Learning Multi-Pattern Vibration Representations for In-the-Wild Bolt Loosening Detection on Power Transmission Towers

 We provide an effective solution to a challenging andpractically important task: bolt-loosening detection onin-service transmission towers under in-the-wild, multi\u0002pattern vibration conditions, where tower types, materials,and noise environments vary substantially.To support this task, we construct, to the best of ourknowledge, the first large-scale labeled multi-pattern vi\u0002bration dataset, comprising 46,696 hammer-excited seg\u0002ments from 346 towers with diverse structures, materials,and environmental conditions.We develop the first baseline tailored to this setting,combining a hybrid learnable filterbank, a learnable STFTencoder, and an attention-based classifier, providing amore suitable solution for multi-pattern tower vibrationclassification than conventional pipelines.