Predictive Maintenance for Electrical Wiring Faults

Electrical systems in aviation are responsible for the effective and safe operation of essential systems of modern aircraft. Aviation maintenance personnel routinely inspect and maintain these systems to guarantee their integrity and reliability. Currently, aircraft inspections are performed manually by maintenance personnel with a flashlight, a mirror, and a notebook. This manual inspection method comes with its disadvantages, such as high worker hours and dependence on the inspector’s expertise to correctly identify and assess all discrepancies they come across. This is a dataset to support automatic optical inspection tool for electrical components using computer vision.

This dataset was built from scratch and hand-labeled. This dataset includes images of a single computer case with multiple configurations of two power supply units, two cooling components, and four SATA cables with several wiring configurations, including various induced faults. Images were collected using a Canon S95 PowerShot camera mounted on a tripod in a windowless room with controlled lighting.
Images were labeled in accordance with FAA best practices by a veteran aviation electrician trained in the FAA's EWIS standards using Intel’s CVAT tool. Localization was provided in the images using bounding boxes, allowing the use of existing object detection tools. Class labels were identified based on documented faults, best practices, and the expert opinion of maintenance personnel. Fault labels included misrouted cables/wires, damaged cables/wires, disconnected plugs and disconnected jacks. Faults were intentionally introduced during data collection by misrouting, unplugging, or damaging cables/wires.