36 results
Filters
Settings
Now showing1 - 10 of 36
- Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , LLM-driven design of physics-constrained constitutive models: Two agents are better than one(Springer, 2026-10-01); ; ; ; Developing constitutive models that capture how materials deform under load traditionally requires years of specialized expertise in continuum mechanics, machine learning, and scientific programming. Large language models (LLMs) have recently been shown to lower this barrier by generating constitutive models on demand, but existing single-agent pipelines lack systematic checks that the resulting models respect fundamental physical laws. To close this gap, we introduce the first multi-agent LLM-driven approach for constitutive model generation: a Creator agent proposes a model tailored to the data, while an Inspector agent critically audits each proposal against nine physical constraints and returns it for refinement whenever a violation is detected. We demonstrate this concept with constitutive artificial neural networks (CANNs) and benchmark it on brain tissue and rubber as isotropic materials, and on porcine skin tissue as a transversely isotropic material with a preferred fiber direction, using two different LLM backbones (Claude Opus 4.7 and Kimi K2.5). Whether a generated model satisfies the physical constraints is assessed numerically, by probing each constraint across a broad sample of deformation states, rotations, and perturbation directions. Adding the Inspector raises the share of exported models that pass all these checks from 90% to 95% for Opus and from 47% to 60% for Kimi. In addition, the generated models are on par with or even surpass expert-designed models in accuracy, extrapolate reliably beyond the training data, and generalize remarkably well to unseen loading paths. Separating generation from inspection thus turns LLM-driven constitutive modeling into a substantially more trustworthy process. The paradigm is deliberately technique-agnostic and scales automatically with advances in LLM capability, opening a promising path toward automated, physics-aware model discovery.Publicationtype: Journal Article9 28 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Different alloy, different story? Impact of transfer learning on predictive models for in silico screening of Mg dissolution modulators(Elsevier, 2026-09-26); ; ; ; Ideally, datasets for training machine learning models consist of thousands of versatile data samples — a scale that is seldom met in materials research to date. Transfer learning (TL) is widely used to mitigate data scarcity in available training sets by leveraging knowledge from related domains, since models pretrained on auxiliary datasets may learn structure–property relationships that are transferable from one to another. In this work, the potential of TL for predicting the performance of small organic molecule dissolution modulators is probed using corrosion inhibition responses of eleven Mg-based materials taken from the previously published ExCorr database. A significant improvement in prediction accuracy is observed when transfer learning is applied. However, for the dataset configuration studied here, the apparent cross-alloy TL gains are shown not to arise from transferable structure–activity relationship (SAR) knowledge but rather from a learned alloy-context signal embedded in the data. Seven Mg-based materials—six commercial alloys plus the high-purity variant HPMg51—form a shared-SAR cluster with strongly correlated inhibitor rankings, such that molecules labelled on one material are readily predictable for others within the cluster. Within this shared-SAR cluster, extensive overlap reduces the effective training pool from approximately 900 records to only around 180 unique molecules, placing the task firmly in a small-data regime. In this setting, an in-context large language model (Gemini 3.1 Pro) is the only approach that successfully learns meaningful SAR, with its pretrained knowledge providing the transfer that the explicit source dataset fails to deliver. Seven Mg-based materials in the ExCorr database form a shared-structure–activity relationship (SAR) cluster (pairwise Spearman r=0.67–0.92). Under molecule-disjoint filtering the cross-alloy TL gain on ExCorr splits into a null inhibitor-SAR contribution and a dominant alloy-context contribution. Cross-alloy overlap shrinks ExCorr’s training pool from ≈900records to ≈180unique molecules, only an in-context LLM (Gemini 3.1 Pro) stays predictive.Publicationtype: Journal Article9 8 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , In-context molecular property prediction with LLMs: a blinding study on memorization and knowledge conflicts(American Chemical Society (ACS), 2026-09-17); ; ; ; The capabilities of large language models (LLMs) have expanded beyond natural language processing to scientific prediction tasks, including molecular property prediction. However, their effectiveness in in-context learning remains ambiguous, particularly given the potential for training data contamination in widely used benchmarks. This paper investigates whether LLMs perform genuine in-context regression on molecular properties or instead rely on verbatim retrieval of memorized target values. Furthermore, we analyze the interplay between pretrained knowledge and in-context information through a series of progressively blinded experiments. We evaluate nine LLM variants across three families (GPT-4.1, GPT-5, Gemini 2.5) on three MoleculeNet data sets (Delaney solubility, Lipophilicity, QM7 atomization energy) using a systematic blinding approach that iteratively reduces available information, complemented by 0-, 60-, and 1000-shot in-context sample sizes as an additional control for information access. To validate the memorization analysis and the blinding experiments, we add a positive and a negative control for the memorization experiments and structural reference baselines for the multishot experiments, as well as bootstrap confidence intervals for all results. We find no evidence of verbatim retrieval on the legacy benchmarks and show that blinding exposes conflicts between pretrained knowledge and in-context information. This work provides a principled framework for evaluating molecular property prediction under controlled information access.Publicationtype: Journal Article3 1 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Solvation-controlled accessibility and irreversible finalization of aminoxy intermediates in TEMPO/Benzothiadiazole nonaqueous redox flow batteries(American Chemical Society (ACS), 2026-08-19); ; ; ; Understanding how reactive intermediates form and evolve during electrochemical operation is essential for improving the stability of molecular energy-storage systems. Here, we investigate the degradation chemistry of an all-organic nonaqueous redox flow battery based on a nitroxide catholyte and a benzothiadiazole anolyte. By combining electrochemical analysis, full-cell cycling experiments, and molecular dynamics simulations, we reveal how the solvation environment governs both the electrochemical accessibility and the chemical fate of transient reduced intermediates. To rationalize these effects, we introduce a chemically informed framework that decouples electrochemical exposure of parasitic pathways from the probability that such events culminate in irreversible product formation. This minimal model quantitatively captures the relationship between coulombic efficiency and capacity loss across a range of ionic liquid electrolytes, providing a unified description of degradation behavior under realistic operating conditions. Molecular-level analysis further shows that electrolyte-dependent solvation and local coordination environments modulate the stabilization and reactivity of key intermediates, thereby controlling the balance between reversible redox processes and permanent inventory loss. These findings establish a direct link between electrolyte solvation structure and electrochemical degradation pathways, offering a general strategy to diagnose and mitigate failure mechanisms in molecular electrochemical systems. More broadly, the approach provides a conceptual framework for connecting reactive intermediate chemistry with macroscopic device performance in organic energy-storage technologies.Publicationtype: Journal Article7 5 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Dirichlet-type absorbing boundary conditions for ordinary state-based peridynamicsWe present time-domain Dirichlet-type absorbing boundary conditions (ABCs) for 2D ordinary state-based peridynamics using the Linear Peridynamic Solid (LPS) model, aimed at efficiently truncating unbounded elastodynamic domains while retaining the flexibility of arbitrary Poisson ratios. The proposed ABCs are built from a semi-analytical far-field representation expressed as a finite superposition of plane-wave modes satisfying the discrete LPS dispersion relation, so that outgoing waves are absorbed consistently with the numerical P- and S-wave branches and their Poisson-ratio-dependent polarizations. A cloud-based collocation and least-squares procedure eliminates the unknown modal amplitudes locally and yields precomputable updating operators that prescribe boundary displacement and velocity directly within explicit time stepping, avoiding spatial derivatives, transform techniques, auxiliary evolution equations, and surface-correction procedures at truncated horizons. Unlike our earlier bond-based ABC formulations, the present work requires a new state based far-field theory, a new discrete dispersion operator with coupled P- and S-wave branches for general Poisson ratios, and an additional treatment of the nonlocal dilatation field. Numerical benchmarks show that the resulting ABCs strongly suppress nonphysical reflections over a broad range of material parameters, providing a simple and robust alternative to absorbing layers.Publicationtype: Journal Article21 19 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , The Cahn–Hilliard residuum and its implications for critical-point wetting at solid solution surfacesAs part of their seminal 1958 analysis of nonuniform solutions, Cahn and Hilliard [J. Chem. Phys. 28, 258–267 (1958)] identified the Laplacian of the composition field as the key contribution to the excess free energy in solids with a conserved network of discrete atomic sites. Integration by parts converted to the now familiar gradient-square form underlying phase-field simulations and studies of critical-point wetting. The residuum from the integration by parts has since been ignored. Here, we examine its implications using critical-point wetting at the surface of an Ising-type solid solution for a representative case. As a benchmark, atomistic Monte Carlo simulation shows a continuous increase in wetting-layer thickness up to the bulk critical temperature. In contrast, the classical gradient-square continuum model predicts a first-order wetting transition. A more realistic Laplacian-based continuum formulation does not readily admit physically meaningful minimization by standard variational methods. As a discrete approach, a plane-by-plane model with an excess energy based on a second-difference (curvature) operator yields solutions consistent with the benchmark. Summation by parts transforms this operator into a first-difference- (gradient-) square form, and neglect of the residual term again leads to an incorrect first-order transition. Thus, the choice between curvature and gradient-square formulations alters the predicted character—first-order transition or continuous evolution in layer thickness—of the wetting, underscoring the fundamental importance of this choice.Publicationtype: Journal Article23 30 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Surface and boundary corrections in peridynamics using optimized nodal influence weights(Wiley, 2026-07-14); ; ; ; Peridynamics (PD), similarly to some other nonlocal continuum theories, exhibits truncated interaction horizons near free surfaces, cracks, and voids in bounded domains. This loss of neighbors causes artificial surface effects and inconsistencies in the derivative and energy operators that persist under discretization refinement, limiting PD accuracy and robustness. First, the analysis and numerical isolation of the effects of horizon truncation on these operators are carried out, showing how it induces surface softening in bond‐based and stiffening/softening in state‐based PD formulations. Then, a purely PD‐based surface correction is proposed to restore full‐horizon behavior while preserving the original horizon radius and bond topology. Each node is assigned a scalar influence weight, and every bond is scaled by the average of the endpoint weights to maintain symmetry. The nodal influence weights are computed in a preprocessing optimization step enforcing agreement between truncated‐ and full‐horizon derivative and energy operators. Unlike existing approaches, the proposed method does not rely on ghost particles, variable horizons, or reference solutions from classical continuum mechanics (CCM). Benchmarks in one, two, and three dimensions, including a dynamic brittle fracture test, show that the optimized nodal influence weights reduce boundary‐induced surface effects, improve energy consistency, and accelerate convergence of PD solutions. Analytical evaluation of the calibration targets over full‐horizons also improves interior volume integration accuracy. The proposed scheme is simple to implement in existing PD codes, incurs only a modest preprocessing cost, performs robustly for flat and curved boundaries on both regular and irregular point sets, produces reliable crack paths, and applies to both bond‐based and ordinary state‐based PD formulations. An open‐source Python implementation, perifit, is provided.Publicationtype: Journal Article44 66 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Generalization of LSTM and CNN autoencoders for anomaly detection across orthogonal and longitudinal turningMachine learning models enable automated process monitoring in manufacturing and support adaptive quality assurance and predictive maintenance systems. Autoencoders have emerged as a promising approach to unsupervised anomaly detection in processes. However, their application to turning processes, particularly their generalizability across different process variants such as varying machine kinematics, cutting geometries, and process variability, remains largely unexplored. Existing studies mainly focus on controlled laboratory environments using single machine tools and standardized parameters, limiting industrial scalability. This work addresses this gap by systematically comparing long short-term memory (LSTM) and convolutional (CNN) based autoencoders developed in collaboration with two research projects. The models are trained and evaluated on two harmonized datasets derived from orthogonal and longitudinal turning processes. The datasets differ with respect to machine tools, kinematics, and cutting tools. Each model is trained on both its native process data and on the respective other dataset to evaluate cross-domain performance. The results on normalized reconstruction errors show that LSTM-based autoencoders yield higher temporal sensitivity, whereas CNN-based autoencoders exhibit greater reconstruction stability but reduced sensitivity to transient dynamics. Moreover, relative reconstruction errors are additionally evaluated in the original physical force units to improve process interpretability. This analysis reveals larger relative deviations for LSTM models due to their heightened sensitivity. Furthermore, the variability of the training dataset is identified as a key factor influencing out-of-distribution generalization performance. The findings demonstrate the generalizability of autoencoder based monitoring approaches across different cutting conditions and provide insights for data-driven and grey-box modeling frameworks in manufacturing.Publicationtype: Journal Article45 7 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , LLMs as designers of physics-constrained neural networks for constitutive modeling: a demonstration on hyperelastic solids(Springer, 2026-06-12); ; ; ; Large language model (LLM)-based frameworks extend beyond agents; they also enable the on-demand creation of specialized scientific and engineering tools. We demonstrate this concept in the field of solid mechanics. There, so-called constitutive models describe the relationship between body deformation and mechanical stress. They are essential for both the scientific understanding and industrial application of materials. However, even recent data-driven methods of constitutive modeling, such as constitutive artificial neural networks (CANNs), still require substantial expert knowledge and human labor. We present a framework in which an LLM generates a CANN on demand, tailored to a given material class and dataset provided by the user. The framework covers LLM-based architecture selection, integration of physical constraints, and complete implementation. Evaluation on three benchmark problems demonstrates that LLM-generated CANNs achieve accuracy and generalization comparable to or greater than manually engineered counterparts, while substantially reducing the expertise required for constitutive modeling.Publicationtype: Journal Article33 25 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , Concurrently coupling particle and continuum simulations to study block copolymer membrane fabrication(2026-06-02); ; ; ; We present a concurrent multiscale simulation framework for membrane fabrication from a block copolymer solution via self-assembly and nonsolvent-induced phase separation, combining simulations of a soft, coarse-grained particle model with the continuum Uneyama-Doi model. The computationally intensive particle model provides a molecularly resolved description of micro- and macrophase separation, including thermal fluctuations, while the computationally efficient continuum model captures the process-driven self-assembly of nonequilibrium membrane morphologies on large length and time scales. Central to the approach is a machine learning-guided adaptive coupling strategy, implemented through a coordinator library, which predicts the evolving spatiotemporal subdomain where high-fidelity particle simulations are required and dynamically allocates computational resources accordingly. The two complementary models are concurrently coupled through a consistent exchange of the solvents' fluxes, enabling the treatment of spatially inhomogeneous, multicomponent systems with diffusive transport, micro- and macrophase separation, and vitrification. This adaptive strategy enables predictive simulations of membrane formation on experimentally and technologically relevant length and time scales, reaching micrometers and minutes. As an application, we examine the influence of polymer concentration in the initial casting solution. The framework is general and extensible, providing a computational tool for investigating nonequilibrium structure formation in complex multicomponent soft-matter systems.Publicationtype: Journal Article18