Through OffNominal Vision, I build software for industrial automation: machine vision and inspection, PLC integration, part traceability, and the operations software manufacturers run on. In the Continual Edge AI Lab under Dr. Sen Lin, I research anomaly detection for industrial inspection: models that stay reliable as production changes and run on edge hardware.
Inspection models are usually evaluated on a fixed test set, but production lines drift: tools wear, materials change, lighting shifts. This work benchmarks which approaches stay reliable as conditions change. Published at ECCV 2026.
Defects are rare and labeled data is expensive. I train representations on normal production images so that anything unusual stands out, including logical errors such as missing or misplaced parts.
Inspection has to run where the parts are. I design compact models that run on embedded hardware at line speed, with no dependence on a data center.
OffNominal Vision is my engineering practice. I build software for industrial automation, specializing in manufacturing.
Automated inspection with deep learning and classical vision: X-ray, 2D and 3D cameras, barcode and QR reading. Experience with Cognex In-Sight, LMI Gocator, Luxonis OAK-D and production X-ray platforms.
Connecting vision and software to the line: PLC logic and I/O on Allen-Bradley and Phoenix Contact PLCnext, Modbus and industrial Ethernet, interlocks and pass/fail handshakes with machine controllers.
UHF RFID and barcode tracking from the reader hardware to the database, so every part carries its history: who worked it, where, and when.
Production scheduling, quoting and order management, and live shop-floor boards, integrated with ERP and accounting systems.
Applications in Python and Rust on Raspberry Pi, NVIDIA Jetson and industrial PCs, built to run unattended and recover on their own.
The servers, databases, APIs and secure remote access that tie stations together and get results to the people who need them.
For a Houston-area manufacturer and its machine shop: RFID operator stations on the production floor, X-ray images filed by part identity, and a scheduling and order management system that is taking over from an off-the-shelf ERP.
Replaced a rule-based vision system on a production X-ray line with a deep-learning pipeline that grades every part, integrated with the machine's recipe software and controls. Also built multi-camera QR tracking for parts trays.
Email is the best way to reach me, about research, a project, or both.