Inspection that catches defects, explains why, and prevents the next
Products made on reels, rolls, and continuous webs, such as industrial films, apparel graphics, webbing, and coiled metals, run at speeds and with frequent spec changes that make defects hard to catch in real time. Problems are often found only after large volumes of material have been produced. Shelfmark, a Pittsburgh-based Physical AI company, provides production intelligence for these lines.
Shelfmark installs cameras at the inspection point and rugged on-prem NVIDIA compute wired to the line. Custom vision models find defects without a reference image across hundreds of SKUs, and each run becomes a defect-mapped digital twin in a browser-based dashboard. The system is fully managed: a single subscription covers the hardware, the platform, and a team that tunes the models as products change.
Beyond detection, Shelfmark connects defect events to inputs like temperature, humidity, and pressure to isolate root causes. In one deployment it traced defects to ambient humidity, and the manufacturer's defect rate dropped 50% after adding controls. Shelfmark reports 99.5% detection accuracy, up to 90% less waste, and half the inspection labor in customer deployments across industrial film, decorated apparel, webbing, and trusses. The long-term goal is production that can optimize itself.