The team, affiliated with the Laboratory of Tribology and Dynamics of Systems at École Centrale de Lyon and CNRS, Tohoku University’s Advanced Institute for Materials Research, and RIKEN iTHEMS, addressed a design problem that becomes increasingly difficult as the size of a textile pattern grows. Even a 10-by-10 weave has more possible yarn arrangements than could realistically be tested individually, and a detailed acoustic simulation can take more than an hour. The researchers sought a way to reduce the number of candidates that required expensive simulation while retaining patterns compatible with basic manufacturing and connectivity requirements.

They began by generating woven patterns that met those structural constraints, then grouped the patterns according to four types of repeating symmetry, including rotational and mirror symmetry. An AI system organized the patterns on a design map, placing structurally similar designs closer together. The researchers simulated a selected set of designs and used those results to train a prediction model.

Once trained, the model could evaluate candidate designs and direct a computer search in less than a minute. Independent, detailed simulations of selected designs were then used to check the predictions, which agreed with the simulation results within 5%. This agreement concerns modeled sound absorption, not measurements from physical acoustic tests.

The study also showed how symmetry can act as a physical design parameter rather than only a visual property. Woven patterns differ in how they distribute yarn crossings, and those crossings shape the microscopic gaps through which air moves. Some arrangements create more winding airflow paths, increasing opportunities for sound energy to be dissipated through friction and heat; others leave straighter and more open paths.

The size of the smallest repeating unit also affects the range of airflow and acoustic behaviors available. Lead author Thanasak Wanglomklang said the approach allows researchers to explore woven structures that would be impractical to examine through trial and error. He emphasized that the method “does not replace physics; it uses physics more strategically to identify promising, manufacturable designs.” Machine learning did not independently determine how sound would interact with a material.

Instead, physics-based simulation supplied training data, symmetry provided an organizing framework, and AI accelerated the identification of designs for closer physical evaluation. The study was published in Communications Materials on 27 August 2026 under the title “AI-assisted symmetry-informed topology optimization of woven materials for broadband sound absorption.” Experimental validation is the next stage.

The team plans to manufacture selected three-dimensional woven prototypes and test their sound absorption experimentally. It also plans to extend the method to multilayer woven materials, larger repeating structures and variations arising from manufacturing processes. Until those tests are completed, the supported result is a faster symmetry-informed route for prioritizing manufacturable candidates in a vast design space, together with close agreement between model predictions and detailed simulation.

The study does not yet demonstrate an improvement in experimentally measured sound absorption.

Subject of Research: Computational optimization of woven textile patterns for broadband sound absorption

Article Title: AI and symmetry accelerate the design of sound-absorbing woven materials

Article References: AI and symmetry accelerate the design of sound-absorbing woven materials. Original publication (Bibliographic details unavailable.)

Image Credits: AI Generated

DOI: Not provided

Keywords: AI-assisted design, sound absorption, woven textiles, symmetry analysis, machine learning, acoustic simulation, materials optimization, topology optimization, multilayer materials, symmetry, accelerate, design