GeoDict Generate Nonwoven Twin GeoApp

The Generate Nonwoven Twin GeoApp is an automated statistical digital twin tool developed by GeoDict for nonwovens and other fiber‑based materials. It starts from real 3D CT scans—using ImportGeo‑Vol for image processing and segmentation, FiberFind‑AI to identify fiber diameter, length, orientation, and curvature—and then automatically generates a statistical digital twin via FiberGeo.Its key features include: basing the model on real CT scans and FiberFind results; preserving essential fiber characteristics such as diameter, length, orientation, and solid volume fraction; automatically optimizing fiber curvature and crimp; optionally matching through‑thickness density distribution; using differential evolution to automatically search for structural parameters; reducing optimization time via downsampling and small representative regions; outputting a 3D Structure.gdt that can be further edited and simulated; and supporting nonwovens, filter media, composites, and fuel cell gas diffusion layers.

What Is a Statistical Digital Twin of Nonwovens?

The Generate Nonwoven Twin GeoApp generates a statistical digital twin—not a voxel‑by‑voxel copy of the original CT scan, but a new random 3D fiber structure that matches the original sample as closely as possible in key geometric characteristics and physical properties, while its local fiber positions and arrangements may differ.

The GeoApp preserves or matches the following structural information: fiber diameter; fiber length; fiber orientation distribution; fiber solid volume fraction; fiber curvature; fiber crimp; and optionally, through‑thickness density distribution.

This statistical twin serves as a digital substitute for the real material, enabling structural parameter studies, performance prediction, and virtual material optimization.

Workflow: From Real CT Scans to Digital Twin

Step 1: Import and Segment CT Scans – Use ImportGeo‑Vol to import CT or µCT data of real fiber materials. Input can be 3D image files or 2D image sequences (PNG, TIF, etc.). ImportGeo‑Vol can adjust voxel size and image orientation, remove noise and artifacts, improve grayscale image quality, and perform segmentation using thresholding, clustering, or AI‑based methods to convert raw grayscale data into a 3D voxel structure containing fiber and pore phases.

Step 2: Identify Fibers with FiberFind‑AI – Use FiberFind‑AI to analyze the segmented fiber structure. It identifies individual fibers and extracts key statistical parameters needed for digital twin modeling, including fiber diameter distribution, fiber orientation distribution, fiber curvature distribution, individual fiber spatial paths, and fiber/binder material identification.The IdentifyFibers.gdr result file generated by FiberFind is one of the main inputs for the Generate Nonwoven Twin GeoApp.

Automatic Optimization of Curvature and Crimp

Fiber structures with curved morphology are difficult to reconstruct accurately through manual settings. The Generate Nonwoven Twin GeoApp solves this through an automatic optimization process.

During optimization, the GeoApp keeps the following main parameters fixed: fiber diameter, fiber length, fiber orientation distribution, and fiber volume fraction. At the same time, it dynamically adjusts fiber curvature and fiber crimp.FiberGeo generates candidate fiber structures based on current parameters, and the GeoApp compares them with the target statistics from FiberFind. Through repeated generation, comparison, and adjustment, it finally produces a statistical digital twin that matches the real sample as closely as possible in random geometric characteristics.

Optional Through‑Thickness Density Distribution Optimization

In complex fiber materials, density is often not uniformly distributed through the thickness. For example, the fiber content at the surface and in the center of a nonwoven may differ, and different layers in a composite may show significant density variations.The Generate Nonwoven Twin GeoApp can further use the Thickness Estimation.gdr result from MatDict to optimize the Z‑direction density distribution of the digital twin.

When using this feature: IdentifyFibers.gdr and Thickness Estimation.gdr must come from the same original structure; a MatDict license is required; enable Optimize Z Density Distribution in the GeoApp; and load the corresponding Thickness Estimation result file.

This feature is particularly suitable for nonwovens, composites, and fuel cell gas diffusion layers with layered, gradient density, or through‑thickness non‑uniform characteristics.

Differential Evolution‑Based Automatic Optimization

Generate Nonwoven Twin uses Differential Evolution to optimize the statistical digital twin. The algorithm maintains multiple candidate fiber structures simultaneously and iteratively mutates and selects them, gradually bringing the candidates closer to the FiberFind analysis results; when through‑thickness density optimization is enabled, it also matches MatDict's 1D statistics.

Users can set three optimization parameters: Population Size (number of candidate structures compared simultaneously in each round), Global Iterations (number of global differential evolution iterations), and Local Iterations (number of local optimizer iterations).The total iteration scale is calculated as: PS × GI × LI.Larger population sizes and more iterations generally explore the parameter space more thoroughly, but also increase the number of FiberGeo structure generations and total computation time.

Reducing Optimization Time via Downsampling

For high‑resolution or large CT scans, performing multiple rounds of structure generation and optimization directly on the full data may require long computation times. The GeoApp provides a Downsampling Factor to reduce structure resolution during the optimization phase, thereby decreasing computational load.

Key points about downsampling: a larger downsampling factor generally speeds up optimization; a larger factor may also reduce optimization accuracy; the number of voxels in the Z‑direction of the original structure must be divisible by the downsampling factor; and the final generated digital twin retains the original structure's resolution.Math2Market also recommends first selecting a small but representative sample region for initial optimization, then generating the full‑size digital twin once suitable parameters are obtained.

Output Results

Upon completion of optimization, GeoDict automatically opens the result file in the Result Viewer.

Structure Generation Report – The Report tab provides: digital twin generation information, structural parameters used, optimization settings, and basic statistical results of the final generated structure.

Optimization Convergence Curves – The Plots sub‑tab in Results shows the convergence process of different optimization parameters, helping users determine whether target parameters are stabilizing, whether curvature matching is converging, whether through‑thickness density distribution has improved, and whether more optimization iterations are needed.

Final Structure Files – The main outputs in the results folder include: Structure.gdt (the final statistical digital twin structure), Result.gdr (the corresponding FiberGeo result file, showing structure generation input parameters), and intermediate results for different optimization targets.The final Structure.gdt can be used directly in GeoDict for subsequent structure analysis, parameter modification, and physical property simulation.

Official Application Examples

1. Nonwoven Statistical Digital Twin – Math2Market demonstrated a nonwoven digital twin automatically generated from FiberFind results, comparing permeability between the original structure and the digital twin. The original nonwoven structure showed a permeability of 5.21 × 10⁻⁹ m², while the statistical digital twin showed 4.92 × 10⁻⁹ m²—a deviation of only about 5.6%.

2. Fiber‑Reinforced Composite Digital Twin – This case comes from the DigiLaugBeh project supported by the German Federal Ministry for Economic Affairs and Climate Action, with original scans provided by Bosch. Digital twins were generated separately for different material layers. The original composite showed orthotropic Young's moduli of 4.067 / 5.457 GPa, while the statistical digital twin showed 4.137 / 5.402 GPa.

3. Filter Media Layer Digital Twin – This case comes from the ElekSim project, supported by the German Federal Ministry for Economic Affairs and Climate Action, with original scans provided by IUTA. The original filter media layer showed an air permeability of 1.700 × 10⁻¹⁰ m², while the statistical digital twin showed 1.071 × 10⁻¹⁰ m².

4. Fuel Cell Gas Diffusion Layer Digital Twin – Math2Market also demonstrated a digital twin of commercial Toray Paper TGP‑H‑030‑05 gas diffusion layer. The example structure does not include binder and consists mainly of straight fibers, optimized using MatDict's through‑thickness 1D statistics. The original GDL structure showed a relative diffusivity of 0.6082, while the statistical digital twin showed 0.6211.

From Digital Twin to Performance Prediction

The Generate Nonwoven Twin GeoApp focuses on modeling and generating the structure. Once the digital twin is complete, it can be combined with other GeoDict modules for virtual material testing.

FlowDict: Flow and Permeability – FlowDict can compute air or liquid permeability, velocity fields, pressure fields, flow resistance, average velocity under given pressure differences, Reynolds number estimates, and Gurley values based on the 3D fiber structure.

ElastoDict: Mechanical Properties – ElastoDict can compute anisotropic stiffness, effective Young's modulus, stress and strain distributions, deformation, and damage/failure directly on voxel microstructures.

SatuDict: Two‑Phase Flow and Saturation‑Dependent Properties – SatuDict can study the distribution of two immiscible fluids (gas and liquid) in fibrous porous materials and compute capillary pressure, relative permeability, relative diffusivity, and saturation‑dependent thermal and electrical conductivity.

Typical Application Areas

Nonwoven Material Development – Build digital twins from µCT scans to compare the effects of different fiber curvatures, orientations, and density distributions on liquid uptake, air permeability, and mechanical properties.

Filter Media Design – Generate statistical structures for air filtration, liquid filtration, and electret filter media, and analyze permeability, pressure drop, and filtration performance.

Protective Face Mask Filter Media – Design, simulate, and optimize mask filter media to achieve high filtration efficiency, low pressure drop, and economical material usage.

Fiber‑Reinforced Composites – Build 3D models with fiber curvature, orientation, and layering from real CT scans, and predict anisotropic mechanical properties.

Fuel Cell Gas Diffusion Layers – Generate digital twins based on fiber parameters and through‑thickness density distribution of real GDL structures, then compute gas diffusion, flow, and liquid water related properties.

Virtual Parametric Studies – Systematically vary fiber curvature, crimp, or density distribution while keeping other statistical parameters fixed, to study the relationship between microstructure and macroscopic properties.

Required GeoDict Modules

Essential Modules:

FiberGeo – Generates fiber structures and the final digital twin based on statistical parameters

FiberFind / FiberFind‑AI – Identifies fibers and provides target parameters including diameter, length, orientation, and curvature

MatDict – Optional; required when optimizing through‑thickness density distribution

Commonly Used Supporting Modules:

ImportGeo‑Vol – CT image import, filtering, and segmentation

FiberFind‑AI – Fiber and binder identification

MatDict – Thickness and density distribution analysis

FiberGeo – Fiber digital twin generation

FlowDict – Permeability and flow simulation

ElastoDict – Mechanical and deformation simulation

SatuDict – Two‑phase flow and saturation‑dependent properties

ExportGeo‑Abaqus – Model export to Abaqus

The specific module combination depends on the material type and the properties to be computed.