تجزیه و تحلیل حساسیت جهانی برای برآورد پارامتر خاک در تونل زنی مکانیزه
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|27163||2014||9 صفحه PDF||سفارش دهید||محاسبه نشده|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Computers and Geotechnics, Volume 56, March 2014, Pages 80–88
The present paper validates two alternative global sensitivity analysis methods, namely variance-based and elementary effect, for the purpose of detecting key subsoil parameters that influence the output of mechanized tunnel finite element simulation. In the elementary effect method, a strategy for considering the dependencies, that result from a set of constraints between different parameters, is proposed. Moreover, because the numerical implementation of variance-based sensitivity estimates, in particular, has been proven to require intensive evaluations of the system under investigation, a practical surrogate modeling technique is utilized. This technique is based on quadratic polynomial regression and represents a reliable approximation of the computationally expensive mechanized tunnel simulation. Furthermore, a convergence analysis based on Central Limit Theorem for the numerical implementation of the methods is introduced. The adopted analysis highlights model evaluations needed for the sensitivity measures to converge, as well as the uncertainty involved in these measures.
There are different methods for performing sensitivity analysis SA; these methods are categorized into two main groups, global and local sensitivity analysis methods. In the local analysis, the partial derivatives of the system response with respect to its input variables are evaluated at a given base (local) point in the input space. Therefore, the information it gives about the sensitivities is totally dependent on the point at which the partial derivatives are evaluated, and hence, it is only suitable for linear models. On the other hand, the global methods explore the input parameter space, and therefore, the information it provides is independent of the model nature. However, the computational effort that it requires is notable in comparison to the local method. A complete review of sensitivity analysis techniques can be found in . The sensitivity analysis techniques played an important role in modeling development in the last two decades. In this regard, due to the sophisticated mathematical models that have been developed to represent complex physical, engineering, social, or economical problems, the sensitivity analysis tools are of great importance for model calibration and for the determination of the key input parameters that governs the system responses. A recent review of applications of SA can be found in  and . Sensitivity analysis techniques have been also employed in geotechnical applications, where, two recent employments can be found in  and . Moreover, local sensitivity analysis for measuring the importance of subsoil parameters in parameter estimation and calibration studies of geotechnical boundary value problems has been utilized in ,  and . However, according to the knowledge of the authors, global sensitivity methods, that have proven to be more efficient than local ones, have not been utilized for parameter estimation of subsoils so far. In this paper we apply two global sensitivity analysis methods, namely Variance-based (VB) ,  and  and elementary effect (EE)  and , for the purpose of detecting the key subsoil parameters to be estimated in the process of system identification in mechanized tunneling. In this process, a computationally expensive numerical simulation of highly nonlinear nature of the geotechnical application with respect to both the physical and geometrical characteristics is carried out. Therefore, in order to make the identification process as efficient and robust as possible, it is favorable to determine the key subsoil parameters that influence the system response. By the elementary effect method we attempt to identify the most important model parameters in relatively small number of evaluation points (simulation runs) that are well distributed in the parameter space. Whereas, in variance-based method a significantly large number of model evaluations is needed for the analysis, therefore, a practical surrogate model for the mechanized tunnel numerical simulation has been utilized. The paper’s outlines are as the following: Section 2 includes the adopted methods with an extension of the elementary effect method to consider constrained model parameters. Also, it includes the metamodel used to substitute the computationally expensive numerical simulation of mechanized tunnel with a concept of convergence analysis for the numerical implementation of the introduced global sensitivity analysis methods. In Section 3 the application of the global sensitivity analysis for ranking the subsoil parameters in mechanized tunnel simulation is showed. Section 4 summarizes the results.
نتیجه گیری انگلیسی
Global sensitivity analysis, in general, has a significant role in geotechnical applications and studies due to the uncertainty embedded in the subsoil properties. In particular, it has a major importance for complex problems, such as mechanized tunneling, which require sophisticated modeling and simulation techniques. In the presented paper two global sensitivity analysis methods have been validated for estimating key model parameters for a mechanized tunnel problem. Moreover, an extension of the elementary effect method is introduced for considering inequality constraints between model parameters that reflect their real world relations. Due to the large number of evaluation points that are required for the variance-based approach, this method is convenient when a robust metamodeling technique is available. On the other hand, the elementary effect method is able to rank the importance of model parameters with considerably low number of evaluation points. Therefore, it can be conducted directly without utilizing surrogate models. In addition to that, the constrained elementary effect is a practical approach that can be utilized for similar engineering problems when the constraints between model parameters constitute physical conditions or limitations. In the context of model parameter estimation utilizing back analysis, ranking input parameters importance is essential for selecting the most sensitive parameters that are to be estimated. Subsequently, this reduces the dimension of the back analysis problem, which has a prominent impact in problems with large dimensions. Moreover, global sensitivity analysis assists the selection of a constitutive model with a certain degree of complexity for the boundary value problem under consideration. For these purposes, both the variance-based and the elementary effect strategies can be utilized. Whereas, for evaluating the propagation of model parameters uncertainties to the outputs of interest, the variance-based method is preferable. This is because it enables quantifying the influence of parameter uncertainty reduction, which might be achieved by further investigations, on the model output variation, and consequently, achieving a certain precision of system responses. Additionally, the variance-based method has the ability to consider different kinds of distributions, not only uniform, that represent the uncertainty of the parameter within its lower and upper bounds.