Machine Learning and Image Processing for Automatic Visual Inspection

The Large Hadron Collider (LHC) is a particle accelerator at the European Organization for Nuclear Research (CERN). The High-Granularity calorimeter (HGCAL) is a planned upgrade to the Compact Muon Solenoid detector (one of the two largest experiments at the LHC). Quality assurance (QA) for the upgrade requires the manual inspection of thousands of circuit boards before their installation into the detector. Visual inspection can take up to 30 minutes and requires expertise and training. Specialized hardware and software packages can be purchased to accelerate this process, but such systems can cost up to tens of thousands of dollars. We present an inexpensive scanner with a gantry-mounted camera system and a custom software pipeline from image capture and preprocessing to hybrid classical- and machine-learning-driven anomaly detection. This system reduces hardware costs to $350 and scan time to 4 minutes. Further, the operation of the visual inspection system requires minimal on-site training experience. These innovations provide a cost-effective and efficient framework for image-based hardware QA in experimental physics and beyond.