Neural Network-Based Identification of Bolt Joint Loosening in Electric Transmission Towers: Localization and Degree Assessment

Bolt joint loosening, a prevalent issue in electric transmission towers, is primarily induced by long-term exposure to environmental loads, thereby undermining their structural integrity and stability. In tower structures, bolt joints are highly correlated, with steel members acting as connecting components to form a cohesive “joint-member-joint” system. Leveraging the dynamic characteristics of this system as inputs, this study develops a fully connected Artificial Neural Network capable of simultaneously identifying the locations and degrees of bolt joint loosening. Based on the mechanical properties of loosened joints, the approach involves forward propagation, backpropagation, and parameter optimization using a dataset generated from finite element simulations. Validation was conducted under various modal orders and loosening conditions. Results indicate that even a basic ANN model can achieve useful accuracy in predicting both the locations and degrees of loosened joints, and increasing the number of modal inputs further improves accuracy, with singlejoint loosening more accurately identified in terms of position and double-joint loosening more accurately predicted in terms of looseness degree.