Long-term structural health monitoring of arch dams under environmental effects: bayesian neural network-based frequency estimation
Measurement: Journal of the International Measurement Confederation, cilt.292, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 292
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.measurement.2026.123318
- Dergi Adı: Measurement: Journal of the International Measurement Confederation
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO), Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Anahtar Kelimeler: Ambient vibration test, Bayesian neural network, Deriner Dam, Dynamic characteristics, Structural health monitoring
- Karadeniz Teknik Üniversitesi Adresli: Evet
Özet
This study investigates the long-term dynamic behavior of Deriner Dam through continuous vibration-based structural health monitoring (SHM) and probabilistic frequency estimation. Ambient vibration measurements were used to identify the first four natural frequencies, while reservoir water level and ambient air temperature were recorded as the principal environmental variables. A Bayesian Neural Network (BNN) was then developed to estimate modal frequencies and quantify prediction uncertainty using Monte Carlo dropout. The model was trained on a 14-month monitoring dataset and subsequently applied to historical environmental records to generate probabilistic reconstructions for periods without vibration measurements. Model performance was evaluated using RMSE, MAE, R2, Prediction Interval Coverage Probability (PICP), and Mean Prediction Interval Width (MPIW). PICP values ranged from approximately 91% to 98% across the four modes, while overall R2 values were 0.776, 0.850, 0.715, and 0.583 for Modes 1–4, respectively. A deterministic MLP baseline showed comparable performance for the lower modes but substantially poorer test-set generalization for Modes 3 and 4. The results indicate that the BNN provides useful uncertainty-aware frequency estimates within the monitored environmental domain. Historical reconstructions should be interpreted as probabilistic estimates conditioned on relationships learned during monitoring rather than as verified measurements. The proposed framework provides a practical basis for uncertainty-informed long-term dam monitoring and engineering decision support.