• Structural Engineering and Mechanics
• /
• v.49 no.1
• /
• pp.111-128
• /
• 2014

The stochastic response surface method (SRSM) and the response surface method (RSM) are often used for structural reliability analysis, especially for reliability problems with implicit performance functions. This paper aims to compare these two methods in terms of fitting the performance function, accuracy and efficiency in estimating probability of failure as well as statistical moments of system output response. The computational procedures of two response surface methods are briefly introduced first. Then their capabilities are demonstrated and compared in detail through two examples. The results indicate that the probability of failure mainly reflects the accuracy of the response surface function (RSF) fitting the performance function in the vicinity of the design point, while the statistical moments of system output response reflect the accuracy of the RSF fitting the performance function in the entire space. In addition, the performance function can be well fitted by the SRSM with an optimal order polynomial chaos expansion both in the entire physical and in the independent standard normal spaces. However, it can be only well fitted by the RSM in the vicinity of the design point. For reliability problems involving random variables with approximate normal distributions, such as normal, lognormal, and Gumbel Max distributions, both the probability of failure and statistical moments of system output response can be accurately estimated by the SRSM, whereas the RSM can only produce the probability of failure with a reasonable accuracy.

• Structural Engineering and Mechanics
• /
• v.75 no.3
• /
• pp.311-325
• /
• 2020

In applying the spectral stochastic finite element methods to the frequency response analysis, the conventional methods are known to give unstable and inaccurate results near the natural frequencies. To address this issue, a new sensitivity based stabilized formulation for stochastic frequency response analysis is proposed in this paper. The main difference over the conventional spectral methods is that the polynomials of random variables are applied to both numerator and denominator in approximating the harmonic response solution. In order to reflect the resonance behavior of the structure, the denominator polynomials is constructed by utilizing the natural frequency sensitivity and the random mode superposition. The numerator is approximated by applying a polynomial chaos expansion, and its coefficients are obtained through the Galerkin or the spectral projection method. Through various numerical studies, it is seen that the proposed method improves accuracy, especially in the vicinities of structural natural frequencies compared to conventional spectral methods.

• Wind and Structures
• /
• v.25 no.6
• /
• pp.589-608
• /
• 2017

A novel probabilistic approach is presented for estimating the equivalent static wind loads that produce a static response of the structure, which is "equivalent" in a probabilistic sense, to the extreme dynamic responses due to the unsteady pressure random field induced by the wind. This approach has especially been developed for complex structures (such as stadium roofs) for which the unsteady pressure field is measured in a boundary layer wind tunnel with a turbulent incident flow. The proposed method deals with the non-Gaussian nature of the unsteady pressure random field and presents a model that yields a good representation of both the quasi-static part and the dynamical part of the structural responses. The proposed approach is experimentally validated with a relatively simple application and is then applied to a stadium roof structure for which experimental measurements of unsteady pressures have been performed in boundary layer wind tunnel.

• Coupled systems mechanics
• /
• v.11 no.5
• /
• pp.411-438
• /
• 2022

In this paper, we present the parameter identification for inelastic and multi-scale problems. First, the theoretical background of several fundamental methods used in the upscaling process is reviewed. Several key definitions including random field, Bayesian theorem, Polynomial chaos expansion (PCE), and Gauss-Markov-Kalman filter are briefly summarized. An illustrative example is given to assimilate fracture energy in a simple inelastic problem with linear hardening and softening phases. Second, the parameter identification using the Gauss-Markov-Kalman filter is employed for a multi-scale problem to identify bulk and shear moduli and other material properties in a macro-scale with the data from a micro-scale as quantities of interest (QoI). The problem can also be viewed as upscaling homogenization.

• Wind and Structures
• /
• v.34 no.1
• /
• pp.127-136
• /
• 2022

In this work a multi-fidelity non-intrusive polynomial chaos (MF-NIPC) has been applied to a structural wind engineering problem in architectural design for the first time. In architectural design it is important to design structures that are safe in a range of wind directions and speeds. For this reason, the computational models used to design buildings and bridges must account for the uncertainties associated with the interaction between the structure and wind. In order to use the numerical simulations for the design, the numerical models must be validated by experi-mental data, and uncertainties contained in the experiments should also be taken into account. Uncertainty Quantifi-cation has been increasingly used for CFD simulations to consider such uncertainties. Typically, CFD simulations are computationally expensive, motivating the increased interest in multi-fidelity methods due to their ability to lev-erage limited data sets of high-fidelity data with evaluations of more computationally inexpensive models. Previous-ly, the multi-fidelity framework has been applied to CFD simulations for the purposes of optimization, rather than for the statistical assessment of candidate design. In this paper MF-NIPC method is applied to flow around a rectan-gular 5:1 cylinder, which has been thoroughly investigated for architectural design. The purpose of UQ is validation of numerical simulation results with experimental data, therefore the radius of curvature of the rectangular cylinder corners and the angle of attack are considered to be random variables, which are known to contain uncertainties when wind tunnel tests are carried out. Computational Fluid Dynamics (CFD) simulations are solved by a solver that employs the Finite Element Method (FEM) for two turbulence modeling approaches of the incompressible Navier-Stokes equations: Unsteady Reynolds Averaged Navier Stokes (URANS) and the Large Eddy simulation (LES). The results of the uncertainty analysis with CFD are compared to experimental data in terms of time-averaged pressure coefficients and bulk parameters. In addition, the accuracy and efficiency of the multi-fidelity framework is demonstrated through a comparison with the results of the high-fidelity model.

• Structural Engineering and Mechanics
• /
• v.45 no.5
• /
• pp.595-611
• /
• 2013

This paper aims to compare three collocation point methods associated with the odd order stochastic response surface method (SRSM) in a systematical and quantitative way. The SRSM with the Hermite polynomial chaos is briefly introduced first. Then, three collocation point methods, namely the point method, the root method and the without origin method underlying the odd order SRSMs are highlighted. Three examples are presented to demonstrate the accuracy and efficiency of the three methods. The results indicate that the condition that the Hermite polynomial information matrix evaluated at the collocation points has a full rank should be satisfied to yield reliability results with a sufficient accuracy. The point method and the without origin method are much more efficient than the root method, especially for the reliability problems involving a large number of random variables or requiring complex finite element analysis. The without origin method can also produce sufficiently accurate reliability results in comparison with the point and root methods. Therefore, the origin often used as a collocation point is not absolutely necessary. The odd order SRSMs with the point method and the without origin method are recommended for the reliability analysis due to their computational accuracy and efficiency. The order of SRSM has a significant influence on the results associated with the three collocation point methods. For normal random variables, the SRSM with an order equaling or exceeding the order of a performance function can produce reliability results with a sufficient accuracy. The order of SRSM should significantly exceed the order of the performance function involving strongly non-normal random variables.

• Nuclear Engineering and Technology
• /
• v.52 no.2
• /
• pp.287-295
• /
• 2020

Group method of data handling (GMDH) is considered one of the earliest deep learning methods. Deep learning gained additional interest in today's applications due to its capability to handle complex and high dimensional problems. In this study, multi-layer GMDH networks are used to perform uncertainty quantification (UQ) and sensitivity analysis (SA) of nuclear reactor simulations. GMDH is utilized as a surrogate/metamodel to replace high fidelity computer models with cheap-to-evaluate surrogate models, which facilitate UQ and SA tasks (e.g. variance decomposition, uncertainty propagation, etc.). GMDH performance is validated through two UQ applications in reactor simulations: (1) low dimensional input space (two-phase flow in a reactor channel), and (2) high dimensional space (8-group homogenized cross-sections). In both applications, GMDH networks show very good performance with small mean absolute and squared errors as well as high accuracy in capturing the target variance. GMDH is utilized afterward to perform UQ tasks such as variance decomposition through Sobol indices, and GMDH-based uncertainty propagation with large number of samples. GMDH performance is also compared to other surrogates including Gaussian processes and polynomial chaos expansions. The comparison shows that GMDH has competitive performance with the other methods for the low dimensional problem, and reliable performance for the high dimensional problem.

• Nuclear Engineering and Technology
• /
• v.53 no.3
• /
• pp.715-731
• /
• 2021

Uncertainties are calculated for pressurized water reactor (PWR) spent nuclear fuel (SNF) characteristics. The deterministic code STREAM is currently being used as an SNF analysis tool to obtain isotopic inventory, radioactivity, decay heat, neutron and gamma source strengths. The SNF analysis capability of STREAM was recently validated. However, the uncertainty analysis is yet to be conducted. To estimate the uncertainty due to nuclear data, STREAM is used to perturb nuclear cross section (XS) and resonance integral (RI) libraries produced by NJOY99. The perturbation of XS and RI involves the stochastic sampling of ENDF/B-VII.1 covariance data. To estimate the uncertainty due to modeling parameters (fuel design and irradiation history), surrogate models are built based on polynomial chaos expansion (PCE) and variance-based sensitivity indices (i.e., Sobol' indices) are employed to perform global sensitivity analysis (GSA). The calculation results indicate that uncertainty of SNF due to modeling parameters are also very important and as a result can contribute significantly to the difference of uncertainties due to nuclear data and modeling parameters. In addition, the surrogate model offers a computationally efficient approach with significantly reduced computation time, to accurately evaluate uncertainties of SNF integral characteristics.

• Structural Engineering and Mechanics
• /
• v.82 no.3
• /
• pp.271-282
• /
• 2022

The active learning surrogate model based on adaptive sampling strategy is increasingly popular in reliability analysis. However, most of the existing sampling strategies adopt the trial and error method to determine the size of the Monte Carlo (MC) candidate sample pool which satisfies the requirement of variation coefficient of failure probability. It will lead to a reduction in the calculation efficiency of reliability analysis. To avoid this defect, a new method for determining the optimal size of the MC candidate sample pool is proposed, and a new structural reliability analysis method combining polynomial chaos-based Kriging model (PC-Kriging) with adaptive sampling region is also proposed (PCK-ASR). Firstly, based on the lower limit of the confidence interval, a new method for estimating the optimal size of the MC candidate sample pool is proposed. Secondly, based on the upper limit of the confidence interval, an adaptive sampling region strategy similar to the radial centralized sampling method is developed. Then, the k-means++ clustering technique and the learning function LIF are used to complete the adaptive design of experiments (DoE). Finally, the effectiveness and accuracy of the PCK-ASR method are verified by three numerical examples and one practical engineering example.

• Proceedings of the Korea Water Resources Association Conference
• /
• 2020.06a
• /
• pp.381-381
• /
• 2020

자료동화(Data Assimilation) 기법은 실시간 수문학적 예측에 있어 정확도 향상을 위해 필수적인 과정이다. 가장 대중적으로 사용되는 기법들 중 하나가 모델 상태변수와 매개변수를 동시에 업데이트할 수 있는 이중 앙상블 칼만 필터(Dual Ensemble Kalman Filter)이다. 이 방법은 정확도 개선 및 적용의 용이성 때문에 많은 연구 분야에서 사용되어져 왔지만, 앙상블을 생성하는 과정에서 상당시간이 소요되는 단점이 존재한다. 본 연구에서는 상태변수와 매개변수를 동시에 업데이트 하면서 홍수 예측의 정확성을 보장할 뿐만 아니라, 앙상블 생성에 있어 계산 효율을 크게 향상시킬 수 있는 기법을 제안한다. Polynomial Chaos Expansion(PCE) 기법을 사용하여 앙상블 칼만 필터를 모방(mimic)할 수 있는 새로운 대체필터(Surrogate Filter)를 개발하는 것을 목표로 한다. 구체적으로 대체필터를 구성하기 위한 다양한 필터를 설계하였다. 첫째 시간에 대해서 PCE가 변화하지 않는 '불변 필터'(즉, 전체 예측기간에 대해 하나의 필터를 사용하여 자료동화할 수 있는 대체필터)와, 매 시간마다 PCE가 변화하는 '시변 필터'(즉, 예측하는 매 시간마다 새로운 필터를 생성해야 하는 대체필터)를 설계하여 적용성, 정확성, 예측성 등을 비교하였다. 또한, PCE의 하이퍼 매개변수를 최적화하기 위한 최적의 프레임 워크가 제안되어, 대체필터를 구축하는 데 효율을 높이고 PCE의 과적합(overfitting) 현상을 피할 수 있도록 하였다. 본 연구에서 제안된 기법은 기존 단일 및 이중 앙상블 칼만 필터(EnKF)의 결과와 비교 검증하였으며, 그 결과는 다음과 같다. (1) 대체필터의 대부분은 원래 EnKF와 비슷한 정도의 불확실성을 설명할 수 있음; (2) 모든 대체 필터는 선행시간이 짧은 경우의 예측에 있어 우수한 결과를 제공하며, 시변 필터가 불변 필터보다 더 정확한 예측 결과를 제공함; (3) 대체필터는 원래 앙상블 칼만필터보다 최대 500배 빠른 속도로 성능을 향상시킬 수 있음. 제안된 대체필터는 자료동화를 수행하는 기존필터와 비슷한 정도의 정확성, 매우 향상된 효율성을 보장함을 확인할 수 있었다.