AI-Based Bolt/Nut Looseness Sensing Using Spectrogram Images of Shaft Vibrations

Traditional infrastructure inspection methods require specialized expertise and involve complex maintenance techniques. Currently, inspectors depend on auditory cues, particularly the sound produced by hammering, to sense loose bolts, relying heavily on their personal experience. However, these methods might become obsolete in the future due to a shortage of skilled personnel. Therefore, AI-based sensing methods are being adopted to address this critical issue more efficiently. Yet, as of now, there is no AI-based solution specifically designed for diagnosing bolt loosening in structures that utilize bolts/nuts. In response, we propose an AI-based approach to identify both the location and the degree of loosening in bolt/nut connections by analyzing the impact sounds. Initially, we convert frequency-limited spectrogram data into a visual representation of the sound emanating from the bolt shaft when it is struck by a hammer on its surroundings. Subsequently, a novel regression detection AI sensing system is employed to determine the location and extent of the bolt loosening using these visualizations. This method can diagnose the degree of looseness in detail, which is not possible with conventional hammering inspection. Through the experiments, we confirmed the usefulness of the regression system with high sensing accuracy and the suitability of lightweight convolutional neural network (CNN)-based models, and gained insight into the dataset. These results highlight the efficiency and reliability of our proposed AI-based sensing method, demonstrating its significant potential for enhancing the safety and maintenance of infrastructural components.