Research on surrogate model of dam numerical simulation with multiple outputs based on adaptive sampling

Area of Science:

• Civil Engineering
• Computational Mechanics
• Numerical Simulation

Background:

• Dam numerical simulation is crucial for understanding structural behavior but computationally intensive for tasks like parameter back analysis.
• Surrogate models reduce computational cost, but optimal Design of Experiments (DoE) for multi-output problems remains a challenge.
• Existing DoE methods often focus on single-output problems, necessitating new approaches for high-dimensional outputs.

Purpose of the Study:

• To propose a novel sequential surrogate modeling approach for dam numerical simulations.
• To address the limitations in optimal Design of Experiments (DoE) for multi-output scenarios.
• To reduce the computational and sampling costs associated with complex dam simulations.

Main Methods:

• Development of a sequential surrogate model utilizing the radial basis function model (RBFM).
• Implementation of a multi-outputs adaptive sampling strategy for efficient data acquisition.
• Validation using benchmark functions for single- and multi-input/output scenarios, followed by application to dam simulations.

Main Results:

• The proposed method demonstrates applicability to both single- and multi-input/output problems.
• Adaptive sampling effectively targets data points based on the surrogate model's functional form.
• Significant reduction in required sampling and computational cost for multi-output dam simulations was achieved.

Conclusions:

• The developed sequential surrogate model with adaptive sampling offers an efficient solution for dam numerical simulations.
• This approach overcomes existing DoE challenges for high-dimensional outputs in engineering simulations.
• The method provides substantial savings in computational resources and time for complex structural analyses.