arXiv subjects

Kikuo Fujita

Publications and source records attributed to Kikuo Fujita.

8 recordsLinked to original sources

arXiv2025

Homogenization-based optimization of wall thickness distribution for TPMS two-fluid heat exchangers

Triply Periodic Minimal Surface (TPMS) structures are attracting growing attention as promising geometries for next-generation high-performance heat exchangers (HXs), due to their continuous flow paths and high surface-area-to-volume ratio that enhance heat transfer performance. Among these, graded TPMS structures with spatially varying thickness have emerged as a potential means to further improve performance. This study proposes an optimization method of wall thickness distribution for TPMS HXs based on an effective porous media model, which allows accurate performance prediction while significantly reducing computational cost. The proposed method is applied to a gyroid two-fluid HX aiming to improve the thermal-hydraulic performance. Furthermore, full-scale numerical simulations of the optimized-thickness design show a 12.2% improvement in the performance evaluation criterion (PEC) compared to the uniform-thickness design. The improvement is primarily attributed to the optimized non-uniform wall thickness, which directs more flow toward the core ends and enhances velocity uniformity. As a result, heat transfer is enhanced at the core ends, leading to more effective use of the entire HX core and improved overall thermal performance.

math.OC↗

arXiv2025

Wasserstein crossover for evolutionary algorithm-based topology optimization

Evolutionary algorithms (EAs) are promising approaches for non-differentiable or strongly multimodal topology optimization problems, but they often suffer from the curse of dimensionality, generally leading to low-resolution optimized results. This limitation stems in part from the difficulty of producing effective offspring through traditional crossover operators, which struggle to recombine complex parent design features in a meaningful way. In this study, we propose a novel crossover operator for topology optimization, termed Wasserstein crossover, and develop a corresponding EA-based optimization framework. Our method leverages a morphing technique based on the Wasserstein distance -- a distance metric between probability distributions derived from the optimal transport theory. Its key idea is to treat material distributions as probability distributions and generate offspring as Wasserstein barycenters, enabling smooth and interpretable interpolation between parent designs while preserving their structural features. The proposed framework incorporates Wasserstein crossover into an EA under a multifidelity design scheme, where low-fidelity optimized initial designs evolve through iterations of Wasserstein crossover and selection based on high-fidelity evaluation. We apply the proposed framework to three topology optimization problems: maximum stress minimization in two- and three-dimensional structural mechanics, and turbulent heat transfer in two-dimensional thermofluids. The results demonstrate that candidate solutions evolve iteratively toward high-performance designs through Wasserstein crossover, highlighting its potential as an effective crossover operator and validating the usefulness of the proposed framework for solving intractable topology optimization problems.

math.OC↗

arXiv2025

Maximum stress minimization via data-driven multifidelity topology design

The maximum stress minimization problem is among the most important topics for structural design. The conventional gradient-based topology optimization methods require transforming the original problem into a pseudo-problem by relaxation techniques. Since their parameters significantly influence optimization, accurately solving the maximum stress minimization problem without using relaxation techniques is expected to achieve extreme performance. This paper focuses on this challenge and investigates whether designs with more avoided stress concentrations can be obtained by solving the original maximum stress minimization problem without relaxation techniques, compared to the solutions obtained by gradient-based topology optimization. We employ data-driven multifidelity topology design (MFTD), a gradient-free topology optimization based on evolutionary algorithms. The basic framework involves generating candidate solutions by solving a low-fidelity optimization problem, evaluating these solutions through high-fidelity forward analysis, and iteratively updating them using a deep generative model without sensitivity analysis. In this study, data-driven MFTD incorporates the optimized designs obtained by solving a gradient-based topology optimization problem with the p-norm stress measure in the initial solutions and solves the original maximum stress minimization problem based on a high-fidelity analysis with a body-fitted mesh. We demonstrate the effectiveness of our proposed approach through the benchmark of L-bracket. As a result of solving the original maximum stress minimization problem with data-driven MFTD, a volume reduction of up to 22.6% was achieved under the same maximum stress value, compared to the initial solution.

math.OC↗

arXiv2024

Deep Concept Identification for Generative Design

A generative design based on topology optimization provides diverse alternatives as entities in a computational model with a high design degree. However, as the diversity of the generated alternatives increases, the cognitive burden on designers to select the most appropriate alternatives also increases. Whereas the concept identification approach, which finds various categories of entities, is an effective means to structure alternatives, evaluation of their similarities is challenging due to shape diversity. To address this challenge, this study proposes a concept identification framework for generative design using deep learning (DL) techniques. One of the key abilities of DL is the automatic learning of different representations of a specific task. Deep concept identification finds various categories that provide insights into the mapping relationships between geometric properties and structural performance through representation learning using DL. The proposed framework generates diverse alternatives using a generative design technique, clusters the alternatives into several categories using a DL technique, and arranges these categories for design practice using a classification model. This study demonstrates its fundamental capabilities by implementing variational deep embedding, a generative and clustering model based on the DL paradigm, and logistic regression as a classification model. A simplified design problem of a two-dimensional bridge structure is applied as a case study to validate the proposed framework. Although designers are required to determine the viewing aspect level by setting the number of concepts, this implementation presents the identified concepts and their relationships in the form of a decision tree based on a specified level.

cs.LG↗

arXiv2024

Data-driven topology design with persistent homology for enhancing population diversity

This paper proposes a selection strategy for enhancing population diversity in data-driven topology design (DDTD), a topology optimization framework based on evolutionary algorithms (EAs) using a deep generative model. While population diversity is essential for global search with EAs, conventional selection operators that preserve diverse solutions based on objective values may still lead to a loss of population diversity in topology optimization problems due to the high dimensionality of design variable space and strong nonlinearity of evaluation functions. Motivated by the idea that topology is what characterizes the inherent diversity among material distributions, we employ a topological data analysis method called persistent homology. As a specific operation, a Wasserstein distance sorting between persistence diagrams is introduced into a selection algorithm to maintain the intrinsic population diversity. We apply the proposed selection operation incorporated into DDTD to a stress-based topology optimization problem as a numerical example. The results confirm that topology can be analyzed using persistent homology and that the proposed selection operation significantly enhances the search performance of DDTD.

math.OC↗

arXiv2021

Data-driven topology design using a deep generative model

In this paper, we propose a sensitivity-free and multi-objective structural design methodology called data-driven topology design. It is schemed to obtain high-performance material distributions from initially given material distributions in a given design domain. Its basic idea is to iterate the following processes: (i) selecting material distributions from a dataset of material distributions according to eliteness, (ii) generating new material distributions using a deep generative model trained with the selected elite material distributions, and (iii) merging the generated material distributions with the dataset. Because of the nature of a deep generative model, the generated material distributions are diverse and inherit features of the training data, that is, the elite material distributions. Therefore, it is expected that some of the generated material distributions are superior to the current elite material distributions, and by merging the generated material distributions with the dataset, the performances of the newly selected elite material distributions are improved. The performances are further improved by iterating the above processes. The usefulness of data-driven topology design is demonstrated through numerical examples.

physics.comp-ph↗

arXiv2020

Topology design of two-fluid heat exchange

Heat exchangers are devices that typically transfer heat between two fluids. The performance of a heat exchanger such as heat transfer rate and pressure loss strongly depends on the flow regime in the heat transfer system. In this paper, we present a density-based topology optimization method for a two-fluid heat exchange system, which achieves a maximum heat transfer rate under fixed pressure loss. We propose a representation model accounting for three states, i.e., two fluids and a solid wall between the two fluids, by using a single design variable field. The key aspect of the proposed model is that mixing of the two fluids can be essentially prevented without any penalty scheme. This is because the solid constantly exists between the two fluids due to the use of the single design variable field. We demonstrate the effectiveness of the proposed approach through three-dimensional numerical examples in which an optimized design is compared with a simple reference design, and the effects of design conditions (i.e., Reynolds number, Prandtl number, design domain size, and flow arrangements) are investigated.

cs.CE↗

arXiv2018

Novel flow field design for vanadium redox flow batteries via topology optimization

This paper presents a three-dimensional topology optimization method for the design of flow field in vanadium redox flow batteries (VRFBs). We focus on generating a novel flow field configuration for VRFBs via topology optimization, which has been attracted attention as a powerful design tool based on numerical optimization. An attractive feature of topology optimization is that a topology optimized configuration can be automatically generated without presetting a promising design candidate. In this paper, we formulate the topology optimization problem as a maximization problem of the electrode surface concentration in the negative electrode during the charging process. The aim of this optimization problem is to obtain a topology optimized flow field that enables the improvement of mass transfer effect in a VRFB. We demonstrate that a novel flow field configuration can be obtained through the numerical investigation. To clarify the performance of the topology optimized flow field, we investigate the mass transfer effect through the comparison with reference flow fields---parallel and interdigitated flow fields---and the topology optimized flow field. In addition, we discuss the power loss that takes account of the polarization loss and pumping power, at various operating conditions.

math.OC↗