Working Towards the Use of Artificial Intelligence to Detect the Unwanted Loosening of Bolts

The common method of bolting, used to secure parts of apparatus together, relies on the bolts having a sufficient preload force in order to ensure mechanical strength. Failing to secure bolted connections to a suitable torque rating can have costly, even dangerous consequences. Despite these checks, bolts can work loose over time through the effects of external forces such as vibration. Using traditional maintenance techniques such as visual inspection to detect a bolt slowly working loose over time can be problematic. For this reason, we have begun researching the use of video-based artificial intelligence (AI) techniques to identify this problem.

Machine learning-based object recognition algorithms require training data in the form of digital images in which the object of interest is highlighted, or “annotated” by a person. It is this annotated data from which the algorithm learns the features of the object. Once trained, the model can then be used to detect the object in previously unseen images. However, accurate training data is expensive and time consuming to produce and, therefore, little is publicly available. 

In order to work towards an AI system that might be able to detect changes in the rotational angle of a bolt over time, we methodically compiled a comprehensive dataset of over 1,100 images. These samples depict bolts at various rotational angles, photographed from different perspectives and focal lengths. In the absence of any similar publicly available dataset, these samples will provide a solid basis for further experimental work.