Engineering bolt loosening response monitoring based on advanced machine learning
TL;DR
This paper proposes a quantitative bolt loosening prediction method based on XGBoost regression, using 1000 vibration signal samples from an engineering case, which effectively quantifies bolt loosening and partially solves the problem of untimely detection in engineering.
Abstract
Bolted connections are critical to structural safety in civil and mechanical engineering, but traditional loosening detection methods are inefficient, manual-dependent and lack real-time monitoring. This paper proposes a quantitative bolt loosening prediction method based on XGBoost regression, using 1000 vibration signal samples (30-dimensional features each) from an engineering case. Loosening degree is defined as the ratio of residual to initial pre-tightening force (0.1336–1.0, lower values indicating severer loosening). Compared with Random Forest, XGBoost achieves 0.9481 R², 0.0382 MAE and 0.0470 RMSE on the test set, with 1.59% higher R², 13.0% lower MAE and 12.6% lower RMSE, showing better generalization. This method effectively quantifies bolt loosening and partially solves the problem of untimely detection in engineering.