Posters

Abstract

The use of mono-wave ultrasonic techniques in bolt maintenance allows for the direct measurement of bolt tension, moving beyond assumptions based on torque. While mono-wave techniques improve traceability and influence life-extension decisions, they require precise baseline measurements of each bolt in its unloaded state for later comparison. This requirement complicates applying the technique to in-situ bolts, particularly when variables such as temperature, clamp length, tensioning method (torque wrench or hydraulic jack), and flange gaps affect the measurements. This work addresses these challenges by implementing a bi-wave ultrasonic approach, eliminating the need for one-to-one baseline measurements. Bi-wave ultrasonic measurements allow data collection from a subset of bolts within a batch, from which a machine learning (ML) pipeline is built and deployed to estimate the tension in the remaining bolts with high accuracy. The pipeline, trained in a lab, can be deployed in the field on previously untested in-situ bolts. Additionally, this work developed and tested a sophisticated pipeline combining signal processing techniques and ML algorithms to estimate bolt tension while accounting for influential variables. The pipeline is designed to understand the relationships between these variables and bolt tension. This understanding is used to automatically detect outliers -- whether caused by measurement errors or environmental factors -- that could result in erroneous tension measurements, thereby removing the need for user interpretation of the signals.