Visual quality control
YOLO · PyTorch · Inspection · Manufacturing
Inspection pipelines with YOLO and PyTorch at Bosch that cut destructive testing from nine tests a day to three.
Overview
Visual inspection and automation for manufacturing, built during three years at Robert Bosch, from October 2022 to September 2025. The work covered inspection systems on the production lines, anomaly detection and the automation of repetitive processes.
What I built
Inspection pipelines with YOLO and PyTorch that check parts on the line, and OCR and Cognex vision inspections that read and verify what is printed on them. Alongside the vision work I automated the optimization of welding parameters with XGBoost, and I built RPA robots that take repetitive work off people’s desks.
Result
The inspection pipelines cut destructive testing from nine tests a day to three. The OCR and Cognex inspections reduced customer complaints by 10%, and the RPA robots saved about 5,000 hours of work a year. The code is private, so this page describes the work rather than linking to it.
Private codebase
Built with
YOLO · PyTorch · OpenCV · Cognex · XGBoost