Browsing by Author "Hammerand, Daniel C."

Now showing 1 - 5 of 5
  • Thumbnail Image
    Accelerating Structural Design and Optimization using Machine Learning
    Singh, Karanpreet (Virginia Tech, 2020-01-13)
    Machine learning techniques promise to greatly accelerate structural design and optimization. In this thesis, deep learning and active learning techniques are applied to different non-convex structural optimization problems. Finite Element Analysis (FEA) based standard optimization methods for aircraft panels with bio-inspired curvilinear stiffeners are computationally expensive. The main reason for employing many of these standard optimization methods is the ease of their integration with FEA. However, each optimization requires multiple computationally expensive FEA evaluations, making their use impractical at times. To accelerate optimization, the use of Deep Neural Networks (DNNs) is proposed to approximate the FEA buckling response. The results show that DNNs obtained an accuracy of 95% for evaluating the buckling load. The DNN accelerated the optimization by a factor of nearly 200. The presented work demonstrates the potential of DNN-based machine learning algorithms for accelerating the optimization of bio-inspired curvilinearly stiffened panels. But, the approach could have disadvantages for being only specific to similar structural design problems, and requiring large datasets for DNNs training. An adaptive machine learning technique called active learning is used in this thesis to accelerate the evolutionary optimization of complex structures. The active learner helps the Genetic Algorithms (GA) by predicting if the possible design is going to satisfy the required constraints or not. The approach does not need a trained surrogate model prior to the optimization. The active learner adaptively improve its own accuracy during the optimization for saving the required number of FEA evaluations. The results show that the approach has the potential to reduce the total required FEA evaluations by more than 50%. Lastly, the machine learning is used to make recommendations for modeling choices while analyzing a structure using FEA. The decisions about the selection of appropriate modeling techniques are usually based on an analyst's judgement based upon their knowledge and intuition from past experience. The machine learning-based approach provides recommendations within seconds, thus, saving significant computational resources for making accurate design choices.
  • Thumbnail Image
    Automation and Expert System Framework for Coupled Shell-Solid Finite Element Modeling of Complex Structures
    Palwankar, Manasi Prafulla (Virginia Tech, 2022-03-25)
    Finite Element (FE) analysis is a powerful numerical technique widely utilized to simulate the real-world response of complex engineering structures. With the advancements in adaptive optimization frameworks, multi-fidelity (coupled shell-solid) FE models are increasingly sought during the early design stages where a large design space is being explored. This is because multi-fidelity models have the potential to provide accurate solutions at a much lower computational cost. However, the time and effort required to create accurate and optimal multi-fidelity models with acceptable meshes for highly complex structures is still significant and is a major bottleneck in the FE modeling process. Additionally, there is a significant level of subjectivity involved in the decision-making about the multi-fidelity element topology due to a high dependence on the analyst's experience and expertise, which often leads to disagreements between analysts regarding the optimal modeling approach and heavy losses due to schedule delays. Moreover, this analyst-to-analyst variability can also result in significantly different final engineering designs. Thus, there is a greater need to accelerate the FE modeling process by automating the development of robust and adaptable multi-fidelity models as well as eliminating the subjectivity and art involved in the development of multi-fidelity models. This dissertation presents techniques and frameworks for accelerating the finite element modeling process of multi-fidelity models. A framework for the automated development of multi-fidelity models with adaptable 2-D/3-D topology using the parameterized full-fidelity and structural fidelity models is presented. Additionally, issues related to the automated meshing of highly complex assemblies is discussed and a strategic volume decomposition technique blueprint is proposed for achieving robust hexahedral meshes in complicated assembly models. A comparison of the full-solid, full-shell, and different multi-fidelity models of a highly complex stiffened thin-walled pressure vessel under external and internal tank pressure is presented. Results reveal that automation of multi-fidelity model generation in an integrated fashion including the geometry creation, meshing and post-processing can result in considerable reduction in cost and efforts. Secondly, the issue of analyst-to-analyst variability is addressed using a Decision Tree (DT) based Fuzzy Inference System (FIS) for recommending optimal 2D-3D element topology for a multi-fidelity model. Specifically, the FIS takes the structural geometry and desired accuracy as inputs (for a range of load cases) and infers the optimal 2D-3D topology distribution. Once developed, the FIS can provide real-time optimal choices along with interpretability that provides confidence to the analyst regarding the modeling choices. The proposed techniques and frameworks can be generalized to more complex problems including non-linear finite element models and as well as adaptable mesh generation schemes.
  • Thumbnail Image
    Development of Surrogate Model for FEM Error Prediction using Deep Learning
    Jain, Siddharth (Virginia Tech, 2022-07-07)
    This research is a proof-of-concept study to develop a surrogate model, using deep learning (DL), to predict solution error for a given model with a given mesh. For this research, we have taken the von Mises stress contours and have predicted two different types of error indicators contours, namely (i) von Mises error indicator (MISESERI), and (ii) energy density error indicator (ENDENERI). Error indicators are designed to identify the solution domain areas where the gradient has not been properly captured. It uses the spatial gradient distribution of the existing solution for a given mesh to estimate the error. Due to poor meshing and nature of the finite element method, these error indicators are leveraged to study and reduce errors in the finite element solution using an adaptive remeshing scheme. Adaptive re-meshing is an iterative and computationally expensive process to reduce the error computed during the post-processing step. To overcome this limitation we propose an approach to replace it using data-driven techniques. We have introduced an image processing-based surrogate model designed to solve an image-to-image regression problem using convolutional neural networks (CNN) that takes a 256 × 256 colored image of von mises stress contour and outputs the required error indicator. To train this model with good generalization performance we have developed four different geometries for each of the three case studies: (i) quarter plate with a hole, (b) simply supported plate with multiple holes, and (c) simply supported stiffened plate. The entire research is implemented in a three phase approach, phase I involves the design and development of a CNN to perform training on stress contour images with their corresponding von Mises stress values volume-averaged over the entire domain. Phase II involves developing a surrogate model to perform image-to-image regression and the final phase III involves extending the capabilities of phase II and making the surrogate model more generalized and robust. The final surrogate model used to train the global dataset of 12,000 images consists of three auto encoders, one encoder-decoder assembly, and two multi-output regression neural networks. With the error of less than 1% in the neural network training shows good memorization and generalization performance. Our final surrogate model takes 15.5 hours to train and less than a minute to predict the error indicators on testing datasets. Thus, this present study can be considered a good first step toward developing an adaptive remeshing scheme using deep neural networks.
  • Thumbnail Image
    Exploring Immersed FEM, Material Design, and Biological Tissue Material Modeling
    Kaudur, Srivatsa Bhat (Virginia Tech, 2024-03-13)
    This thesis utilizes the Immersed Interface Finite Element Method (IIFEM) for shape optimization and material design, while also investigating the modeling and parameterization of lung tissue for Diver Underwater Explosion (UNDEX) simulations. In the first part, a shape optimization scheme utilizing a four-noded rectangular immersed-interface element is presented. This method eliminates the need for interface fitted mesh or mesh morphing, reducing computational costs while maintaining solution accuracy. Analytical design sensitivity analysis is performed to obtain gradients for the optimization formulation, and various parametrization techniques are explored. The effectiveness of the approach is demonstrated through verification and case studies. For material design, the study combines topological shape optimization with IIFEM, providing a computational approach for architecting materials with desired effective properties. Numerical homogenization evaluates effective properties, and level set-based topology optimization evolves boundaries within the unit cell to generate optimal periodic microstructures. The design space is parameterized using radial basis functions, facilitating a gradient-based optimization algorithm for optimal coefficients. The method produces geometries with smooth boundaries and distinct interfaces, demonstrated through numerical examples. The thesis then delves into modeling the mechanical response of lung tissues, particularly focusing on hyperelastic and hyperviscoelastic models. The research adopts a phased approach, emphasizing hyperelastic model parametrization while reserving hyperviscoelastic model parametrization for future studies. Alternative methods are used for parametrization, circumventing direct experimental tests on biological materials. Representative material properties are sourced from literature or refit from existing experimental data, incorporating both empirically derived data and practical values suitable for simulations. Damage parameter quantification relies on asserted quantitative relationships between injury levels and the regions or percentages of affected lung tissue.
  • Thumbnail Image
    Geometrically-Linear and Nonlinear Analysis of Linear Viscoelastic Composites Using the Finite Element Method
    Hammerand, Daniel C. (Virginia Tech, 1999-08-25)
    Over the past several decades, the use of composite materials has grown considerably. Typically, fiber-reinforced polymer-matrix composites are modeled as being linear elastic. However, it is well-known that polymers are viscoelastic in nature. Furthermore, the analysis of complex structures requires a numerical approach such as the finite element method. In the present work, a triangular flat shell element for linear elastic composites is extended to model linear viscoelastic composites. Although polymers are usually modeled as being incompressible, here they are modeled as compressible. Furthermore, the macroscopic constitutive properties for fiber-reinforced composites are assumed to be known and are not determined using the matrix and fiber properties along with the fiber volume fraction. Hygrothermo-rheologically simple materials are considered for which a change in the hygrothermal environment results in a horizontal shifting of the relaxation moduli curves on a log time scale, in addition to the usual hygrothermal loads. Both the temperature and moisture are taken to be prescribed. Hence, the heat energy generated by the viscoelastic deformations is not considered. When the deformations and rotations are small under an applied load history, the usual engineering stress and strain measures can be used and the time history of a viscoelastic deformation process is determined using the original geometry of the structure. If, however, sufficiently large loads are applied, the deflections and rotations will be large leading to changes in the structural stiffness characteristics and possibly the internal loads carried throughout the structure. Hence, in such a case, nonlinear effects must be taken into account and the appropriate stress and strain measures must be used. Although a geometrically-nonlinear finite element code could always be used to compute geometrically-linear deformation processes, it is inefficient to use such a code for small deformations, due to the continual generation of the assembled internal load vector, tangent stiffness matrix, and deformation-dependent external load vectors. Rather, for small deformations, the appropriate deformation-independent stiffness matrices and load vectors to be used for all times can be determined once at the start of the analysis. Of course, the time-dependent viscoelastic effects need to be correctly taken into account in both types of analyses. The present work details both geometrically-linear and nonlinear triangular flat shell formulations for linear viscoelastic composites. The accuracy and capability of the formulations are shown through a range of numerical examples involving beams, rings, plates, and shells.