Organizer Affiliation: Prof. Wael A. Altabey (Professor), Department of Mechanical Engineering, Faculty of Engineering, Alexandria, University, Alexandria, 21544, Egypt
Email: wael.altabey@gmail.com
A balance between model complexity, accuracy, and computational cost is a central concern in numerical simulations. High-quality finite element (FE) mesh generation remains one of the most computationally expensive and bottlenecked phases in numerical analysis, particularly for complex geometries with multi-scale features. Traditional meshing techniques often require extensive manual intervention or suffer from element distortion, leading to numerical instability and reduced solver accuracy. This special session presents recent advancements in intelligent meshing representations that leverage machine learning algorithms, geometric deep learning, and adaptive neural representations to automate and optimize the meshing process. Ultimately, these advancements bridge the gap between computer-aided design (CAD) paradigms and automated high-fidelity engineering simulations, paving the way for next-generation, autonomous numerical analysis.
This special session aims to bring together researchers, mathematicians, and software developers at the intersection of computational mechanics and artificial intelligence to explore next-generation intelligent meshing representations. We welcome contributions focusing on the deployment of machine learning architectures—such as Graph Neural Networks (GNNs), geometric deep learning, and adaptive neural fields—to fundamentally redefine how geometries are discretized and evaluated for finite element analysis (FEA). By investigating intelligent mesh sizing functions, physics-informed neural network (PINN) guided adaptive refinement, and instant error prediction, this session intends to showcase techniques that eliminate the manual bottleneck, drastically optimize wall-clock computational time, and guarantee high-fidelity convergence.
Ultimately, this session serves as a collaborative forum to map out the transition from classical, rule-based meshing algorithms toward autonomous, end-to-end, high-quality simulation workflows capable of driving next-generation engineering paradigms.
Contributions on the following theme are welcome (but they need not be limited to this list):
Theme A: AI-Driven & Neural Mesh Representations
- Graph Neural Networks (GNNs) for non-Euclidean mesh topology optimization and node connectivity prediction.
- Geometric Deep Learning architectures interfacing directly with CAD Boundary Representations (B-Reps).
- Neural Sizing Functions and neural field definitions for instantaneous 2D and 3D geometric discretization.
- Deep learning models for structural, fluid, and multi-physics grid generation.
Theme B: Automated Quality Control & Error Estimation
- Machine Learning Classifiers for instantaneous detection of element distortion, skewness, and critical Jacobian metrics.
- Data-driven a posteriori error estimation to bypass expensive solver-driven iterative loops.
- Physics-Informed Neural Networks (PINNs) coupled with finite element solvers for localized, high-gradient boundary layer refinement.
- Surrogate modeling for mesh independence validation and mesh sensitivity analysis.
Theme C: Numerical Performance & Solver Convergence
- Impact of intelligent representations on the condition number of global stiffness matrices and iterative solver stability.
- Comparative benchmarks: Traditional adaptivity vs. machine-learning-driven adaptive refinement.
- Reduction of preprocessing wall-clock times in large-scale industrial structural, thermal, and fluid simulations.
Theme D: Autonomous & Advanced Design Pipelines
- Integration of intelligent meshing within automated Topology Optimization and generative inverse design frameworks.
- Agentic AI and LLM-assisted pipelines for autonomous CAD-to-Mesh-to-Solution orchestration.
- Handling extreme mesh distortions in large-deformation mechanics, crack propagation, and fluid-structure interactions (FSI).