Active learning for structural reliability analysis with multiple limit state functions through variance-enhanced PC-Kriging surrogate models

J. Moran A.,P.G. Morato,P. Rigo

DOI: https://doi.org/10.48550/arXiv.2302.12074

2023-02-23

Abstract:Existing active strategies for training surrogate models yield accurate structural reliability estimates by aiming at design space regions in the vicinity of a specified limit state function. In many practical engineering applications, various damage conditions, e.g. repair, failure, should be probabilistically characterized, thus demanding the estimation of multiple performance functions. In this work, we investigate the capability of active learning approaches for efficiently selecting training samples under a limited computational budget while still preserving the accuracy associated with multiple surrogated limit states. Specifically, PC-Kriging-based surrogate models are actively trained considering a variance correction derived from leave-one-out cross-validation error information, whereas the sequential learning scheme relies on U-function-derived metrics. The proposed active learning approaches are tested in a highly nonlinear structural reliability setting, whereas in a more practical application, failure and repair events are stochastically predicted in the aftermath of a ship collision against an offshore wind substructure. The results show that a balanced computational budget administration can be effectively achieved by successively targeting the specified multiple limit state functions within a unified active learning scheme.

Machine Learning,Artificial Intelligence,Applications,Computation

What problem does this paper attempt to address?

This paper attempts to solve the problem of how to efficiently select training samples under a limited computational budget when multiple limit state functions need to be evaluated in structural reliability analysis. Specifically, the paper focuses on how to effectively manage computational resources while ensuring the prediction accuracy of multiple limit state functions through active learning methods. This is particularly important in practical engineering applications, for example, when evaluating the probability characteristics of a structure under different damage conditions, multiple performance functions need to be considered simultaneously.

### Main Research Contents
1. **Background Introduction**:
- Structural reliability analysis usually involves the evaluation of one or more performance functions, which describe the probability of failure or damage events that may occur in a system under specific conditions.
- Although high - fidelity engineering simulations (such as finite element analysis, computational fluid dynamics) are accurate, they are computationally expensive. Therefore, using surrogate models (such as Kriging models) to approximate system responses is an effective solution.
- Surrogate models not only provide point estimates but also provide a measure of prediction uncertainty, which is especially important for structural reliability analysis.

2. **Active Learning Method**:
- Active learning is a machine - learning technique that sequentially selects training samples to improve the accuracy of surrogate models, especially when computational resources are limited.
- The paper proposes an active learning strategy based on the PC - Kriging surrogate model, which combines variance enhancement and U - function - derived indicators to balance the selection of training points among multiple limit state functions.

3. **Method Details**:
- **PC - Kriging Surrogate Model**: It combines Gaussian processes and polynomial chaos expansion and can provide an estimate of the mathematical relationship of the system and its uncertainty.
- **Variance Enhancement**: The prediction variance is corrected by leave - one - out cross - validation error information to improve the accuracy of training point selection.
- **U - Function**: It is used to evaluate candidate points in the design space, especially those close to the limit state boundary area, to achieve a balance between exploration and exploitation.

4. **Experimental Verification**:
- The paper verifies the effectiveness of the proposed method through two numerical experiments:
- **Analytical Reliability Problem**: Two limit state functions \( g_1 \) and \( g_2 \) are defined, and the performance of different active learning strategies is compared.
- **Offshore Wind Power Structure Collision Accident**: The repair and failure probabilities of offshore wind power structures after ship collisions are evaluated, and the active learning strategy in the multi - limit - state - function setting is verified.

### Conclusion
The active learning strategy proposed in the paper, which combines variance enhancement and the U - function, can effectively manage the selection of training samples under a limited computational budget, thereby achieving balanced prediction accuracy among multiple limit state functions. The experimental results show that this strategy performs well in both analytical and practical engineering problems and can significantly improve the efficiency and accuracy of structural reliability analysis.