Reliability-based design optimization with progressive surrogate models

Reliability-based design optimization (RBDO) has traditionally been solved as a nested (bilevel) optimization problem, which is a computationally expensive approach. Unilevel and decoupled approaches for solving the RBDO problem have also been suggested in the past to improve the computational efficiency. However, these approaches also require a large number of response evaluations during optimization. To alleviate the computational burden, surrogate models have been used for reliability evaluation. These approaches involve construction of surrogate models for the reliability computation at each point visited by the optimizer in the design variable space. In this article, a novel approach to solving the RBDO problem is proposed based on a progressive sensitivity surrogate model. The sensitivity surrogate models are built in the design variable space outside the optimization loop using the kriging method or the moving least squares (MLS) method based on sample points generated from low-discrepancy sampling (LDS) to estimate the most probable point of failure (MPP). During the iterative deterministic optimization, the MPP is estimated from the surrogate model for each design point visited by the optimizer. The surrogate sensitivity model is also progressively updated for each new iteration of deterministic optimization by adding new points and their responses. Four example problems are presented showing the relative merits of the kriging and MLS approaches and the overall accuracy and improved efficiency of the proposed approach.     

Time-variant reliability global sensitivity analysis with single-loop Kriging model combined with importance sampling

This paper develops a novel failure probability-based global sensitivity index by introducing the Bayes formula into the moment-independent global sensitivity index to approximate the effect of input random variables or stochastic processes on the time-variant reliability. The proposed global sensitivity index can estimate the effect of uncertain inputs on the time-variant reliability by comparing the difference between the unconditional probability density function of input variables and the conditional probability density function in failure state of input variables. Furthermore, a single-loop active learning Kriging method combined with metamodel-based importance sampling is employed to improve the computational efficiency. The accuracy of the results obtained by Kriging model is verified by the reference results provided by the Monte Carlo simulation. Four examples are investigated to demonstrate the significance of the proposed failure probability-based global sensitivity index and the effectiveness of the computational method.     

A global sensitivity index based on Fréchet derivative and its efficient numerical analysis

Sensitivity analysis plays an important role in reliability evaluation, structural optimization and structural design, etc. The local sensitivity, i.e., the partial derivative of the quantity of interest in terms of parameters or basic variables, is inadequate when the basic variables are random in nature. Therefore, global sensitivity such as the Sobol’ indices based on the decomposition of variance and the moment-independent importance measure, among others, have been extensively studied. However, these indices are usually computationally expensive, and the information provided by them has some limitations for decision making. Specifically, all these indices are positive, and therefore they cannot reveal whether the effects of a basic variable on the quantity of interest are positive or adverse. In the present paper, a novel global sensitivity index is proposed when randomness is involved in structural parameters. Specifically, a functional perspective is firstly advocated, where the probability density function (PDF) of the output quantity of interest is regarded as the output of an operator on the PDF of the source basic random variables. The Fréchet derivative is then naturally taken as a measure for the global sensitivity. In some sense such functional perspective provides a unified perspective on the concepts of global sensitivity and local sensitivity. In the case the change of the PDF of a basic random variable is due to the change of parameters of the PDF of the basic random variable, the computation of the Fréchet-derivative-based global sensitivity index can be implemented with high efficiency by incorporating the probability density evolution method (PDEM) and change of probability measure (COM). The numerical algorithms are elaborated. Several examples are illustrated, demonstrating the effectiveness of the proposed method.     

Data-driven reliability assessment with scarce samples considering multidimensional dependence

This study proposes a data-driven method for assessing reliability, based on the scarce input dataset with multidimensional correlation. Since considering the distribution parameters estimated from the scarce dataset as those of the population may lead to epistemic uncertainty, the bootstrap resampling algorithm is adopted to infer the distribution parameters as interval parameters. To account for the variable dependence, vine copula theory is utilized to construct the joint probability density function (PDF) of input variables, and maximum likelihood estimation (MLE) and Akaike information criterion (AIC) analysis are employed to select optimal copulas based on the samples for the vine structure. Subsequently, the failure probability bounds of a response function are calculated based on the constructed joint PDF with interval distribution parameters by the active learning Kriging (AK) method combining the sparse grid integration (SGI) method. Finally, several examples are provided to demonstrate the feasibility and efficiency of the proposed method.     

Robustification of noise radar detection

A study of robust detection scheme for continuous wave noise radars is presented. The probability density function (PDF) of the noise at the input of the radar is not usually Gaussian and has heavy tails generated by impulse interferences. Although the PDF of interferences at the output of the noise radar correlator is Gaussian, impulse interferences increase the processing floor, and thus decrease the overall radar sensitivity. The proposed robustification applied to the correlator?s input signal increases the radar sensitivity in the presence of impulse interferences, and does not introduce any significant losses if the input noise is purely Gaussian.     

结构影响线识别：反问题可识别性分析与降维贝叶斯不确定性量化

结构影响线识别是移动荷载下既有结构评估的理论基础,其本质上是基于系统输入-输出含噪数据反向对静力系统指定截面的响应函数进行识别。已有研究虽然取得了进展,但它们在以下两个方面存在局限性：缺乏反问题可识别性分析;缺乏不确定性量化。反问题可识别性分析是为了厘清系统识别的参数的解的情况。不确定性量化是基于测量输入-输出含噪数据估计影响线参数的后验概率密度函数。针对上述两个局限性,该文在贝叶斯概率框架的基础上开展关于影响线识别的反问题可识别性分析与贝叶斯不确定性量化。该文进行基于直接参数化的影响线识别,包括系统输入与输出、反问题可识别性分析、参数最优值。经分析得出：一方面,直接参数化无法保证全局模型可识别;另一方面,现有方法即使是全局模型可识别的情况下也无法进行不确定性量化。为保证反问题是全局模型可识别且同时获取参数后验概率密度函数,该文提出基于降维贝叶斯不确定性量化的影响线后验识别,包括系统输入与输出重构、反问题可识别性分析、后验概率密度函数。该文进行模拟数据下新光大桥吊杆拉力影响线识别,与实测及模拟数据下简支梁桥应变影响线识别,验证提出方法的有效性。     

The generalized EPC method for the non-stationary probabilistic response of nonlinear dynamical system

It is well known that the non-stationary response probability density function (PDF) plays an important role in the reliability and failure analysis. With current approximate or numerical methods, the non-stationary approximate PDF is generally obtained in terms of the ones at the discrete time instants. Repeated computation must be conducted for the ones at other time instants since the response PDF at only one time instant can be obtained after performing the solution procedure. Thus, the computational efficiency suffers a major setback. In this paper, the exponential polynomial closure (EPC) method is further improved and enhanced to obtain the completely non-stationary PDF solution, which is distributed continuously in the time domain. It takes the temporal base function into the PDF approximation. The unknown coefficients in the EPC solution are generalized to be explicit functions of the time parameter. With the least squares method, the explicit time functions can be determined based on the few simulated response PDF values. This implementation makes the continues PDF distribution in the time domain available, which greatly improves the computational efficiency without the repeating computation. Two typical nonlinear systems under stationary and non-stationary random excitations are taken as examples to illustrate the efficiency of the proposed method. Numerical results show that the results obtained by the proposed method agree well with the simulated results. In addition, the relationship between the explicit time function and the modulate function is discussed.     

Implementation and evaluation of nonparametric regression procedures for sensitivity analysis of computationally demanding models

The analysis of many physical and engineering problems involves running complex computational models (simulation models, computer codes). With problems of this type, it is important to understand the relationships between the input variables (whose values are often imprecisely known) and the output. The goal of sensitivity analysis (SA) is to study this relationship and identify the most significant factors or variables affecting the results of the model. In this presentation, an improvement on existing methods for SA of complex computer models is described for use when the model is too computationally expensive for a standard Monte-Carlo analysis. In these situations, a meta-model or surrogate model can be used to estimate the necessary sensitivity index for each input. A sensitivity index is a measure of the variance in the response that is due to the uncertainty in an input. Most existing approaches to this problem either do not work well with a large number of input variables and/or they ignore the error involved in estimating a sensitivity index. Here, a new approach to sensitivity index estimation using meta-models and bootstrap confidence intervals is described that provides solutions to these drawbacks. Further, an efficient yet effective approach to incorporate this methodology into an actual SA is presented. Several simulated and real examples illustrate the utility of this approach. This framework can be extended to uncertainty analysis as well.     

模糊分布参数的全局灵敏度分析新方法

为合理度量随机输入变量分布参数的模糊性对输出性能统计特征的影响,提出了模糊分布参数的全局灵敏度效应指标,并研究了指标的高效求解方法。首先,分析了不确定性从模糊分布参数至模型输出响应统计特征的传递机理,以输出性能期望响应为例,利用输出均值的无条件隶属函数与给定模糊分布参数取值条件下的隶属函数的平均差异来度量模糊分布参数的影响,建立了模糊分布参数的全局灵敏度效应指标。其次,为减少所提指标的计算成本、提高计算效率,采用了扩展蒙特卡罗模拟法(EMCS)来估算输入变量分布参数与模型输出响应统计特征的函数关系。最后通过对算例的计算,验证该文所提方法的准确性和高效性。     

Evaluation of neural network models with generalized sensitivity analysis

A sensitivity analysis method for discovering characteristic features of the input data using neural network classification models has been devised. The sensitivity is the gradient of the neural network model response function, and because neural network models are nonlinear, the gradient depends on the point where it is evaluated. Two criteria are used for measuring the sensitivity. The first criterion calculates the sensitivity or gradient of the neural network output with respect to the average of the objects that comprise each class. The second criterion measures the average sensitivity of the class objects. The sensitivity analysis was applied to temperature-constrained cascade correlation network models and evaluated with sets of synthetic data and experimental mobility spectra. The neural network models were built using temperature-constrained cascade correlation networks (TCCCNs). A weight constraint was devised for the output units of the network models. This method implements weight decay with conjugate gradient training and yields more sensitive neural network models. Temperature-constrained hidden units furnish more sensitive network models than networks without constraints. By comparing the sensitivities of the class mean input and the mean sensitivity for all the inputs of a class, the individual input variables may be assessed for linearity. If these two sensitivities for an input variable differ by a constant factor, then that variable is modeled by a simple linear relationship. If the two sensitivities vary by a nonconstant scale factor, then the variable is modeled by higher order functions in the network. The sensitivity method was used to diagnose errors in the training data, and the test for linearity indicated a TCCCN architecture that had better predictability.     

Relative contributions of aleatory and epistemic uncertainty sources in time series prediction

This paper develops a novel computational framework to compute the Sobol indices that quantify the relative contributions of various uncertainty sources towards the system response prediction uncertainty. In the presence of both aleatory and epistemic uncertainty, two challenges are addressed in this paper for the model-based computation of the Sobol indices: due to data uncertainty, input distributions are not precisely known; and due to model uncertainty, the model output is uncertain even for a fixed realization of the input. An auxiliary variable method based on the probability integral transform is introduced to distinguish and represent each uncertainty source explicitly, whether aleatory or epistemic. The auxiliary variables facilitate building a deterministic relationship between the uncertainty sources and the output, which is needed in the Sobol indices computation. The proposed framework is developed for two types of model inputs: random variable input and time series input. A Bayesian autoregressive moving average (ARMA) approach is chosen to model the time series input due to its capability to represent both natural variability and epistemic uncertainty due to limited data. A novel controlled-seed computational technique based on pseudo-random number generation is proposed to efficiently represent the natural variability in the time series input. This controlled-seed method significantly accelerates the Sobol indices computation under time series input, and makes it computationally affordable.     

基于时变NARMA模型的非线性时变系统辨识

在非线性自回归滑动平均模型NARMA(NonlinearAutoRegressiveMovingAverage)中引入时间变量,将其扩展为时变NARMA模型,用Taylor展开将模型中的非线性函数展开为关于输入输出的多项式,得到关于参数线性时变的多项式形式的时变NARMA模型,再用基序列拟合模型的时变参数得到关于参数线性时不变的模型,最后用递推最小二乘法估计模型参数。仿真算例证明,与小波网络方法相比,辨识精度高,计算量小。     

Research on fatigue reliability prediction model and structural improvement of welded drive axle housing based on master S-N Curve method

Fatigue reliability prediction of welded structures is mainly based on nominal stress or hot spot stress method, but there are some problems such as grid sensitivity and joint geometry dependence. The Master S-N curve method can solve these problems well, but the corresponding reliability model needs to be studied. In this paper, the fatigue reliability model of welded structures based on the Master S-N curve method is studied. Considering the randomness of life and the correlation of failure, a reliability model is proposed, which reduces the computational burden by establishing a median damage-random threshold rule. Taking the welded drive axle housing as an object, the system reliability is analyzed under the bench test condition, and verified by the experimental data. After the verification, this method is used to predict the reliability of the axle housing under variable amplitude loading collected in the test field, and the results are verified by Monte Carlo (MC) method. When the P-S-N curves are parallel, the model is accurate, which is the characteristic of the Master S-N curve method. This method only needs to input the median damage value of the weak part, which is easy to be applied. This method can speed up the reliability prediction cycle of welded structures, which is beneficial to product innovation and optimal design. Finally, an improved design scheme is proposed for the weak parts of welding, and the effects of welding leg width, welding depth, and closed weld on fatigue life are revealed.     

Dependence of the dose estimate on the time pattern of intake by the example of tritiated water intakes

The uncertainties related to activity measurement and time pattern of intake in routine monitoring of internal exposure are considered through the example of tritiated water intakes. For this purpose, a combination of intake-to-bioassay and bioassay-to-intake calculations with Monte Carlo integration technique is introduced as a method of investigation. The time pattern of intake and the measured activity are defined as random input quantities. The probability density functions (PDFs) of the input quantities are defined and a Monte Carlo integration is performed to obtain the PDF of the output quantity which is either the value of intake estimated from a measured value of activity or the estimated activity from a given value of intake. Different possible estimates of the intake are considered: some represent the parameters of the PDF of the output quantity, others are derived from the commonly used constant chronic, I(CC), and mid-point, I(1/2), methods. The combinations of activity and intake estimates that would provide a stable estimate of the initial intake in intake-to-bioassay and bioassay-to-intake calculations were studied. Several intake estimates satisfying this requirement can be chosen depending on the task to be solved by adjusting the proper activity estimate.     

基于DEA的评价模型及其输入输出指标确定

产品包装方案的评价是一个多目标规划问题.应用数据包络分析原理(DEA),通过对包装设计方案的横向比较,构建产品包装方案的评价模型.采用两步法对评价模型求解,首先对输入变量进行比例收缩(或对输出变量进行比例扩张),求出各方案的相对效率;然后计算各方案每项指标的松弛变量.结合包装行业的特点,将越小越好的指标作为输入指标、越大越好的指标作为输出指标,建立了相应的输入输出指标体系.     

基于失效概率积分和马尔可夫链模拟的可靠性灵敏度分析

可靠性灵敏度可以被表达为失效概率对基本随机变量分布参数的偏导数的形式,利用失效概率为基本变量的联合概率密度函数在失效域上的积分表达式,并且利用马尔可夫链能够高效模拟感兴趣区域样本的性质,一种针对单个失效模式和系统多个失效模式的可靠性灵敏度分析方法被提出。由于可靠性参数灵敏度可以表达为一个与联合概率密度函数相关的函数在失效域中的数学期望的形式,所提方法采用马尔可夫链来高效模拟失效域中的样本,进而采用样本均值替代总体均值的方法来得到可靠性灵敏度的估计值。与已有的基于Monte-Carlo模拟的可靠性灵敏度分析方法相比,所提方法在保证计算精度的基础上计算效率有显著提高,尤其是针对小失效概率的可靠性灵敏度分析问题。该算例充分说明了所提方法的合理可行性。     

A new computational method of a moment-independent uncertainty importance measure

For a risk assessment model, the uncertainty in input parameters is propagated through the model and leads to the uncertainty in the model output. The study of how the uncertainty in the output of a model can be apportioned to the uncertainty in the model inputs is the job of sensitivity analysis. Saltelli [Sensitivity analysis for importance assessment. Risk Analysis 2002;22(3):579-90] pointed out that a good sensitivity indicator should be global, quantitative and model free. Borgonovo [A new uncertainty importance measure. Reliability Engineering and System Safety 2007;92(6):771-84] further extended these three requirements by adding the fourth feature, moment-independence, and proposed a new sensitivity measure, δ_i. It evaluates the influence of the input uncertainty on the entire output distribution without reference to any specific moment of the model output. In this paper, a new computational method of δ_i is proposed. It is conceptually simple and easier to implement. The feasibility of this new method is proved by applying it to two examples.     

A hybrid finite element method for near bodies of revolution (EM scattering)

A genetic formulation for a hybrid finite element solution for three-dimensional electromagnetic scattering is given using the equivalent current approach. The major computational tasks involved in monostatic scattering calculations are analyzed and compared as a function of the method of implementing the near-field radiation condition, i.e. method of moments, model expansion, and body of revolution (BOR). A method utilizing a BOR formulation that addresses these computational issues is given. This BOR implementation utilizes Hermite cubic basis functions and a variable number of modes per basis function in order to achieve the greatest efficiency. The combined field integral equation formulation is used to eliminate nonphysical resonance of the mesh boundary. Examples are given showing the efficiency and accuracy of this BOR code by itself, and as part of this hybrid finite-element method.<>     

Robustness analysis of radial base function and multi-layered feed-forward neural network models

In this paper, two popular types of neural network models (radial base function (RBF) and multi-layered feed-forward (MLF) networks) trained by the generalized delta rule, are tested on their robustness to random errors in input space. A method is proposed to estimate the sensitivity of network outputs to the amplitude of random errors in the input space, sampled from known normal distributions. An additional parameter can be extracted to give a general indication about the bias on the network predictions. The modelling performances of MLF and RBF neural networks have been tested on a variety of simulated function approximation problems. Since the results of the proposed validation method strongly depend on the configuration of the networks and the data used, little can be said about robustness as an intrinsic quality of the neural network model. However, given a data set where ‘pure’ errors from input and output space are specified, the method can be applied to select a neural network model which optimally approximates the nonlinear relations between objects in input and output space. The proposed method has been applied to a nonlinear modelling problem from industrial chemical practice. Since MLF and RBF networks are based on different concepts from biological neural processes, a brief theoretical introduction is given.     

A PDF discretization scheme in wavenumber–frequency joint spectrum for simulating multivariate random fluctuating wind fields

The wavenumber–frequency joint spectrum-based spectral representation method is an effective method for simulating multivariate random fluctuating wind fields. However, the simulation efficiency may not be quite satisfactory when the uniform discretizations in both frequency and wavenumber domains are employed. In this paper, a probability density function discretization scheme in wavenumber domain, which is actually an uneven discretization scheme, is proposed to alleviate the computational burden. In the proposed scheme, the wavenumber–frequency joint spectrum at a characteristic frequency is first taken as a probability density function. Then, the wavenumber domain is unevenly discretized according to the probability density function. Since the wavenumber–frequency joint spectrum may decay very fast in wavenumber domain, a uniform discretization is further applied in large intervals in wavenumber domain. Finally, the unevenly discrete wavenumbers and wavenumber intervals are substituted into the wavenumber–frequency joint spectrum-based spectral representation method for generating samples of multivariate random fluctuating wind fields, where the frequency domain is still uniformly discretized. The accuracy and efficiency of the proposed discretization scheme are compared with the uniform and conventional uneven discretization schemes. Practical applications of the proposed method are also investigated in various engineering scenarios.