A vision-based hole quality assessment technique for robotic drilling of composite materials using a hybrid classification model

Robotic drilling has advantages over traditional computer numerical control machines due to its flexibility and dexterity and the potential for rapid production and process automation. The dexterity and reach of the robotic drill end-effector enable the efficient drilling of large composite components, such as aircraft wing structures. Due to the anisotropy and inhomogeneity of fibre-reinforced polymer composite materials, drilling remains a challenging task. Inspection of the drilled hole is required at the end of the process to ensure that the final product is free from defects. Typically, such inspections require the parts to be transferred to a dedicated inspection station, which is a time-consuming non-value-added task and impractical for large components. In the interest of an efficient and sustainable manufacturing process, this work proposes a hybrid classification model implemented with a robotic drilling system to investigate the quality of drilled holes in situ. The classifier is trained and tested with a random selection of drilled holes, and the most accurate classifier is implemented. The selected classifier returns 90% overall prediction accuracy on unseen drilled holes. This machine learning–based approach, using a convolutional neural network and support vector machine classifier, can significantly improve inspection reliability while reducing production time for drilled composite components. This is the first study that demonstrates a hole quality assessment technique for robotic drilling of composite material in situ at scale.