Presentations

Abstract

The implementation of mono-wave ultrasonic techniques to bolt maintenance programs has enabled the direct measurement of the tension in bolts rather than an assumed value from a converted applied torque. Further enhancing digital traceability that could impact life extension decisions and maintenance schedules. Mono-wave techniques, however, suffer from the need for precise baseline measurements of each bolt, when it is not under tension, that are later referenced against to calculate bolt elongation and tension. Such one-to-one measurements make applying techniques to in-situ bolts difficult. The complex nature of the variables that impact these measurements, such as temperature, clamp length, tensioning method (torque wrench or hydraulic jack), and the presence of flange gaps provides further complications. The presented work looks to resolve these issues in several ways. First, implementing a bi-wave ultrasonic approach eliminates the need for one-to-one measurements. Bi-wave ultrasonic measurements allow for data to be collected on a sub-set of bolts within a batch, a model to be built from data collected on that subset, and deployment of that model on the remaining bolts of the batch with high levels of accuracy. A model can be trained in a lab and deployed to the field on previously untested in-situ bolts. Second, this work implements machine learning algorithms to understand the relationships between the variables impacting ultrasonic measurements and bolt tension. This understanding is then utilized to detect when incoming data is an "outlier" that would cause erroneous measurements and prevent a tension measurement from being made. Eliminating the need for signal interpretation by a user. A combination of machine learning and deep learning models are then trained to directly estimate bolt tension while accounting for the impactful variables.