Research on cross-domain bolt looseness detection of cylindrical shell structures based on a distorted similitude model

The distorted similitude model, which retains certain dimensions while scaling down others, has attracted attention in structural health monitoring due to its reduced size, ease of condition modification, and advantages in data collection. However, a major challenge arises in detection of bolt loosening states of a prototype using vibration data from the distorted similitude model, as the data distributions between the model and the prototype differ substantially. To address this challenge, a cross-domain bolt looseness detection method based on distorted scaling laws and domain adaptation networks is proposed. In the case of minimizing distortion to twice the axial length of the bolted cylindrical shell, bolt looseness detection using data from the distorted similitude model achieves only 20.8% accuracy. In contrast, the proposed method achieves 100% accuracy. For five bolt loosening scenarios, the average maximum mean discrepancy (MMD) between the All-Pole group delay coefficients of the distorted similitude model and the prototype is 72.6. The established scaling laws reduce this discrepancy to 47.5, demonstrating their effectiveness in aligning data distributions. Both domain discriminator loss and MMD loss contribute to overcoming inter-domain distribution differences. The combined effect of the scaling laws and domain adaptation network ensures that bolt loosening features from the distorted similitude model and the prototype are consistently aligned in the data space. This study reports the first successful demonstration of vibration-based detection for prototype bolt loosening using a distorted similitude model, thereby establishing a novel paradigm in similitude model research.