Article
Engineering, Multidisciplinary
John D. Jakeman, Sam Friedman, Michael S. Eldred, Lorenzo Tamellini, Alex A. Gorodetsky, Doug Allaire
Summary: We present an adaptive algorithm for constructing surrogate models of multi-disciplinary systems composed of a set of coupled components. The algorithm introduces coupling variables with unknown distributions to independently build surrogates for each component. The surrogates are then combined to form an integrated-surrogate, which greatly reduces the cost compared to the original model. The algorithm minimizes the error in the integrated-surrogate by allocating training data based on the contribution of each component-surrogate. Extensive numerical results demonstrate the accuracy and efficiency of the algorithm compared to black-box methods.
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
(2022)
Article
Computer Science, Interdisciplinary Applications
Shuo Wang, Yin Liu, Qi Zhou, Yongliang Yuan, Liye Lv, Xueguan Song
Summary: An adaptive MFS-MLS model is proposed to combine the advantages of high-fidelity and low-fidelity simulations by giving adaptive weightings to different HF samples, achieving competitive performance with high computational efficiency.
STRUCTURAL AND MULTIDISCIPLINARY OPTIMIZATION
(2021)
Article
Engineering, Aerospace
Hongyan Bu, Liming Song, Zhendong Guo, Jun LI
Summary: This paper investigates the accuracy issue of the Maximum Likelihood Estimation (MLE) in selecting the scale factor for the Bayesian Multi-Fidelity Surrogate (MFS) model. An alternative approach is proposed to select the scale factor by minimizing the posterior variance of the discrepancy function. Experimental results demonstrate that the proposed approach achieves better accuracy and robustness compared to MLE.
CHINESE JOURNAL OF AERONAUTICS
(2022)
Article
Computer Science, Interdisciplinary Applications
Qi Zhou, Jinhong Wu, Tao Xue, Peng Jin
Summary: The paper introduces a two-stage adaptive multi-fidelity surrogate model-assisted multi-objective genetic algorithm (AMFS-MOGA), which involves obtaining a preliminary Pareto frontier using low-fidelity model data in the first stage and constructing an initial MFS model based on samples selected from the preliminary Pareto set in the second stage. The fitness values of individuals are evaluated using the MFS model, which is adaptively updated according to prediction uncertainty and population diversity. The effectiveness of the proposed approach is demonstrated through benchmark tests and design optimization, showing comparable results to traditional methods while significantly reducing computational costs.
ENGINEERING WITH COMPUTERS
(2021)
Article
Engineering, Mechanical
Kunpeng Li, Xiwang He, Liye Lv, Jiaxiang Zhu, Guangbo Hao, Haiyang Li, Xueguan Song
Summary: This paper proposes a single-fidelity surrogate model with a hierarchical structure called Nonlinearity Integrated Correlation Mapping Surrogate (NI-CMS) model, which overcomes the expensive modeling cost by establishing a low-fidelity model and correcting it using the idea of a multi-fidelity surrogate model. The performance of the proposed model is evaluated through numerical test functions and a surrogate-based digital twin experiment, demonstrating its advantages over existing models.
JOURNAL OF MECHANICAL DESIGN
(2023)
Article
Engineering, Multidisciplinary
J. M. Winter, J. W. J. Kaiser, S. Adami, I. S. Akhatov, N. A. Adams
Summary: In this work, a novel framework combining multi-fidelity Gaussian Process modeling techniques with input-space warping is proposed for the cost-efficient construction of a stochastic surrogate model. The framework achieves high computational efficiency by combining a large number of cheap estimates with a few expensive measurements. It is based on coarse-grid approximations of high-fidelity numerical simulations. The framework is applied to generate a surrogate model for crystal growth velocities in directional dendritic solidification, and its accuracy is assessed using cross-validation techniques.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2022)
Article
Computer Science, Interdisciplinary Applications
Shuai Zhang, Pengwei Liang, Yong Pang, Jianji Li, Xueguan Song
Summary: The paper proposes a novel method for determining the scale factor in multi-fidelity surrogate models, which outperforms other models in terms of prediction accuracy and robustness. It explores the impact of different cost ratios and proportions on model performance, providing a new approach to the design and optimization of engineering problems.
STRUCTURAL AND MULTIDISCIPLINARY OPTIMIZATION
(2022)
Article
Computer Science, Artificial Intelligence
Lili Zhang, Yuda Wu, Ping Jiang, Seung-Kyum Choi, Qi Zhou
Summary: The NHLF-Co-Kriging method proposed in this work is able to handle multiple non-hierarchical low-fidelity models effectively by optimizing scale factors, providing more accurate MF surrogate models under a limited computational budget.
ADVANCED ENGINEERING INFORMATICS
(2022)
Article
Energy & Fuels
Lian Wang, Yuedong Yao, Tao Zhang, Caspar Daniel Adenutsi, Guoxiang Zhao, Fengpeng Lai
Summary: This paper proposes a self-adaptive multi-fidelity surrogate-assisted multi-objective production optimization algorithm (SAMFS-MOPO) to reduce the computational burden and improve the accuracy of the surrogate model. By using two fidelity samples to establish a multi-fidelity surrogate model, and using i-updating and g-updating strategies to update the model during the optimization process, this method demonstrates superior performance in convergence, diversity, and efficiency compared to other conventional methods.
JOURNAL OF PETROLEUM SCIENCE AND ENGINEERING
(2022)
Article
Engineering, Multidisciplinary
Paolo Conti, Mengwu Guo, Andrea Manzoni, Jan S. Hesthaven
Summary: When evaluating quantities of interest dependent on differential equation solutions, there is a trade-off between accuracy and efficiency. The proposed multi-fidelity surrogate modeling approach leverages low-fidelity data to improve approximations with limited high-fidelity data, using long short-term memory (LSTM) networks for parameterized, time-dependent problems. This approach enhances output predictions for unseen parameter values and forward in time simultaneously, and outperforms feed-forward neural networks in terms of accuracy.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2023)
Article
Engineering, Aerospace
Quan Lin, Jiexiang Hu, Qi Zhou
Summary: This paper proposes two parallel multi-objective Bayesian optimization (MOBO) approaches based on multi-fidelity surrogate modeling to improve the optimization efficiency. The approaches utilize cheap auxiliary low-fidelity data for better performance. The modified hypervolume expected improvement function is used to determine the updating points and fidelity levels, and two parallel computing strategies are developed for multi-point sampling. Additionally, a constraint handling strategy is introduced for constrained problems. The proposed approaches are validated through numerical benchmark examples and real-world applications, showing significant improvements in terms of efficiency and overall performance compared to state-of-the-art MOBO methods.
AEROSPACE SCIENCE AND TECHNOLOGY
(2023)
Article
Computer Science, Artificial Intelligence
Zhili Tang, Liang Xu, Shaojun Luo
Summary: This paper proposes an adaptive dynamic surrogate-assisted evolutionary computation approach based on variable search region to improve the computational efficiency in high-dimensional search space. The method focuses on constructing a local surrogate model to find the optimal solution and achieves highly accurate results in less than five adaptive iterations. It is successfully applied to the aerodynamic shape design optimization and greatly improves the computational efficiency compared to traditional static global approximation surrogate models.
APPLIED SOFT COMPUTING
(2022)
Article
Engineering, Aerospace
Zengcong LI, Kuo Tian, Shu Zhang, Bo Wang
Summary: To accelerate multi-objective optimization for expensive engineering cases, this study presents a KE-VFS-CMA-ES model which is based on knowledge extraction and variable-fidelity surrogate-assisted covariance matrix adaptation evolution strategy. In the first part, the KE-VFS model is established by using a low-fidelity surrogate model to obtain low-fidelity non-dominated solutions, extracting knowledge using the K-means clustering algorithm and space-filling strategy, and combining high-fidelity and low-fidelity sample points. In the second part, a novel model management based on the MHVI criterion and pre-screening strategy is proposed. The results demonstrate the excellent efficiency, robustness, and applicability of KE-VFS-CMA-ES compared to other multi-objective optimization algorithms.
CHINESE JOURNAL OF AERONAUTICS
(2023)
Article
Computer Science, Interdisciplinary Applications
Lili Zhang, Seung-Kyum Choi, Tingli Xie, Ping Jiang, Jiexiang Hu, Jasuk Koo
Summary: This study constructs Kriging-based multi-fidelity surrogate models to accelerate fatigue analysis of welded joints, taking into consideration different fidelity levels of finite element models for improved accuracy. By using verification methods and uncertainty quantification metrics, the most suitable surrogate models are determined and ranked using the EW-TOPSIS technique.
STRUCTURAL AND MULTIDISCIPLINARY OPTIMIZATION
(2021)
Article
Computer Science, Interdisciplinary Applications
Huan Zhao, Zheng-Hong Gao, Lu Xia
Summary: Surrogate models are widely used in uncertainty-based design optimization, but they often have accuracy and sensitivity issues. To address these challenges, a UBDO framework based on multi-fidelity polynomial chaos-Kriging is proposed, with particular superiority for complex aerodynamic applications.
COMPUTERS & FLUIDS
(2022)