curve. The reported test RMSE = 11.69 MPa and R² = 0.9850 thus certify in-curve interpolation, and the five-fold CV is likewise point-level ('solely by interpolation accuracy'). The headline generalization claim is statistically forced by the training set rather than demonstrated on unseen deformation conditions.

full rationale

The stress-fitting results themselves are not circular: the network is genuinely trained on the experimental flow-stress data and achieves low error on held-out points. However, two of the paper's advertised 'predictions' are enforced by construction rather than independently confirmed. First, the DRX fraction output is regularized to the hand-set JMAK curve and to saturation/monotonicity; Fig. 8's near-perfect overlap with the JMAK reference is the penalty functioning, and the paper itself disclaims quantitative microstructural validation. Second, the Zener-Hollomon monotonic trend is a mathematical consequence of the imposed thermal-softening and strain-rate-sensitivity constraints plus the explicit Arrhenius-consistency penalty, so presenting it as an emergent physical validation is circular. Third, the generalization claim rests on a point-level data split rather than a condition-level split, so the RMSE/R² metrics certify interpolation along the 30 measured flow curves rather than prediction of new thermomechanical conditions. Self-citations to the authors' prior alloy-characterization work [6-8,22] are used as experimental provenance, not as load-bearing theoretical assumptions, so no separate self-citation circularity is found. Overall, the central numerical stress fit has independent content, but the DRX-kinetics and Zener-Hollomon 'validations' are definitional, and the generalization claim is overstated. Score 6 reflects this partial, construction-level circularity.

Assumptions & free parameters 5 free parameters · 6 assumptions · 1 invented entities

The central stress-fit claim rests on standard ML ingredients and abundant data, but the advertised DRX kinetics rest on a hand-picked JMAK prior, a global critical-strain rule, and an unvalidated latent fraction. The physics constraints are enforced as soft penalties whose weights are tuned manually, so the ledger is dominated by parametric assumptions rather than independent experimental evidence.

free parameters (5)
  • JMAK k = 2.0
    Hand-picked 'universal estimate' in Eq. 1.4; controls the DRX prior that shapes all X_DRX predictions.
  • JMAK n = 2.0
    Hand-picked 'universal estimate' in Eq. 1.4; together with k fixes the sigmoidal kinetics imposed on the DRX head.
  • critical strain ratio εc/εp = 0.6
    Eq. 1.3 sets a global DRX-onset strain. The paper itself admits at 950 °C a single εc misplaces the peak, and at 1000–1050 °C low strain rates show no peak at all.
  • physics loss weights = λ(t): 1e-5→1e-2; L_DRX=1.0, L_Avrami=1.0, L_couple=1.0, L_sat=0.5, L_Arr=0.5
    Hand-tuned balance in Eq. 1.2 and Table 2; determines how strongly the DRX and Arrhenius priors bind the advertised 'predictions'.
  • dropout rate τ = 0.05
    Used for MC-dropout uncertainty; no calibration or coverage analysis is provided, and the resulting 95% CI is much narrower than the test RMSE.
assumptions (6)
  • domain assumption Zener–Hollomon and Arrhenius thermal-activation kinetics describe hot-deformation flow stress
    Section 1 invokes Z = ε̇ exp(Q/RT); used to claim Z–H consistency, though no Arrhenius equation is trained.
  • ad hoc to paper JMAK/Avrami sigmoidal X_DRX with universal k=2, n=2 applies across all 30 conditions
    Eq. 1.4; the paper itself notes this is a zeroth-order estimate and that strain-rate dependence of DRX is largely suppressed by the prior.
  • ad hoc to paper A single global critical strain εc = 0.6εp marks the onset of DRX and of enforced flow softening
    Eq. 1.3; the paper concedes at 950 °C the actual peak strain is larger than the global εc, so the constraint mislocates the softening onset.
  • domain assumption Monotone sign constraints (∂σ/∂T < 0, ∂σ/∂lnε̇ > 0, hardening then softening) hold at every point in the thermomechanical window
    Table 1 and Eq. 1.2; at 1000 °C/0.01 s⁻¹ the reported curves rise monotonically without a clear peak, so the post-εc softening constraint cannot literally bind.
  • domain assumption The recorded ~7,100 points from 30 compression tests are accurate and representative of the alloy's constitutive response
    Section 2.1; no friction-correction details, no repeat-specimen information, and the comparison to EBSD/optical data is deferred.
  • standard math Neural-network universal approximation and automatic differentiation compute correct gradients for the chosen architecture
    Section 2.2; standard but unproved background assumption for the specific width/depth/optimizer configuration.
invented entities (1)
  • X_DRX (dynamic recrystallization fraction output)
    purpose: Auxiliary network output used to condition the stress decoder and to claim prediction of recrystallization kinetics.
    No EBSD, optical metallography, or X-ray measurements are used as training targets or validation reference points; Section 4.8 explicitly states the outputs are qualitative and require future microstructural validation. Its behavior is regularization-dominated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Physics-Informed Residual Deep Learning for Constitutive Modeling of Hot Deformation and Dynamic Recrystallization in a Mo-Rich $\alpha+\beta$ Titanium Alloy." pith.science (2026). https://pith.science/paper/2L2JOHW5

@misc{pith2026260713467,
  author       = {Pith},
  title        = {Pith review of: Physics-Informed Residual Deep Learning for Constitutive Modeling of Hot Deformation and Dynamic Recrystallization in a Mo-Rich $\alpha+\beta$ Titanium Alloy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2L2JOHW5}},
  note         = {Machine review of arXiv:2607.13467}
}
abstract

Accurate constitutive modeling of hot deformation behavior is essential for designing thermomechanical processes in advanced structural alloys. Conventional Arrhenius-type and empirical models do not adequately capture the combined effects of strain hardening, dynamic recovery (DRV), and dynamic recrystallization (DRX) across broad processing conditions. In this study, two Stacked Residual Physics-Informed Neural Networks (STAR-PINNs) were developed to simulate the hot deformation response of a Mo-rich $\alpha+\beta$ titanium alloy (Ti-6Al-4Mo-1V-0.1Si). The Enhanced STAR-PINN incorporated thermomechanical constitutive constraints, while the DRX-Aware STAR-PINN employed a dual-output architecture to account for recrystallization kinetics. Both models used a shared residual encoder trained on experimental flow stress data collected at temperatures from 800 to 1050 degrees C and strain rates between 0.01 and 10 per second. Physics-informed constraints, including thermal softening, strain-rate sensitivity, strain hardening, and post-peak softening, were enforced through automatic differentiation. The DRX-Aware model further integrated JMAK-Avrami regularization, DRX saturation constraints, and Arrhenius-based consistency with tunable parameters, directly linking the predicted DRX fraction to stress output via latent-feature fusion. The DRX-Aware STAR-PINN achieved RMSE = 11.69 MPa, MAE = 4.83 MPa, R^2 = 0.9850, and a cross-validated RMSE of 12.47 +/- 0.26 MPa. This model accurately reproduced temperature-dependent flow curves, DRX kinetics, and Zener-Hollomon relationships, while maintaining physically consistent constitutive behavior. These results demonstrate that physics-informed deep learning provides a robust and interpretable framework for constitutive modeling, offering a practical approach for advanced process modeling of titanium alloys.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

61 extracted references · 23 canonical work pages

  1. [61]

    Dodge, S

    S. Dodge, S. Barmada, A. Formisano, A STacked Adaptive Residual PINN (STAR-PINN) Approach to 2D Time-Domain Magnetic Diffusion in Nonlinear Materials, IEEE Access 13 (2025) 141380–141394. https://doi.org/10.1109/ACCESS.2025.3597869

  2. [1]

    Banerjee, J.C

    D. Banerjee, J.C. Williams, Perspectives on Titanium Science and Technology, Acta Mater. 61 (2013) 844–879. https://doi.org/10.1016/j.actamat.2012.10.043

  3. [2]

    Boyer, An overview on the use of titanium in the aerospace industry, Mater

    R.R. Boyer, An overview on the use of titanium in the aerospace industry, Mater. Sci. Eng. A 213 (1996) 103–114. https://doi.org/10.1016/0921-5093(96)10233-1

  4. [3]

    https://doi.org/10.1007/978-3-540-73036-1

    Titanium, Springer Berlin Heidelberg, Berlin, Heidelberg, 2007. https://doi.org/10.1007/978-3-540-73036-1

  5. [4]

    Peters, J

    M. Peters, J. Kumpfert, C.H. Ward, C. Leyens, Titanium Alloys for Aerospace Applications, Adv. Eng. Mater. 5 (2003) 419 –427. https://doi.org/10.1002/adem.200310095

  6. [5]

    Leyens, M

    C. Leyens, M. Peters, eds., Titanium and Titanium Alloys: Fundamentals and Applications, 1st ed., Wiley, 2003. https://doi.org/10.1002/3527602119

  7. [6]

    Mahadule, R.K

    D. Mahadule, R.K. Khatirkar, A. Gupta, R. Kumar, Effect of heating temperature and cooling rate on the microstructure and mechanical properties of a Mo-rich two phase α + β titanium alloy, J. Mater. Res. 36 (2021) 751 –763. https://doi.org/10.1557/s43578-020- 00100-6

  8. [7]

    Joshi, D

    P.S. Joshi, D. Mahadule, R.K. Khatirkar, Effect of cooling rate and temperature on variant selection in a Mo -rich two -phase α + β titanium alloy, Can. Metall. Q. (2026) 1 –14. https://doi.org/10.1080/00084433.2025.2611656

Show all 61 references
  1. [8]

    Mahadule, P.S

    D. Mahadule, P.S. Joshi, R.K. Khatirkar, Evolution of Microstructure and Texture During Hot Deformation of the Ti -6Al-4Mo-1 V -0.1Si Alloy, Met. Mater. Int. (2025). https://doi.org/10.1007/s12540-025-01974-z

  2. [9]

    Sellars, W.J

    C.M. Sellars, W.J. McTegart, On the mechanism of hot deformation, Acta Metall. 14 (1966) 1136–1138. https://doi.org/10.1016/0001-6160(66)90207-0

  3. [10]

    Zener, J.H

    C. Zener, J.H. Hollomon, Effect of Strain Rate Upon Plastic Flow of Steel, J. Appl. Phys. 15 (1944) 22–32. https://doi.org/10.1063/1.1707363

  4. [11]

    Garofalo, An empirical relation defining the stress dependence of minimum creep rate in metals, (1963) 351

    F. Garofalo, An empirical relation defining the stress dependence of minimum creep rate in metals, (1963) 351

  5. [12]

    Y . Liu, H. Zhang, L. Hua, F. Huang, K. Ji, Y . Chen, J. Mao, Deformation behavior and microstructural evolution of Ti-6Al-4 V alloy under compression with confining pressure, Int. J. Plast. 198 (2026) 104610. https://doi.org/10.1016/j.ijplas.2026.104610

  6. [13]

    Jonas, C.M

    J.J. Jonas, C.M. Sellars, W.J.McG. Tegart, Strength and structure under hot -working conditions, Metall. Rev. 14 (1969) 1–24. https://doi.org/10.1179/mtlr.1969.14.1.1. 30

  7. [14]

    Sakai, J.J

    T. Sakai, J.J. Jonas, Overview no. 35 Dynamic recrystallization: Mechanical and microstructural considerations, Acta Metall. 32 (1984) 189 –209. https://doi.org/10.1016/0001-6160(84)90049-X

  8. [15]

    Cook, W.H

    J. Cook, W.H. G.R., A constitutive model and data for metals subjected to large strains, high strain rates and high temperatures, 21 (1983) 541--547

  9. [16]

    Poliak, J.J

    E.I. Poliak, J.J. Jonas, A one-parameter approach to determining the critical conditions for the initiation of dynamic recrystallization, Acta Mater. 44 (1996) 127 –136. https://doi.org/10.1016/1359-6454(95)00146-7

  10. [17]

    Seshacharyulu, S.C

    T. Seshacharyulu, S.C. Medeiros, W.G. Frazier, Y .V .R.K. Prasad, Microstructural mechanisms during hot working of commercial grade Ti -6Al-4V with lamellar starting structure, Mater. Sci. Eng. A 325 (2002) 112 –125. https://doi.org/10.1016/S0921 - 5093(01)01448-4

  11. [18]

    J. Cai, F. Li, T. Liu, B. Chen, M. He, Constitutive equations for elevated temperature flow stress of Ti–6Al–4V alloy considering the effect of strain, Mater. Des. 32 (2011) 1144 –

  12. [19]

    G. Quan, G. Luo, J. Liang, D. Wu, A. Mao, Q. Liu, Modelling for the dynamic recrystallization evolution of Ti–6Al–4V alloy in two-phase temperature range and a wide strain rate range, Comput. Mater. Sci. 97 (2015) 136 –147. https://doi.org/10.1016/j.commatsci.2014.10.009

  13. [20]

    Lin, X.-M

    Y .C. Lin, X.-M. Chen, A critical review of experimental results and constitutive descriptions for metals and alloys in hot working, Mater. Des. 32 (2011) 1733 –1759. https://doi.org/10.1016/j.matdes.2010.11.048

  14. [21]

    Safari, M

    A. Safari, M. Imran, S. Weiss, A Comparative Study on Modified Johnson –Cook and Arrhenius-Type Constitutive Models to Predict the Hot Deformation Behaviour of Molybdenum-Hafnium-Carbide Alloy, J. Mater. Eng. Perform. 30 (2021) 1945 –1956. https://doi.org/10.1007/s11665-021-05464-2

  15. [22]

    Joshi, D

    P.S. Joshi, D. Mahadule, R.K. Khatirkar, A machine learning and microstructural synergy for flow stress prediction in hot -deformed Ti–6Al–4Mo–1V–0.1Si alloy, Mater. Chem. Phys. 350 (2026) 131943. https://doi.org/10.1016/j.matchemphys.2025.131943

  16. [23]

    Ghaboussi, J.H

    J. Ghaboussi, J.H. Garrett, X. Wu, Knowledge‐Based Modeling of Material Behavior with Neural Networks, J. Eng. Mech. 117 (1991) 132 –153. https://doi.org/10.1061/(ASCE)0733-9399(1991)117:1(132)

  17. [24]

    Huber, Ch

    N. Huber, Ch. Tsakmakis, Determination of constitutive properties fromspherical indentation data using neural networks. Part i:the case of pure kinematic hardening in plasticity laws, J. Mech. Phys. Solids 47 (1999) 1569 –1588. https://doi.org/10.1016/S0022-5096(98)00109-4

  18. [25]

    Sun, W.D

    Y . Sun, W.D. Zeng, Y .Q. Zhao, Y .L. Qi, X. Ma, Y .F. Han, Development of constitutive relationship model of Ti600 alloy using artificial neural network, Comput. Mater. Sci. 48 (2010) 686–691. https://doi.org/10.1016/j.commatsci.2010.03.007

  19. [26]

    G. Ji, F. Li, Q. Li, H. Li, Z. Li, A comparative study on Arrhenius-type constitutive model and artificial neural network model to predict high-temperature deformation behaviour in Aermet100 steel, Mater. Sci. Eng. A 528 (2011) 4774 –4782. https://doi.org/10.1016/j.msea.2011.0...

  20. [27]

    Peng, K.L

    P.W. Peng, K.L. Ou, C.Y . Chao, Y .N. Pan, C.H. Wang, Research of microstructure and mechanical behavior on duplex (α + β) Ti-4.8Al-2.5Mo-1.4V alloy, J. Alloys Compd. 490 (2010) 661–666. https://doi.org/10.1016/j.jallcom.2009.10.133

  21. [28]

    Sabokpa, A

    O. Sabokpa, A. Zarei -Hanzaki, H.R. Abedi, N. Haghdadi, Artificial neural network modeling to predict the high temperature flow behavior of an AZ81 magnesium alloy, Mater. Des. 39 (2012) 390–396. https://doi.org/10.1016/j.matdes.2012.03.002

  22. [29]

    Fan, H.C

    J.K. Fan, H.C. Kou, M.J. Lai, B. Tang, H. Chang, J.S. Li, Characterization of hot deformation behavior of a new near beta titanium alloy: Ti -7333, Mater. Des. 49 (2013) 945–952. https://doi.org/10.1016/j.matdes.2013.02.044

  23. [30]

    Logarzo, G

    H.J. Logarzo, G. Capuano, J.J. Rimoli, Smart constitutive laws: Inelastic homogenization through machine learning, Comput. Methods Appl. Mech. Eng. 373 (2021) 113482. https://doi.org/10.1016/j.cma.2020.113482

  24. [31]

    Pandya, C.C

    K.S. Pandya, C.C. Roth, D. Mohr, Strain rate and temperature dependent fracture of aluminum alloy 7075: Experiments and neural network modeling, Int. J. Plast. 135 (2020) 102788. https://doi.org/10.1016/j.ijplas.2020.102788

  25. [32]

    Mozaffar, R

    M. Mozaffar, R. Bostanabad, W. Chen, K. Ehmann, J. Cao, M.A. Bessa, Deep learning predicts path -dependent plasticity, Proc. Natl. Acad. Sci. 116 (2019) 26414 –26420. https://doi.org/10.1073/pnas.1911815116

  26. [33]

    Gorji, M

    M.B. Gorji, M. Mozaffar, J.N. Heidenreich, J. Cao, D. Mohr, On the potential of recurrent neural networks for modeling path dependent plasticity, J. Mech. Phys. Solids 143 (2020) 103972. https://doi.org/10.1016/j.jmps.2020.103972

  27. [34]

    Raissi, P

    M. Raissi, P. Perdikaris, G.E. Karniadakis, Physics -informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations, J. Comput. Phys. 378 (2019) 686 –707. https://doi.org/10.1016/j.jcp.2018.10.045

  28. [36]

    Cuomo, V .S

    S. Cuomo, V .S. di Cola, F. Giampaolo, G. Rozza, M. Raissi, F. Piccialli, Scientific Machine Learning through Physics-Informed Neural Networks: Where we are and What’s next, (2022). https://doi.org/10.48550/ARXIV .2201.05624

  29. [37]

    Haghighat, M

    E. Haghighat, M. Raissi, A. Moure, H. Gomez, R. Juanes, A physics -informed deep learning framework for inversion and surrogate modeling in solid mechanics, Comput. Methods Appl. Mech. Eng. 379 (2021) 113741. https://doi.org/10.1016/j.cma.2021.113741

  30. [38]

    Mao, A.D

    Z. Mao, A.D. Jagtap, G.E. Karniadakis, Physics-informed neural networks for high-speed flows, Comput. Methods Appl. Mech. Eng. 360 (2020) 112789. https://doi.org/10.1016/j.cma.2019.112789

  31. [39]

    Zhang, M

    E. Zhang, M. Dao, G.E. Karniadakis, S. Suresh, Analyses of internal structures and defects in materials using physics -informed neural networks, Sci. Adv. 8 (2022) eabk0644. https://doi.org/10.1126/sciadv.abk0644

  32. [40]

    Jagtap, K

    A.D. Jagtap, K. Kawaguchi, G.E. Karniadakis, Adaptive activation functions accelerate convergence in deep and physics-informed neural networks, J. Comput. Phys. 404 (2020) 109136. https://doi.org/10.1016/j.jcp.2019.109136. 32

  33. [41]

    Karniadakis, I.G

    G.E. Karniadakis, I.G. Kevrekidis, L. Lu, P. Perdikaris, S. Wang, L. Yang, Physics - informed machine learning, Nat. Rev. Phys. 3 (2021) 422 –440. https://doi.org/10.1038/s42254-021-00314-5

  34. [42]

    K. He, X. Zhang, S. Ren, J. Sun, Deep Residual Learning for Image Recognition, in: 2016 IEEE Conf. Comput. Vis. Pattern Recognit. CVPR, IEEE, Las Vegas, NV , USA, 2016: pp. 770–778. https://doi.org/10.1109/CVPR.2016.90

  35. [43]

    Eshkofti, M

    K. Eshkofti, M. Barreau, Vanishing Stacked-Residual PINN for State Reconstruction of Hyperbolic Systems, (2025). https://doi.org/10.48550/ARXIV .2503.14222

  36. [44]

    LeCun, Y

    Y . LeCun, Y . Bengio, G. Hinton, Deep learning, Nature 521 (2015) 436 –444. https://doi.org/10.1038/nature14539

  37. [45]

    Heaton, Ian Goodfellow, Yoshua Bengio, and Aaron Courville: Deep learning: The MIT Press, 2016, 800 pp, ISBN: 0262035618, Genet

    J. Heaton, Ian Goodfellow, Yoshua Bengio, and Aaron Courville: Deep learning: The MIT Press, 2016, 800 pp, ISBN: 0262035618, Genet. Program. Evolvable Mach. 19 (2018) 305–307. https://doi.org/10.1007/s10710-017-9314-z

  38. [46]

    J.L. Ba, J.R. Kiros, G.E. Hinton, Layer Normalization, (2016). https://doi.org/10.48550/ARXIV .1607.06450

  39. [47]

    Ramachandran, B

    P. Ramachandran, B. Zoph, Q.V . Le, Searching for Activation Functions, (2017). https://doi.org/10.48550/ARXIV .1710.05941

  40. [48]

    Loshchilov, F

    I. Loshchilov, F. Hutter, Decoupled Weight Decay Regularization, (2017). https://doi.org/10.48550/ARXIV .1711.05101

  41. [49]

    Loshchilov, F

    I. Loshchilov, F. Hutter, SGDR: Stochastic Gradient Descent with Warm Restarts, (2016). https://doi.org/10.48550/ARXIV .1608.03983

  42. [50]

    Lakshminarayanan, A

    B. Lakshminarayanan, A. Pritzel, C. Blundell, Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles, (2016). https://doi.org/10.48550/ARXIV .1612.01474

  43. [51]

    Ghahramani, Zoubin Yarin, Dropout as a Bayesian approximation: representing model uncertainty in deep learning, (2016) 1050–1059

    G. Ghahramani, Zoubin Yarin, Dropout as a Bayesian approximation: representing model uncertainty in deep learning, (2016) 1050–1059

  44. [52]

    Salakhutdinov, Ruslan Nitish and Hinton, Geoffrey and Krizhevsky, Alex and Sutskever, Ilya, Dropout: a simple way to prevent neural networks from overfitting, (2014) 1929–1958

    S. Salakhutdinov, Ruslan Nitish and Hinton, Geoffrey and Krizhevsky, Alex and Sutskever, Ilya, Dropout: a simple way to prevent neural networks from overfitting, (2014) 1929–1958

  45. [53]

    Avrami, Kinetics of Phase Change

    M. Avrami, Kinetics of Phase Change. I General Theory, J. Chem. Phys. 7 (1939) 1103–

  46. [54]

    Reaction Kinetics in Processes of Nucleation and Growth

    K. Barmak, A Commentary on: “Reaction Kinetics in Processes of Nucleation and Growth”*, Metall. Mater. Trans. B 49 (2018) 3616–3680. https://doi.org/10.1007/s11663- 010-9421-1

  47. [56]

    Sellars C M, The Physical Metallurgy of Hot Working, Sheffield, England, 1979

  48. [57]

    Sakai, A

    T. Sakai, A. Belyakov, R. Kaibyshev, H. Miura, J.J. Jonas, Dynamic and post -dynamic recrystallization under hot, cold and severe plastic deformation conditions, Prog. Mater. Sci. 60 (2014) 130–207. https://doi.org/10.1016/j.pmatsci.2013.09.002

  49. [58]

    Hallberg, Approaches to Modeling of Recrystallization, Metals 1 (2011) 16 –48

    H. Hallberg, Approaches to Modeling of Recrystallization, Metals 1 (2011) 16 –48. https://doi.org/10.3390/met1010016

  50. [59]

    Ding, Z.X

    R. Ding, Z.X. Guo, Coupled quantitative simulation of microstructural evolution and plastic flow during dynamic recrystallization, Acta Mater. 49 (2001) 3163 –3175. https://doi.org/10.1016/S1359-6454(01)00233-6. 33

  51. [60]

    Caruana, Multitask Learning, Mach

    R. Caruana, Multitask Learning, Mach. Learn. 28 (1997) 41 –75. https://doi.org/10.1023/A:1007379606734

  52. [1112]

    https://doi.org/10.1063/1.1750380

  53. [1151]

    https://doi.org/10.1016/j.matdes.2010.11.004

Pith tools

Reviewed August 2, 2026 · model on record in the stance chip above.