From Filter Media Images to Performance Prediction: Combining AI and Physics-Based Simulation to design next generation filter media

<p class="font_8">Developing high-performance filter media requires understanding how their complex </p>
<p class="font_8">microstructure affects pressure drop, filtration efficiency, particle deposition, and filter lifetime. </p>
<p class="font_8">Artificial intelligence (AI) can support this process by extracting quantitative information from </p>
<p class="font_8">experimental image data and making it available for digital material development.</p>
<p class="font_8"><br></p>
<p class="font_8">This presentation shows how AI and physics-based simulation can be combined to support filtration </p>
<p class="font_8">R&D from microstructure characterization to performance prediction. AI-assisted image processing </p>
<p class="font_8">and segmentation transform micro-CT and other 3D image data into digital representations of real </p>
<p class="font_8">filter media. For fibrous materials, neural networks can identify individual fibers and determine </p>
<p class="font_8">characteristics such as fiber diameter, orientation, length, and curvature. This data can also serve </p>
<p class="font_8">as a basis to create representative digital models.</p>
<p class="font_8"><br></p>
<p class="font_8">Physics-based simulations then connect these resulting structures to filtration performance. </p>
<p class="font_8">Properties such as permeability, pressure drop, filter efficiency, particle deposition, clogging </p>
<p class="font_8">behavior, and filter capacity can be predicted for existing or virtually modified media. Structural </p>
<p class="font_8">parameters such as fiber diameter, orientation, porosity, gradients, thickness, and pleat geometry </p>
<p class="font_8">can then be varied in the model systematically.</p>
<p class="font_8"><br></p>
<p class="font_8">This combination of AI-based characterization, digital modeling, and simulation supports the </p>
<p class="font_8">efficient exploration of new filter designs with a focus on meeting performance requirements</p>

Speaker Bio

Dr. Andreas Wiegmann is CEO of Math2Market GmbH in Kaiserslautern, Germany, and President of Math2Market North America, Inc. He holds a Diploma in Mathematics with a minor in Computer Science from the Technical University of Karlsruhe (now KIT) and a PhD in Mathematics from the University of Washington, Seattle. USA. After postdoctoral research at UC Berkeley and Lawrence Berkeley National Laboratory, he joined the German Fraunhofer Institute for Industrial Mathematics (ITWM) in 1999.

His work on random 3D material models and image-based material property computations lead to the creation of the GeoDict software platform in 2001 and the founding of Math2Market in 2011. Today, GeoDict is used by more than 500 renowned companies and research institutions worldwide for digital material R&D. Andreas has authored more than 180 peer reviewed publications, with around 8,200 citations.

Presentation time

December 9, 2026
10:40 AM to 11:50 AM EST

Abstract

Developing high-performance filter media requires understanding how their complex

microstructure affects pressure drop, filtration efficiency, particle deposition, and filter lifetime.

Artificial intelligence (AI) can support this process by extracting quantitative information from

experimental image data and making it available for digital material development.

This presentation shows how AI and physics-based simulation can be combined to support filtration

R&D from microstructure characterization to performance prediction. AI-assisted image processing

and segmentation transform micro-CT and other 3D image data into digital representations of real

filter media. For fibrous materials, neural networks can identify individual fibers and determine

characteristics such as fiber diameter, orientation, length, and curvature. This data can also serve

as a basis to create representative digital models.

Physics-based simulations then connect these resulting structures to filtration performance.

Properties such as permeability, pressure drop, filter efficiency, particle deposition, clogging

behavior, and filter capacity can be predicted for existing or virtually modified media. Structural

parameters such as fiber diameter, orientation, porosity, gradients, thickness, and pleat geometry

can then be varied in the model systematically.

This combination of AI-based characterization, digital modeling, and simulation supports the

efficient exploration of new filter designs with a focus on meeting performance requirements