The inspection of aerospace components coated with HVOF tungsten carbide is currently performed manually at Collins Aerospace. These cylindrical rods, used in landing gear and actuators, must be free of defects such as scratches, cracks, or porosity, since any flaw can compromise mechanical strength. However, manual inspection is slow, costly, operator-dependent, and lacks digital traceability, making it unsustainable as demand and certification requirements increase.
The AVIWC project, developed by Alta Scuola Politecnica with Collins Aerospace, explored automated alternatives. A dataset of 6,000 microscope images was used to train a Convolutional Neural Network (ResNet-18), achieving over 97% precision in detecting scratches and 99% accuracy on defect-free rods. Limited labeled data, however, prevented reliable detection of cracks and porosity. In addition, a site visit revealed that manual rod rotation is a key bottleneck.
The team therefore designed a low-cost mechatronic prototype (≈€300) using a stepper motor and Arduino, enabling smooth and repeatable rotation. This allows to reduce operator workload, standardize image capture, and create the basis for scalable AI inspection.
The project concludes that a hybrid approach—mechanical automation, AI, and human oversight—delivers significant benefits. It improves efficiency, reduces costs, ensures digital traceability, and aligns with aerospace certification standards, paving the way for sustainable industrial adoption.