Lumafield Introduces Quality Agent, a Physical AI Technology that Automates Defect Detection and Root-Cause Analysis for Manufacturers

Built on the first large-scale AI foundation model trained on industrial X-ray CT data, Quality Agent monitors every part and every quality signal in production, detects deviations from known-good product, and delivers root-cause evidence to quality teams before defects escape the factory

SAN FRANCISCO, Sept. 03, 2026 (GLOBE NEWSWIRE) -- Lumafield, a leading manufacturing intelligence platform, today announced Lumafield Quality Agent, an agentic AI system that automates defect detection and root-cause analysis across complex manufacturing processes. Quality Agent runs natively inside Lumafield’s Voyager platform and is built on the first large-scale foundation model trained on industrial X-ray CT data.

Manufacturers use a variety of inspection methods to ensure quality before products leave the factory: human visual inspection, surface-level automated optical inspection (AOI), and destructive cross-sectioning. But the failures that reach customers and trigger recalls are typically novel—”unknown unknowns” that no inspection step was designed to detect. These defects pass through automated production lines undetected until products fail in the field. According to Lumafield’s Cost of Quality Report, over 42% of manufacturers spend at least 5% of their total revenue on quality-related costs, and 58% estimate that at least a quarter of their true quality costs go unaccounted for.

Quality Agent inverts the traditional inspection model. Rather than checking parts against a predefined list of failure modes, it learns the characteristics of known-good products and catches any deviation from that state, with no requirement that a defect be anticipated in advance. It applies that detection at a scale human inspection cannot sustain, monitoring every part across every available data stream: X-ray and CT imagery, machine vision feeds, maintenance logs, and environmental sensors. When Quality Agent identifies something new, it escalates the finding to a quality engineer with the supporting evidence needed to classify it as acceptable variation, a new control plan entry, or a problem requiring containment. The system is designed to augment the powers of quality teams, not replace them.

“Quality teams don't need general-purpose AI retrofitted onto the factory floor; they need domain-specific intelligence built for the physical world,” said Andreas Bastian, Co-Founder and CTO at Lumafield. “Quality Agent is powered by a foundation model trained on hundreds of thousands of real industrial parts. By pairing our volumetric CT data expertise with agentic workflows, quality teams can now catch complex structural defects they didn't even know to look for, trace their origin immediately, and stop escapes before they turn into costly recalls.”

The foundation model underlying Quality Agent is trained on Lumafield’s corpus of industrial CT data, a volumetric record of manufactured products that general-purpose AI models cannot meaningfully interpret, because this type of data is not present in large commercial model training datasets.

In one investigation spanning 1,054 lithium-ion battery cells from 10 manufacturers, Lumafield’s model reliably separated cells by manufacturer, and even discovered that two brands were actually the same cells, produced by the same OEM and resold under different branding by a rewrap vendor.

“Manufacturing is a far harder problem than it is given credit for, and robots alone will not automate it,” said Bastian. “Making something is straightforward. Making safe products that perform exactly as intended is difficult, and it depends on understanding quality and the processes behind it. Automating quality and automating process control are central to automating manufacturing as a whole.”

To learn more about Lumafield Quality Agent or request a demonstration, visit https://www.lumafield.com/manufacturing-ai/quality-agent

About Lumafield

Lumafield is a manufacturing intelligence platform that gives engineers the power to see inside their work in unprecedented detail, at every stage of the product development process. Lumafield’s industrial X-ray CT scanners and cloud-based analysis software have transformed both product development and high-volume manufacturing in industries as diverse as medical devices, automotive, aerospace and defence, electronics, and consumer packaging.

Founded in 2019 and headquartered in San Francisco, with additional offices in Cambridge, Mass. and Los Angeles, Lumafield has received funding from investors including Lux Capital, Kleiner Perkins, DCVC, Spark Capital, Matter Venture Partners, IVP, G2 Venture Partners, Wellington Management, Haystack Ventures, Tony Fadell’s Build Collective, and Figma founder Dylan Field.

Media Contact

Email press@lumafield.com for interviews or additional assets.