Small-data inverse design is challenging in engineering informatics when observations are heterogeneous, mixed-type, and constrained by physical relations among design variables. This work proposes a topology-aware surrogate framework guided by an Incremental Transformer (INCRT) for physics-constrained inverse design, applied to geopolymer mixture design. The method integrates intrinsic-dimensionality analysis, mixed-variable design-space representation, tabular surrogate prediction, INCRT-based manifold rationalisation, and constrained inverse optimisation. Using a public benchmark of fly-ash and slag-based geopolymer concrete mixtures with compressive-strength and carbon-emission targets, the high-dimensional design space proves strongly redundant, organising around fewer effective mixture regimes. Compressive strength requires nonlinear tabular surrogates, while carbon emission is largely determined by composition and well recovered by regularised linear models. INCRT thus acts not as a replacement for tabular predictors but as a rationalisation layer providing prototype regimes and a manifold-support score for inverse design. Three strategies are compared: unconstrained surrogate optimisation, physics-constrained optimisation, and topology-aware physics-constrained optimisation. Unconstrained optimisation can match target strength but may yield physically invalid or off-manifold candidates; physics-only constraints do not always ensure data support. The topology-aware strategy yields candidates balancing target compliance, carbon reduction, physical admissibility, and proximity to the learned feasible manifold. The framework aims not to replace experimental validation but to support screening of credible candidate mixtures from small, mixed, physically constrained engineering datasets.
A machine learning-assisted inverse parameter prediction framework that couples a Random Forest-based inverse surrogate model with compliance-minimization topology optimization and suggests that the framework may serve as a rapid design initialization tool for lightweight structural applications.
Dailin Li, K. Jiang, Miao He et al.· AIP Advances· 0 citations
Basalt fiber reinforced concrete (BFRC) has recently attracted increased attention for improving durability, mechanical strength, and chemical resistance concerning harsh environmental conditions. The most noticeable gap is left in being able to predict long-term performance accurately and optimize that performance because of the complexities that arise from the multiscale interactions between fibers, matrix, and environmental stressors. This study, therefore, offers a highly unified and multiscale machine learning framework by pulling together five disaggregated analytical models into a single predictive-optimization pipeline prearranged for basalt fiber reinforced concrete. The physics-augmented graph attention transformer network (P-GATNet) is expected to embed interfacial physics within graph-based message passing to capture load-driven mechanical responses, resulting in highly accurate strength and fracture evolution predictions (e.g., flexural R² ≈ 0.97). The spectral decomposition assisted degradation model uses Hilbert-Huang-based spectral analysis, which then decouples degradation mechanisms for accurately forecasting alkali resistance and damage kinetics with an error of less than 4.5%. The multi-agent physics reinforcement optimizer (MAPRO) jointly optimizes the strength and chemical performance by modeling competing failure mechanisms via cooperative agents. For improved representation of features, the deep morphological encoder with multi-modal fusion (DME-MMF) marries image-derived morphological embeddings with experimental tabular data, thus enhancing the interpretability and accuracy of predictions. Lastly, the transformer-based inverse composite generator enables reverse material design by producing feasible basalt fiber reinforced concrete formulations that satisfy predetermined strength and durability targets at an approximate success rate of 93%. This approach improves predictive fidelity, interpretability, and design in basalt fiber reinforced concrete.
V. Vairagade· Journal of Materials Science...· 0 citations
An artificial intelligence framework that incorporates a physics-constrained inverse-design system (PHICS) built upon a directed acyclic graph (DAG) architecture for CPCMs, integrating interface, phase, and carrier engineering.
Dongliang Ding, Min-hao Zou, Ruoyu Huang et al.· National Science Review· 0 citations
A physics-guided generative AI framework for property-targeted porous media design, combining a property-aware variational autoencoder, a conditional latent diffusion model, and an independently trained differentiable structure-to-property surrogate is proposed.
Novel methodologies in machine learning enable non-destructive prediction of mechanical properties through the use of either material components or material imaging. Predictive technologies are continuously being developed in the field of mechanical and materials engineering. However, when it comes to inverse problems, the identification of the optimal microstructural content to create the material with the desired properties is often complicated and ill posed. In fact, the inverse problem is not unique, and multiple material combinations can lead to the desired properties. Working with biobased materials adds a layer of complexity to the inverse problem since in most cases biosourced components are mixtures rather than pure components. This work establishes a methodology for the prediction of material properties based on imaging of lignin-containing polyurethane rigid foams and solving the inverse problem by generating the required lignin-based formulation leading to the desired mechanical properties. The methodology employs a novel learnable kernel principal component analysis, based on the combination of the non-linear model order reduction technique with machine learning autoencoders mapping the physical data into a large dimensional space. The results are material generation through required mechanical properties, mechanical properties identification from microstructure imaging, microstructure generation from mechanical properties, and (bio)chemical combinations. The predictions of the mechanical properties are within less than 2.3% on average from the actual properties on both training and testing sets, while the inception shows an accuracy of 0.893 on the train set and 0.891 on the test set.
C. Ghnatios, I. Hage, J. E. Quinsaat et al.· Advanced Modeling and Simula...· 0 citations
Bio‐based epoxy resins from resources like cardanol are promising sustainable alternatives, yet they typically exhibit weaker thermomechanical properties than conventional, petroleum‐based ones. To balance performance with sustainability, a machine learning (ML) framework for optimizing partially bio‐based formulations in a 13‐component design space is presented. Complementary ML strategies were empolyed: Bayesian optimization (BO) for glass transition temperature () maximization, active learning and random design for diverse dataset generation, and inverse design (ID) for the virtual screening of targeted properties. The Pareto front was refined through strategic manual experiments and virtual screening. The results demonstrate that BO effectively navigates high‐dimensional formulation spaces to find high formulations (160°C) after only 22 new experiments. The final ML model achieved a mean absolute error of ca 11°C and an coefficient of 0.91 for the prediction of , after iterative model refinement. This work establishes a data‐efficient approach for accelerating the development of sustainable bio‐based materials with tailored properties.
Nathalie Wunder, Rodrigo Q. Albuquerque, Florian Max et al.· Journal of Applied Polymer S...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
