University of Thessaly (UTH)
Management of uncertainties in structural response and reliability simulations using measured data
Abstract
dc:descriptionThe subject of current PhD thesis is the development of methods for the modeling and management of uncertainties in structural simulations. Emphasis is given on developing the probabilistic-statistical framework that is used for (a) quantifying and calibrating uncertainty models of mechanical systems and external excitation forces, in structural dynamics based on vibration measurements, as well as (b) propagating these modeling uncertainties in structural dynamics simulations to achieve updated robust predictions of system performance, reliability and safety. The tools for identifying system and uncertainty models as well as performing robust prediction analyses are the Bayesian Inference, Laplace methods of asymptotic approximation and more accurate stochastic simulation algorithms, such as Transitional Markov Chain Monte Carlo. These tools involve solving optimization problems, generating samples for tracing and then populating the important uncertainty region in the parameter space, as well as evaluating integrals over high-dimensional spaces of the uncertain model parameters. A moderate to very large number of repeated system analyses are required to be performed over the space of uncertain parameters. Consequently, the computational demands depend highly on the number of system analyses and the time required for performing a system analysis. For such large-order finite element models the computational demands in implementing asymptotic approximations as well as stochastic simulation techniques may be excessive. This study integrates an efficient Component Mode Synthesis technique that takes into account the FE model parameterization to substantially alleviate the computational burden associated with the Bayesian methodology. Another innovative aspect of this thesis is the use of measurements in order to improve the mathematical models that simulate the mechanical system, the excitation and the uncertainties that arise, based on the Bayesian Inference. Finally, the computational efficiency of the proposed techniques is demonstrated through applications (a) in structural health monitoring, damage identification and updating structural reliability of civil infrastructure, and (b) in predicting fatigue for metallic structures through a limited number of acceleration measurements. Part A: From the beginning of the current PhD thesis it became clear that in order to perform the analysis and management of uncertainties for complex structural models such as large DOF models involving hundreds of thousands or even million DOF and models including localized nonlinearities, it was necessary the analysis to be integrated with smart techniques for reducing the time consuming analysis that appear in reliability simulations. For this reason, much effort of the research was devoted to investigate efficient techniques for reducing the number of degrees of freedom of the FE model simulating the structural system. Specifically, component mode synthesis (CMS) techniques are widely used to carry out system analyses in a significantly reduced space of generalized coordinates, thus alleviating the computational burden involved in the implementation of methods for management of uncertainties. In this work, a novel framework is presented for integrating the Craig-Bampton technique into existing FE model updating formulations in order to reduce the time consuming operations involved in reanalyses of large-order models of hundreds of thousands or even millions degrees of freedom. The proposed method exploits the fact that in FE model parameterization schemes the stiffness matrix of the structure often depends linearly on the parameters of the model and also that a parameter usually represents a global property (e.g. the modulus of elasticity) of a substructure. The division of the structure into components is then guided by the FE parameterization scheme so that the stiffness matrix that arise for each one of the introduced components to depend linearly on only one of the parameters to be estimated. The methodology proved to be very effective for the management of uncertainties in the dynamic structural analysis of large-scale models, since it led to a significant reduction both of the degrees of freedom and of the time consumed for the management of uncertainties by three and more orders of magnitude. Through the implementation of the framework to large order models of structures involving hundreds of thousands of degrees of freedom, such as the linear model of Metsovo bridge, it was demonstrated that the computational effort was reduced drastically from days to a few seconds. This drastic reduction of the computational time without compromising the accuracy allowed for first time the uncertainty quantification and propagation of large order complex structures.The above reduction techniques proved to be a very useful tool for the research conducted in the context of current thesis due to the reduction of the computational effort of reanalyses involved in stochastic optimizations and simulations algorithms implemented for performing management of uncertainties in structural response and reliability simulations the reduction of degrees of freedom is a very important tool for the preparation of this thesis as it significantly reduces the computational cost for the management of a large number of repeat analyzes required in optimization algorithms and stochastic simulation algorithms used in the management of uncertainties in finite element models. Finally the above techniques were very useful for the development of a framework for management of uncertainties that appear in structural health monitoring (SHM).Part B: In the context of this thesis, a Bayesian framework for model parameter estimation and class selection, based on vibration measurements, was developed. Specifically, the theoretical framework for the parameter estimation problem for linear models using the modal characteristics (eigenfrequencies and modeshapes) was first formulated and then implemented in software. The results of the application of the proposed Bayesian Framework, implemented in the developed software, for linear large DOF models were of high accuracy.αποτελέσματα υψηλής ακρίβειας. Moreover, different approaches of the Bayesian Framework, presented in relevant literature were implemented in software and tested for their accuracy. Another field of great interest of this thesis was the development of Bayesian tools, including asymptotic approximations and stochastic simulation algorithms, for the quantification and management of uncertainties in parameter estimation for both linear and nonlinear models. Among the stochastic simulation algorithms available Transitional MCMC is one of the most promising and for this reason it was chosen to be used. The Transitional MCMC is an algorithm of general use and can be applied to both linear and non-linear models, as it was demonstrated in this thesis through its application to a small scale experimental model of a vehicle with nonlinear wheel and suspension components review for the model updating of the parameters of the nonlinear.In this case of this nonlinear model, the Bayesian framework was extended to cover the case that the measurements are taken to be either response time histories or response spectra functions instead of the modal characteristics that were used in case of linear models. For the case that measured data is response time histories, the proposed methodologies were implemented in software, taking into account different approaches for calculating the prediction error that appears according to Bayesian formulation. The Bayesian framework for uncertainty quantification, calibration and propagation was successfully implemented in the case of the experimental vehicle that was a combination of linear and nonlinear components. Drastic reduction in computational effort to manageable levels was achieved using component mode synthesis techniques that are presented in the context of this thesis, for the linear vehicle frame obtaining a drastic reduction in the DOFs from 50,000 to less than 100. Finally, based on the measured output quantities of interest of the components of the vehicle frame, interesting results were concluded about the methodology that the estimates of the model parameter values and their uncertainties for each component can be used to build the model for the combined wheel-suspension-frame structure.Part C: A novel framework is proposed for estimating damage accumulation due to fatigue in the entire body of a metallic structure using vibration measurements from a limited number of sensors. Fatigue is estimated using Palmgren-Miner damage rule, S-N curves, rainflow cycle counting of the variable amplitude time histories of the stress components, or frequency domain stochastic fatigue methods based on PSD of the stress components. These methods can be applied to any point in the structure and construct the complete fatigue map of the entire structure, provided that the stress response characteristics (time histories or PSDs) at all desirable points are available. These stress response characteristics are predicted from limited number of vibration sensors using a high fidelity finite element model and different prediction methods, including Kalman filter type techniques, kriging approximations and modal expansion methods. The effectiveness of the proposed methods is demonstrated using simulated data from a chain-like spring-mass model and a small-scale model of a vehicle structure. The proposed framework can be used to construct fatigue damage accumulation and lifetime prediction maps consistent with the actual operational conditions provided by a monitoring system. These maps are useful for designing optimal fatigue-based maintenance strategies for metallic structures taking into account all uncertainties in modeling and fatigue predictions.
Degree
thesis:*- Grantor dc:publisher
- University of Thessaly (UTH)
- Year dc:date
- 2015
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Papadioti, Dimitra-Christina
- Παπαδιώτη, Δήμητρα-Χριστίνα
Subjects
dc:subject × 12Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/36144