Artificial Intelligence in Nonwovens: From Materials Science to the Factory Floor

AI in nonwovens is following a pattern playing out across manufacturing broadly: investment is accelerating faster than readiness. Manufacturers have doubled their AI spending over the past year, but only 37% say they’re fully prepared to operationalize it at scale, and 62% of AI projects are still stuck in pilot or development, according to a 2026 global survey from Riverbed of manufacturing IT and business leaders. Nonwovens is no exception. From R&D labs experimenting with autonomous discovery to production lines adopting vision systems and predictive maintenance, the industry is edging closer to practical deployments of AI in nonwovens. The promise is real, but so is the readiness gap.

AI in materials science: a glimpse of what’s possible

At the research frontier, organizations like the National Institute of Standards and Technology (NIST) are developing autonomous laboratories where AI and robotics design, execute, and analyze experiments without human intervention. Microscopy and X-ray images that once required painstaking manual review are now interpreted by machine-learning models capable of flagging defects or structural features in real time.

That frontier moved from research proposal to physical building this year: in January 2026, Radical AI opened what New York state officials called the state’s first fully autonomous materials science laboratory, at the Brooklyn Navy Yard. The facility runs roughly 100 AI-driven experiments a day — AI screens billions of candidate materials while a robotic lab synthesizes and tests the most promising ones, feeding results back into the model in a continuous loop. “Our new facility will run materials experiments at a pace and scale that traditional R&D cannot match,” said Joseph Krause, Radical AI’s CEO and co-founder. The lab’s initial focus is aerospace, energy, infrastructure, defense, and manufacturing materials, not nonwovens specifically, but it’s a concrete marker of how fast AI in materials science is moving from pilot to production.

For nonwovens, this hints at a future where AI could accelerate development of biodegradable fibers, PFAS-free treatments, or antimicrobial finishes, speeding R&D cycles from years to months. In essence, it offers a window into how materials innovation may be reimagined. But for now, most nonwovens producers are not running self-driving labs; their AI journey tends to begin closer to the line.

AI is not new to the conversation, either. In 2018, Suominen introduced Intelligent Nonwovens, embedding digital patterns into wipes that could be recognized by smartphones and interpreted using AI. The goal was traceability, counterfeit protection, and consumer engagement. It was one of the sector’s first public attempts to link AI directly to nonwoven substrates. Yet the concept has not visibly expanded since, a reminder that bold digital ideas can be floated, but scaling them to industrial reality is another matter.

Setting the groundwork for AI

AI is most visible today in factory automation and inspection, though in varying degrees of maturity. ANDRITZ’s Metris Copilot, built on Microsoft Azure’s OpenAI Service, integrates anomaly detection and operator assistance directly into line controls. “ANDRITZ is helping to shape the future by using Microsoft Azure to further enhance its autonomous factory solutions,” said Ralph Haupter, President of Microsoft EMEA, when the partnership was announced in 2024. “The deep integration of their products with Azure cloud services is the type of technological innovation that drives sustainable and efficient change in the industry.” By INDEX26 this past May, ANDRITZ was describing the broader Metris platform in more explicit AI terms, pairing production management and process optimization “based on state-of-the-art AI” with its existing Advanced Control Expert (ACE™) and condition-monitoring tools, a sign of how quickly OEM messaging, if not always deployment, is shifting.

DiloGroup offers one of the more concrete examples of self-learning AI already at work on a nonwovens line rather than just in a research pilot. Its Smart Machine Assistant, built with Siemens on the Xcelerator platform, analyzes the complex, interdependent parameters of needlefelt production and recommends adjustments to operators in real time, aiming to cut material and energy waste while holding product quality steady. “The potential of digitalization allows us to open up new fields of application at an early stage,” said Rebekka Dilo, Head of Technical Application Centre at DiloGroup. “Through the partnership with Siemens, we can take this step into the future.” It’s still advisory rather than autonomous, an operator makes the final call, but it’s a real step past pure rule-based control.

In parallel, Trützschler’s T-ONE provides what it calls a digital work environment for recipe management, process monitoring, and data analysis. In practice, it delivers measurable gains in speed and waste reduction, but the improvements stem largely from digitalization and process discipline rather than genuine artificial intelligence. It’s better understood as a stepping stone toward AI-enabled nonwovens production lines. It has been shown to increase production speed by more than 50% and reduce waste by up to 30%.

Similar stories play out with Uster, ISRA Vision, and Mahlo, which supply vision and sensor systems capable of catching defects such as holes, thin spots, or contamination and adjusting parameters in real time. Most of these remain rule-based systems with closed-loop control. They provide tangible value, but the real leap will come when more of these platforms evolve into self-learning models, like DiloGroup’s, that predict and prevent defects before they occur rather than just react to them.

AI in inspection

Conversation around AI application is gaining visibility at industry events. At Cinte Techtextil China 2025, AiDLab hosted an AI in Automated Textile Material Inspection panel. Speakers included Prof. Calvin Wong (CEO, AiDLab), Cheng Yik Hung (Hong Kong Polytechnic University), Eric Sham (AiDLab), and Dorothy Yeung (AiDLab).

AiDLab’s track record is more than academic: WiseEye 2.0 achieves over 90 percent defect detection accuracy at 60 m/min across complex textile structures. The AiTIS system, deployed with Banitore, inspects masks at 500 units per minute with accuracy exceeding 99 percent, a successful commercial deployment of AI in nonwoven healthcare manufacturing.

For nonwovens converters, these projects demonstrate that AI-powered inspection is no longer a lab demo; it’s crossing into high-speed, consumer-critical production environments.

The nonwovens AI readiness gap

The Riverbed numbers cited above are cross-industry, not nonwovens-specific, and that’s the point: no one has published an industry-wide, statistically sound picture of how many nonwovens producers and converters are actually using AI today, what they’re spending, or what’s really holding back wider adoption. The case studies above are proof of concept, not proof of scale.

Three questions sit at the center of that gap, and none of them have a clear answer yet:

1. Can AI predict bond uniformity before it fails a quality audit? AiDLab’s WiseEye 2.0 shows that high accuracy in textiles is possible, but applying that precision to spunbond or meltblown bond uniformity remains unproven at commercial scale.
2. Will algorithms optimize energy consumption on spunbond and meltblown lines at a time of volatile power costs? Metris Copilot and DiloGroup’s Smart Machine Assistant both offer efficiency insights, but the shift from operator guidance to autonomous energy optimization hasn’t happened yet.
3. Who owns the torrents of data generated by AI-enabled platforms, the OEM, the converter, or the brand? Suominen’s Intelligent Nonwovens hinted at the potential and the risks of materials linked directly to data streams, a debate that remains unsettled.

For now, these questions matter less for the answers they lack than for how they frame the path ahead. Closing this gap will take more than another vendor pilot demo. It will take converters, OEMs, and brand owners actually comparing notes on what’s working, what isn’t, and what it costs, the kind of ground-truth data the nonwovens industry doesn’t have yet.

What this means for nonwovens converters

For nonwovens producers, the pragmatic path is to treat today’s tools as enablers rather than endpoints. Deploy vision systems to cut waste, digitalize recipe management for consistency, and pilot predictive maintenance for downtime reduction, but go in with clear-eyed expectations about where the technology actually is versus where the marketing says it is. At the same time, keep an eye on how materials scientists are reshaping the foundations of discovery; labs like Radical AI’s in Brooklyn are a preview of tools that could eventually reach fiber and finish development.

In the world of nonwovens, hype tends to travel faster than adoption. The challenge for converters will be to balance experimentation with pragmatism, and to prepare for a future where the real intelligence is not in the buzzwords, but in the execution. The race is less about adopting AI quickly and more about proving where it actually delivers value, and right now, the industry doesn’t have the data to say for sure where that is.