An image dataset for use in detecting unwanted bolt rotation.

In any industrial system, ensuring that the engineered components therein are in working order is essential for the safety of workers and for efficient and cost-effective running. However, due to factors such as stress, deformation, and corrosion, individual components degrade over time, eventually leading to failure. Whilst there exist several public training datasets for use in bolt detection, there is none in the area of bolts or other mechanical fixings changing over time. We prepared a novel dataset of over 1,100 images depicting a bolted apparatus from different angles, and with varying degrees of bolt rotation. The images were taken in laboratory conditions, with carefully measured variations. As far as we know, no other such dataset exists.