Long-term viscoelastic deformation monitoring of a concrete dam: A multi-output surrogate model approach for parameter identification

• Published in: Engineering structures 2022-09, Vol.266, p.114553, Article 114553
• Main Authors: Lin, Chaoning, Li, Tongchun, Chen, Siyu, Yuan, Li, van Gelder, P.H.A.J.M., Yorke-Smith, Neil
• Format: Article
• Language: English
• Subjects: Computer applications; Concrete dam; Concrete dams; Dam safety; Dams; Deformation; Detention dams; Finite element method; Gaussian process; Heuristic methods; Inverse analysis; Inversion; Mathematical models; Mechanical properties; Monitoring; Multi-output Gaussian process; Optimization; Parameter identification; Surrogate model; Viscoelasticity
• ISSN: 0141-0296; 1873-7323
• DOI: 10.1016/j.engstruct.2022.114553

•A novel surrogate model-assisted inversion method is proposed for identifying viscoelastic parameters of dam system.•The sensitivity of the training sample size, parameter range and output quantity of the MOGP surrogate model is investigated.•The computational accuracy and efficiency of the proposed inversion method are proved.•A physics-based monitoring model is calibrated for long-term deformation prediction of the concrete dam. Dam safety monitoring has become an important topic and is critical for evaluating a dam’s safety status. This study focuses on identifying the mechanical properties of a concrete dam from long-term viscoelastic deformation monitoring data. A novel inversion framework is proposed in which a surrogate model, instead of the finite element model, is placed inside the optimization loop. First, a multi-output surrogate model based on Gaussian process is trained by using data from a finite element simulation in the creep regime. In order to efficiently create a high-precision and reliable surrogate model, three test instances are conducted to investigate the impact of sample size, parameter range and output quantity on the performance of the surrogate model. Subsequently, a meta-heuristic optimization, multi-verse optimizer, is employed to identify the unknown viscoelastic parameters. The results illustrate that the identified properties allow predictions on dam displacement which are consistent with the monitoring data. Compared with the traditional inversion method based on finite element modelling, the proposed inversion method based on the multi-output surrogate model not only achieves accurate estimation of mechanical parameters but also greatly improves computational efficiency.