Solidification-Driven 3D Generative Models: A Framework for Temperature-Field-Controlled, Phase-Transition-Based Three-Dimensional Structure Generation
Abstract
Three-dimensional (3D) generative modeling has enabled the synthesis of geometric structures, spatial representations, and complex visual content through approaches such as 3D diffusion models, neural radiance fields (NeRFs), 3D Gaussian splatting, and point-cloud generative models. However, these approaches do not necessarily treat the physical process through which a liquid transforms into a solid as the primary computational mechanism responsible for generating a 3D structure. This paper proposes Solidification-Driven 3D Generative Models (SD3GM), a conceptual framework in which solidification is treated as a generative computational process rather than merely a physical phenomenon to be simulated or a constraint imposed on an already generated object. The proposed framework generates three-dimensional structures through the coupled evolution of temperature fields, phase-transition variables, solidification fronts, nucleation events, and spatial competition between growing solid regions. Instead of directly predicting a final geometry, the model constructs a structure progressively as the material transitions from a liquid-dominated state toward a solid-dominated state. A thermal-field module determines the spatial distribution of cooling conditions, a phase-transition module estimates local solidification states, and a front-evolution module governs the advancement and interaction of solidifying regions. A geometry-decoding module converts the resulting physical state into explicit 3D representations, including meshes, voxels, point clouds, and implicit surfaces. The framework also supports inverse generation, in which a target geometry or structural property guides the selection of thermal boundary conditions, cooling schedules, nucleation parameters, and material-specific settings. A mathematical formulation, modular architecture, training objectives, and experimental evaluation protocol are presented. This paper establishes a proposed methodology rather than a validated generative system; its performance, generality, and novelty relative to existing solidification simulation and generative modeling methods remain to be established experimentally.
Related
- Phase-Transition-Driven 3D Generative Computing: A State-Evolution Framework for Constructing Three-Dimensional Geometry through Nucleation, Crystallization, Solidification, and Phase Separation
- Numerical Study on Dendritic Growth During Solidification in Vibration-Assisted GMAW of Q450NQR1 Steel Using a Coupled Temperature–Cellular Automaton Model
- Increasing high-temperature creep life via improved directional solidification: A novel Ni-based single crystal superalloy reinforced by in-situ MC carbide fibers
- Robust generative transition-state models for unseen chemistry
- Viscosity-Informed Generative Actor-Critic for High-Dimensional Stochastic Optimal Control
BibTeX
@misc{zhang2026solidificationdriven,
title = {{Solidification-Driven 3D Generative Models: A Framework for Temperature-Field-Controlled, Phase-Transition-Based Three-Dimensional Structure Generation}},
author = {Jincheng Zhang},
howpublished = {Zenodo},
year = {2026},
month = oct,
doi = {10.5281/zenodo.23635902},
url = {https://zenodo.org/records/23635902},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Solidification-Driven 3D Generative Models: A Framework for Temperature-Field-Controlled, Phase-Transition-Based Three-Dimensional Structure Generation
AU - Jincheng Zhang
PY - 2026
DA - 2026/10/25
PB - Zenodo
DO - 10.5281/zenodo.23635902
UR - https://zenodo.org/records/23635902
AB - Three-dimensional (3D) generative modeling has enabled the synthesis of geometric structures, spatial representations, and complex visual content through approaches such as 3D diffusion models, neural radiance fields (NeRFs), 3D Gaussian splatting, and point-cloud generative models. However, these approaches do not necessarily treat the physical process through which a liquid transforms into a solid as the primary computational mechanism responsible for generating a 3D structure. This paper proposes Solidification-Driven 3D Generative Models (SD3GM), a conceptual framework in which solidification is treated as a generative computational process rather than merely a physical phenomenon to be simulated or a constraint imposed on an already generated object. The proposed framework generates three-dimensional structures through the coupled evolution of temperature fields, phase-transition variables, solidification fronts, nucleation events, and spatial competition between growing solid regions. Instead of directly predicting a final geometry, the model constructs a structure progressively as the material transitions from a liquid-dominated state toward a solid-dominated state. A thermal-field module determines the spatial distribution of cooling conditions, a phase-transition module estimates local solidification states, and a front-evolution module governs the advancement and interaction of solidifying regions. A geometry-decoding module converts the resulting physical state into explicit 3D representations, including meshes, voxels, point clouds, and implicit surfaces. The framework also supports inverse generation, in which a target geometry or structural property guides the selection of thermal boundary conditions, cooling schedules, nucleation parameters, and material-specific settings. A mathematical formulation, modular architecture, training objectives, and experimental evaluation protocol are presented. This paper establishes a proposed methodology rather than a validated generative system; its performance, generality, and novelty relative to existing solidification simulation and generative modeling methods remain to be established experimentally.
ER -