Computer Science

The advanced magnetic resonance (MR) image reconstructions such as the compressed sensing and subspace-based imaging are considered as large-scale, iterative, optimization problems. Given the large number of reconstructions required by the practical clinical usage, the computation time of these advanced reconstruction methods is often unacceptable. In this work, we propose using Google's Tensor Processing Units (TPUs) to accelerate the MR image reconstruction. TPU is an application-specific integrated circuit (ASIC) for machine learning applications, which has recently been used to solve large-scale scientific computing problems. As proof-of-concept, we implement the alternating direction method of multipliers (ADMM) in TensorFlow to reconstruct images on TPUs. The reconstruction is based on multi-channel, sparsely sampled, and radial-trajectory k -space data with sparsity constraints. The forward and inverse non-uniform Fourier transform operations are formulated in terms of matrix multiplications as in the discrete Fourier transform. The sparsifying transform and its adjoint operations are formulated as convolutions. The data decomposition is applied to the measured k -space data such that the aforementioned tensor operations are localized within individual TPU cores. The data decomposition and the inter-core communication strategy are designed in accordance with the TPU interconnect network topology in order to minimize the communication time. The accuracy and the high parallel efficiency of the proposed TPU-based image reconstruction method are demonstrated through numerical examples.

A common workflow for many engineering design problems requires the evaluation of the design system to be investigated under a range of conditions. These conditions usually involve a combination of several parameters. To perform a complete evaluation of a single candidate configuration, it may be necessary to perform hundreds to thousands of simulations. This can be computationally very expensive, particularly if several configurations need to be evaluated, as in the case of the mathematical optimization of a design problem. Although the simulations are extremely complex, generally, there is a high degree of redundancy in them, as many of the cases vary only slightly from one another. This redundancy can be exploited by omitting some simulations that are uninformative, thereby reducing the number of simulations required to obtain a reasonable approximation of the complete system. The decision of which simulations are useful is made through the use of machine learning techniques, which allow us to estimate the results of "yet-to-be-performed" simulations from the ones that are already performed. In this study, we present the results of one such technique, namely active learning, to provide an approximate result of an entire offshore riser design simulation portfolio from a subset that is 80% smaller than the original one. These results are expected to facilitate a significant speed-up in the offshore riser design.

Stabilised mixed velocity-pressure formulations are one of the widely-used finite element schemes for computing the numerical solutions of laminar incompressible Navier-Stokes. In these formulations, the Newton-Raphson scheme is employed to solve the nonlinearity in the convection term. One fundamental issue with this approach is the computational cost incurred in the Newton-Raphson iterations at every load/time step. In this paper, we present an iteration-free mixed finite element formulation for incompressible Navier-Stokes that preserves second-order temporal accuracy of the generalised-alpha and related schemes for both velocity and pressure fields. First, we demonstrate the second-order temporal accuracy using numerical convergence studies for an example with a manufactured solution. Later, we assess the accuracy and the computational benefits of the proposed scheme by studying the benchmark example of flow past a fixed circular cylinder. Towards showcasing the applicability of the proposed technique in a wider context, the inf-sup stable P2-P1 pair for the formulation without stabilisation is also considered. Finally, the resulting benefits of using the proposed scheme for fluid-structure interaction problems is illustrated using two benchmark examples in fluid-flexible structure interaction.

Optimizing fluid-dynamic performance is an important engineering task. Traditionally, experts design shapes based on empirical estimations and verify them through expensive experiments. This costly process, both in terms of time and space, may only explore a limited number of shapes and lead to sub-optimal designs. In this research, a test-proven deep learning architecture is applied to predict the performance under various restrictions and search for better shapes by optimizing the learned prediction function. The major challenge is the vast amount of data points Deep Neural Network (DNN) demands, which is improvident to simulate. To remedy this drawback, a Frequentist active learning is used to explore regions of the output space that DNN predicts promising. This operation reduces the number of data samples demanded from ~8000 to 625. The final stage, a user interface, made the model capable of optimizing with given user input of minimum area and viscosity. Flood fill is used to define a boundary area function so that the optimal shape does not bypass the minimum area. Stochastic Gradient Langevin Dynamics (SGLD) is employed to make sure the ultimate shape is optimized while circumventing the required area. Jointly, shapes with extremely low drags are found explored by a practical user interface with no human domain knowledge and modest computation overhead.

In this work, we extend the recently proposed adaptive phase field method to model fracture in orthotropic functionally graded materials (FGMs). A recovery type error indicator combined with quadtree decomposition is employed for adaptive mesh refinement. The proposed approach is capable of capturing the fracture process with a localized mesh refinement that provides notable gains in computational efficiency. The implementation is validated against experimental data and other numerical experiments on orthotropic materials with different material orientations. The results reveal an increase in the stiffness and the maximum force with increasing material orientation angle. The study is then extended to the analysis of orthotropic FGMs. It is observed that, if the gradation in fracture properties is neglected, the material gradient plays a secondary role, with the fracture behaviour being dominated by the orthotropy of the material. However, when the toughness increases along the crack propagation path, a substantial gain in fracture resistance is observed.

Metamaterials exhibit materials response deviation from conventional elasticity. This phenomenon is captured by the generalized elasticity as a result of extending the theory at the expense of introducing additional parameters. These parameters are linked to internal length scales. Describing on a macroscopic level a material possessing a substructure at a microscopic length scale calls for introducing additional constitutive parameters. Therefore, in principle, an asymptotic homogenization is feasible to determine these parameters given an accurate knowledge on the substructure. Especially in additive manufacturing, known under the infill ratio, topology optimization introduces a substructure leading to higher order terms in mechanical response. Hence, weight reduction creates a metamaterial with an accurately known substructure. Herein, we develop a computational scheme using both scales for numerically identifying metamaterials parameters. As a specific example we apply it on a honeycomb substructure and discuss the infill ratio. Such a computational approach is applicable to a wide class substructures and makes use of open-source codes; we make it publicly available for a transparent scientific exchange.

With more efficient structures, last trends in aeronautics have witnessed an increased flexibility of wings, calling for adequate design and optimization approaches. To correctly model the coupled physics, aerostructural optimization has progressively become more important, being nowadays performed also considering higher-fidelity discipline methods, i.e., CFD for aerodynamics and FEM for structures. In this paper a methodology for high-fidelity gradient-based aerostructural optimization of wings, including aerodynamic and structural nonlinearities, is presented. The main key feature of the method is its modularity: each discipline solver, independently employing algorithmic differentiation for the evaluation of adjoint-based sensitivities, is interfaced at high-level by means of a wrapper to both solve the aerostructural primal problem and evaluate exact discrete gradients of the coupled problem. The implemented capability, ad-hoc created to demonstrate the methodology, and freely available within the open-source SU2 multiphysics suite, is applied to perform aerostructural optimization of aeroelastic test cases based on the ONERA M6 and NASA CRM wings. Single-point optimizations, employing Euler or RANS flow models, are carried out to find wing optimal outer mold line in terms of aerodynamic efficiency. Results remark the importance of taking into account the aerostructural coupling when performing wing shape optimization.

The current design of aerodynamic shapes, like airfoils, involves computationally intensive simulations to explore the possible design space. Usually, such design relies on the prior definition of design parameters and places restrictions on synthesizing novel shapes. In this work, we propose a data-driven shape encoding and generating method, which automatically learns representations from existing airfoils and uses the learned representations to generate new airfoils. The representations are then used in the optimization of synthesized airfoil shapes based on their aerodynamic performance. Our model is built upon VAEGAN, a neural network that combines Variational Autoencoder with Generative Adversarial Network and is trained by the gradient-based technique. Our model can (1) encode the existing airfoil into a latent vector and reconstruct the airfoil from that, (2) generate novel airfoils by randomly sampling the latent vectors and mapping the vectors to the airfoil coordinate domain, and (3) synthesize airfoils with desired aerodynamic properties by optimizing learned features via a genetic algorithm. Our experiments show that the learned features encode shape information thoroughly and comprehensively without predefined design parameters. By interpolating/extrapolating feature vectors or sampling from Gaussian noises, the model can automatically synthesize novel airfoil shapes, some of which possess competitive or even better aerodynamic properties comparing with training airfoils. By optimizing shape on learned features via a genetic algorithm, synthesized airfoils can evolve to have specific aerodynamic properties, which can guide designing aerodynamic products effectively and efficiently.

Level set-based immersed boundary techniques operate on nonconforming meshes while providing a crisp definition of interface and external boundaries. In such techniques, an isocontour of a level set field interpolated from nodal level set values defines a problem's geometry. If the interface is explicitly tracked, the intersected elements are typically divided into sub-elements to which a phase needs to be assigned. Due to loss of information in the discretization of the level set field, certain geometrical configurations allow for ambiguous phase assignment of sub-elements, and thus ambiguous definition of the interface. The study presented here focuses on analyzing these topological ambiguities in embedded geometries constructed from discretized level set fields on hexahedral meshes. The analysis is performed on three-dimensional problems where several intersection configurations can significantly affect the problem's topology. This is in contrast to two-dimensional problems where ambiguous topological features exist only in one intersection configuration and identifying and resolving them is straightforward. A set of rules that resolve these ambiguities for two-phase problems is proposed, and algorithms for their implementations are provided. The influence of these rules on the evolution of the geometry in the optimization process is investigated with linear elastic topology optimization problems. These problems are solved by an explicit level set topology optimization framework that uses the extended finite element method to predict physical responses. This study shows that the choice of a rule to resolve topological features can result in drastically different final geometries. However, for the problems studied in this paper, the performances of the optimized design do not differ.

This paper introduces a simple formulation for topology optimization problems ensuring manufacturability by machining. The method distinguishes itself from existing methods by using the advection-diffusion equation with Robin boundary conditions to perform a filtering of the design variables. The proposed approach is less computationally expensive than the traditional methods used. Furthermore, the approach is easy to implement on unstructured meshes and in a distributed memory setting. Finally, the proposed approach can be performed with few to no continuation steps in any system parameters. Applications are demonstrated with topology optimization on unstructured meshes with up to 64 million elements and up to 29 milling tool directions.