Tapping and listening: a new approach to bolt looseness monitoring

Abstract

Bolted joints are among the most common building blocks used across different types of structures, and are often the key components that sew all other structural parts together. Monitoring and assessment of looseness in bolted structures is one of the most attractive topics in mechanical, aerospace, and civil engineering. This paper presents a new percussion-based non-destructive approach to determine the health condition of bolted joints with the help of machine learning. The proposed method is very similar to the percussive diagnostic techniques used in clinical examinations to diagnose the health of patients. Due to the different interfacial properties among the bolts, nuts and the host structure, bolted joints can generate unique sounds when it is excited by impacts, such as from tapping. Power spectrum density, as a signal feature, was used to recognize and classify recorded tapping data. A machine learning model using the decision tree method was employed to identify the bolt looseness level. Experiments demonstrated that the newly proposed method for bolt looseness detection is very easy to implement by ‘listening to tapping’ and the monitoring accuracy is very high. With the rapid in robotics, the proposed approach has great potential to be implemented with intimately weaving robotics and machine learning to produce a cyber-physical system that can automatically inspect and determine the health of a structure.