{"id":"1a5c2236-de10-4f1e-947a-3392c19044a1","arxiv_id":"2606.23402","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":3.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Physics-informed CNNs (PICNNs with soft PDE penalty and PInteCNNs with hard-coded solver) predict wood thermal responses from multimodal data, claiming better accuracy and interpretability than data-driven models on three wood datasets.","lead":"The paper introduces physics-informed convolutional neural networks that embed a 2D steady-state heat transfer equation to predict pixel-level thermal responses of heterogeneous wood from RGB images and temperature maps. A smart generalist might read it to see how adding physical constraints to machine learning can improve predictions for complex real-world materials in architecture.","discovery_kind":"new_application","skeptic_critique":{"model":"grok-4.3","headline":"Adequacy of the normalized 2D steady-state heat equation as model for wood samples under experimental conditions","rationale":"The reader's weakest assumption is precisely the load-bearing modeling choice. Because the full manuscript was not supplied in the initial review, the concern remains untested; confirming or refuting the PDE adequacy via the residual check would directly resolve the UNVERDICTED status without requiring changes to other aspects of the argument.","tokens_in":1743,"tokens_out":350,"duration_ms":15499,"concrete_test":"On the held-out test images, compute the pointwise residual of the normalized 2D steady-state operator applied to the model-predicted temperature field; compare the distribution of residuals against the same operator applied to a data-driven baseline and against measurement noise level. If the physics-informed residuals are not systematically smaller or if they correlate with prediction error, the embedded equation is not an adequate model.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that embedding the derived PDE (normalized 2D steady-state heat transfer) supplies a useful inductive bias that improves accuracy, interpretability, and parameter extraction over pure data-driven baselines. This holds only if the PDE is a sufficiently accurate description of the actual thermal process in the Poplar, Grandis-CC, and Grandis-RC samples. The abstract states the equation is obtained by normalization from the general heat equation, but provides no detail on whether transient effects, moisture-dependent conductivity, anisotropy, or out-of-plane heat flow remain negligible under the testbed conditions. If any of these are material, the soft penalty or hard-coded solver will enforce an incorrect constraint, potentially degrading rather than enhancing performance.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript proposes two physics-informed frameworks, PICNNs (soft penalty on the PDE) and PInteCNNs (hard-coded analytical solver), that embed a normalized 2D steady-state heat transfer equation derived from the general heat equation to predict pixel-level thermal responses of spatially heterogeneous wood from RGB images and temperature maps. It introduces three real-world multimodal datasets (Poplar, Grandis-CC, Grandis-RC) and claims that the physics-informed approaches outperform purely data-driven baselines in accuracy, interpretability, and extraction of physical parameters while capturing intra-species diversity.","tokens_in":1900,"tokens_out":379,"duration_ms":17078,"significance":"If the quantitative results demonstrate clear gains over baselines with proper validation and if the chosen PDE is shown to be an adequate model, the work would illustrate a practical way to inject thermodynamic constraints into image-to-field prediction tasks, offering a template for interpretable modeling of heterogeneous materials.","major_comments":[{"comment":"Abstract: the central claims of outperformance and interpretable parameter extraction are asserted without any quantitative metrics, error bars, baseline comparisons, or validation protocol details, rendering it impossible to evaluate whether the data support the claims.","section":"Abstract"},{"comment":"PDE model (derivation and experimental justification): the normalized 2D steady-state heat transfer equation is adopted as the inductive bias, yet the manuscript provides no analysis or evidence showing that transient effects, moisture-dependent conductivity, anisotropy, or out-of-plane heat flow remain negligible for the Poplar, Grandis-CC, and Grandis-RC samples under the reported testbed conditions. If any of these are material, the soft or hard enforcement of the PDE would impose an incorrect constraint.","section":"PDE derivation and model section"}],"minor_comments":[],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive feedback. We address the major comments point by point below.","responses":[{"response":"We agree that the abstract should include quantitative support. In the revised manuscript we will add specific metrics (e.g., mean squared error reductions with standard deviations across the three datasets), explicit baseline comparisons, and a concise description of the validation protocol.","revision_made":"yes","referee_comment":"[Abstract] Abstract: the central claims of outperformance and interpretable parameter extraction are asserted without any quantitative metrics, error bars, baseline comparisons, or validation protocol details, rendering it impossible to evaluate whether the data support the claims."},{"response":"We acknowledge the need to justify the modeling assumptions. The testbed protocol used equilibration periods and controlled moisture to approximate steady-state conditions; we will add a dedicated subsection with literature references on wood thermal behavior and any sensitivity checks feasible from the existing datasets to support that the neglected effects remain small.","revision_made":"yes","referee_comment":"[PDE derivation and model section] PDE model (derivation and experimental justification): the normalized 2D steady-state heat transfer equation is adopted as the inductive bias, yet the manuscript provides no analysis or evidence showing that transient effects, moisture-dependent conductivity, anisotropy, or out-of-plane heat flow remain negligible for the Poplar, Grandis-CC, and Grandis-RC samples under the reported testbed conditions. If any of these are material, the soft or hard enforcement of the PDE would impose an incorrect constraint."}],"tokens_in":1422,"tokens_out":343,"duration_ms":18690,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The core of the paper is taking existing physics-informed CNN ideas and pointing them at RGB-to-thermal mapping for three wood species. They collect real multimodal data for Poplar, Grandis cross-cut and radial-cut samples, then compare a soft-penalty version against one that hard-codes an analytical solver for the normalized 2D steady-state heat equation. Collecting the datasets is the concrete piece of work here; domain-specific data for heterogeneous natural materials is still scarce.\n\nThe abstract states that the physics embedding improves accuracy, interpretability and parameter extraction over pure data-driven baselines, yet it gives no error values, no baseline numbers, no ablation on the penalty weight, and no check on whether the extracted parameters match independent measurements. Without those, it is impossible to know whether the claimed balance actually occurs.\n\nThe stress-test point about the PDE itself is on target. Wood is anisotropic, moisture changes conductivity, and the testbed may have transient or out-of-plane effects. If the normalized steady-state equation is only an approximation, the hard-coded solver will enforce the wrong constraint and the soft-penalty version will trade off against it. The abstract does not report any validation that these effects are negligible, so the inductive-bias argument stays untested.\n\nThis is for people who need thermal models of wood in architecture or materials engineering and are already comfortable with PINNs. A reader looking for a new method will not find one; a reader looking for usable datasets and a worked example on a narrow material class might. The work is coherent on its own terms and cites the relevant PINN literature, so it clears the bar for serious refereeing even though the current write-up is too thin on evidence.","headline":"This applies standard PINN techniques to wood thermal prediction with three new datasets, but the abstract supplies zero numbers so the outperformance claim cannot be checked.","tokens_in":2378,"tokens_out":417,"would_cite":false,"duration_ms":20065,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Physics-informed convolutional networks that embed the steady-state heat equation outperform data-driven models on predicting pixel-level thermal responses of heterogeneous wood from RGB images.","keywords":["physics-informed neural networks","wood thermal properties","heat transfer equation","convolutional neural networks","material heterogeneity","thermal prediction","deep learning"],"falsifier":"A held-out set of wood samples where a standard convolutional neural network produces lower prediction error on thermal maps than either the penalty-based or hard-coded physics models, or where extracted parameters fail to correlate with separate lab measurements of thermal conductivity.","tokens_in":2661,"feed_emoji":"🌡️","tokens_out":590,"duration_ms":19596,"temperature":0.7,"pith_summary":"The paper develops two physics-informed neural network architectures that incorporate a normalized 2D steady-state heat transfer equation to predict how wood materials of different species respond to temperature changes at the pixel level. Traditional data-driven methods treat the mapping from visual images to thermal maps as a black box and can overfit to noise, but these approaches enforce thermodynamic consistency either as a loss penalty or by building a solver into the network. By testing on real datasets of poplar and eucalyptus wood samples, the work shows that adding physical constraints improves both prediction quality and the ability to extract meaningful parameters like thermal conductivity variations within a species. This matters because wood's natural variability challenges assumptions of uniform material properties in applications like building design and energy efficiency.","feed_headline":"Physics constraints boost neural net accuracy on wood heat maps","feed_subtitle":"Models enforcing the steady-state heat equation handle wood variations better than black-box predictors and yield physical parameters.","key_machinery":"PICNNs that embed physics as a soft penalty term in the loss function and PInteCNNs that hard-code an analytical approximator-predictor-corrector solver for the heat equation directly into the CNN.","core_discovery":"Embedding physical inductive biases successfully balances predictive accuracy, physical interpretability, and intra-species diversity, outperforming data-driven approaches in handling complex wood material heterogeneity and enabling the extraction of interpretable physical parameters.","pith_inferences":["This method may generalize to other materials with spatially varying properties where PDEs are known.","Hard-coded physics integration could enable more efficient inference compared to penalty-based methods.","Future work might extend the steady-state assumption to time-dependent heat transfer for dynamic scenarios."],"forward_implications":["Improved prediction of thermal responses while respecting physical laws for heterogeneous materials.","Ability to extract interpretable physical parameters such as thermal conductivity variations.","Handling of intra-species diversity in wood without assuming material homogeneity.","Outperformance over pure data-driven CNNs on real-world multimodal wood datasets."],"fun_headline_variants":["Physics constraints embedded in CNNs for wood heat map prediction","Integrating heat transfer equations into neural nets for wood analysis","Physics informed frameworks predict wood thermal responses from images","PDE constrained models yield interpretable wood heat parameters"],"cache_read_input_tokens":64,"weakest_assumption_plain":"The normalized 2D steady-state heat transfer equation derived from the general heat equation is an adequate model for the thermal behavior of the tested wood samples under the experimental conditions.","fun_headline_variants_meta":{"raw":{"variants":["Physics constraints embedded in CNNs for wood heat map prediction","Integrating heat transfer equations into neural nets for wood analysis","Physics informed frameworks predict wood thermal responses from images","PDE constrained models yield interpretable wood heat parameters"]},"model":"grok-4.3","cost_usd":0.005689,"raw_usage":{"total_tokens":2713,"prompt_tokens":660,"num_sources_used":0,"completion_tokens":62,"cost_in_usd_ticks":56887000,"prompt_tokens_details":{"text_tokens":660,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1991,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":660,"tokens_out":62,"duration_ms":13425,"temperature":1.0,"reasoning_tokens":1991,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-26T08:46:44.979703+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A held-out set of wood samples where a standard convolutional neural network produces lower prediction error on thermal maps than either the penalty-based or hard-coded physics models, or where extracted parameters fail to correlate with separate lab measurements of thermal conductivity.","supporting_citations":[],"review_version":1}