Article
Engineering, Industrial
Nathan Hertlein, Philip R. Buskohl, Andrew Gillman, Kumar Vemaganti, Sam Anand
Summary: This study introduces a deep learning-based framework that predicts near optimal designs by learning latent similarities between runs in a training set using a conditional generative adversarial network (cGAN). The cGAN predictions show good agreement with true optima, and for greater accuracy, predictions can be further improved by applying a small number of topology optimization iterations.
JOURNAL OF MANUFACTURING SYSTEMS
(2021)
Article
Materials Science, Multidisciplinary
Nathan K. Brown, Anthony P. Garland, Georges M. Fadel, Gang Li
Summary: This paper investigates the use of reinforcement learning algorithms to automate the design of 2D discretized topologies. The results show that a deep RL agent can learn and apply generalized design strategies, demonstrating good performance in multi-objective design tasks.
MATERIALS & DESIGN
(2022)
Article
Computer Science, Interdisciplinary Applications
Soyoung Yoo, Sunghee Lee, Seongsin Kim, Kwang Hyeon Hwang, Jong Ho Park, Namwoo Kang
Summary: The study introduces a deep learning-based CAD/CAE framework for automatic 3D CAD design generation and engineering performance evaluation in the conceptual design phase, demonstrating the practical incorporation of AI into end-use product design projects and providing engineers and industrial designers with a rapid design tool.
STRUCTURAL AND MULTIDISCIPLINARY OPTIMIZATION
(2021)
Article
Engineering, Multidisciplinary
Waad Almasri, Dimitri Bettebghor, Faouzi Adjed, Florence Danglade, Fakhreddine Ababsa
Summary: This study integrates layout and mechanical constraints in the mechanical design process using deep learning technology and proposes a DL-layout-driven solution trained via a generative adversarial network framework. The solution can quickly generate mechanically valid designs conforming with layout constraints and has the capability to generate multiple shapes based on different input constraints.
ENGINEERING OPTIMIZATION
(2022)
Article
Automation & Control Systems
Nikita O. Starodubcev, Nikolay O. Nikitin, Elizaveta A. Andronova, Konstantin G. Gavaza, Denis O. Sidorenko, Anna Kalyuzhnaya
Summary: Generative design techniques have been widely applied in various fields, particularly in engineering, to automate initial stages of designing structures, minimizing routine work. However, existing approaches are limited by problem specificity and lack flexibility in method selection. To address these issues, a general approach named GEFEST (Generative Evolution For Encoded STructure) was proposed, providing sampling, estimation, and optimization principles for adaptable problem solutions. Experimental studies confirmed the effectiveness of GEFEST in coastal engineering, microfluidics, thermodynamics, and oil field planning, achieving improvements of 12%, 9%, 8%, and 7% respectively over baseline and state-of-the-art solutions.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2023)
Article
Computer Science, Interdisciplinary Applications
Minsik Seo, Seungjae Min
Summary: In this paper, a novel deep learning-aided material representation scheme is proposed for multi-scale topology optimization. This method establishes a general-purpose mapping from a low-dimensional variable to microstructural images using a deep generative model and a regression model. The generator and predictor networks are then integrated into the optimization process to reduce design variables and eliminate homogenization computations. The proposed method enables faster convergence and automatic satisfaction of complicated geometrical constraints.
ADVANCES IN ENGINEERING SOFTWARE
(2022)
Article
Biochemistry & Molecular Biology
Kostas Papadopoulos, Kathryn A. Giblin, Jon Paul Janet, Atanas Patronov, Ola Engkvist
Summary: The deep generative model trained with reinforcement learning using 3D shape and pharmacophore similarity scoring component has been shown to efficiently discover new leads in molecular design without relying on other information. Comparison with 2D QSAR models indicates that the 3D similarity based model produces more diverse outputs and allows for scaffold hopping and generation of novel series. Combining the two scoring components for training the generative model leads to more efficient exploration of desirable chemical space.
BIOORGANIC & MEDICINAL CHEMISTRY
(2021)
Article
Engineering, Multidisciplinary
Dengcheng Yan, Wenxin Xie, Yiwen Zhang, Qiang He, Yun Yang
Summary: This work proposes a deep reinforcement learning-based framework for hypernetwork dismantling, which can be applied to real-world tasks. By generating synthetic hypernetworks and conducting trial-and-error tasks, the dismantling strategy is continuously optimized. Experimental results demonstrate the effectiveness of the proposed framework.
IEEE TRANSACTIONS ON NETWORK SCIENCE AND ENGINEERING
(2022)
Article
Chemistry, Physical
Yong Zhao, Edirisuriya M. Dilanga Siriwardane, Zhenyao Wu, Nihang Fu, Mohammed Al-Fahdi, Ming Hu, Jianjun Hu
Summary: We propose a deep learning based Physics Guided Crystal Generative Model (PGCGM) for efficient crystal material design. Our model increases the generation validity by more than 700% compared to FTCP, one of the latest structure generators and by more than 45% compared to our previous CubicGAN model. The generated structures are validated using Density Functional Theory (DFT) calculations, with 1869 out of 2000 materials successfully optimized and deposited into the Carolina Materials Database, showing thermodynamic stability and potential synthesizability with negative formation energy and energy-above-hull less than 0.25 eV/atom for 39.6% and 5.3% of the materials, respectively.
NPJ COMPUTATIONAL MATERIALS
(2023)
Article
Food Science & Technology
Luana P. Queiroz, Carine M. Rebello, Erbet A. Costa, Vinicius V. Santana, Bruno C. L. Rodrigues, Alirio E. Rodrigues, Ana M. Ribeiro, Idelfonso B. R. Nogueira
Summary: This work proposes a novel framework based on scientific machine learning to tackle an emerging problem in flavor engineering and industry. By combining generative and reinforcement learning models, new flavor molecules are designed. The potential of the framework is validated by evaluating the synthesis accessibility, number of atoms, and likeness to natural or pseudo-natural products of the designed molecules.
Article
Nanoscience & Nanotechnology
Sean Hooten, Raymond G. Beausoleil, Thomas Van Vaerenbergh
Summary: The PHORCED technique utilizes a probabilistic generative neural network interfaced with an electromagnetic solver to assist in the design of photonic devices, showing better performing designs than local gradient-based inverse design methods. Additionally, transfer learning with PHORCED demonstrates that neural networks trained for specific tasks can be re-trained for different tasks with significantly fewer simulations.
Article
Computer Science, Artificial Intelligence
Hyo-Seok Hwang, Minhyeok Lee, Junhee Seok
Summary: In optical engineering, designing devices or systems with desired properties is important yet challenging. This paper proposes a deep reinforcement learning-based inverse design framework that utilizes a deep learning simulator to reduce training time and provides multiple design candidates to satisfy target properties.
APPLIED SOFT COMPUTING
(2022)
Article
Computer Science, Artificial Intelligence
Xin Li, Haojie Lei, Li Zhang, Mingzhong Wang
Summary: This paper explores interpretable Deep Reinforcement Learning (DRL) by representing policy using Differentiable Inductive Logic Programming (DILP). The research focuses on the optimization perspective of DILP-based policy learning and proposes using Mirror Descent for policy optimization. The theoretical and empirical studies verify the effectiveness of the proposed approach.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Chemistry, Multidisciplinary
Bongsung Bae, Haelee Bae, Hojung Nam
Summary: Computational drug design has advanced significantly in recent years, especially in the development of AI models for generating new chemical compounds with desired properties and enhanced binding affinity to target proteins. This study introduces LOGICS, a novel generative framework for designing target-focused chemical structures, which outperforms state-of-the-art models in optimizing binding affinity and generating more diverse de novo chemical structures.
JOURNAL OF CHEMINFORMATICS
(2023)
Article
Telecommunications
Meryem Simsek, Oner Orhan, Marcel Nassar, Oguz Elibol, Hosein Nikopour
Summary: As cellular networks continue to grow in density, traditional fiber backhaul access to each cell site becomes difficult. Utilizing millimeter wave communication and beamforming, high-speed wireless backhaul can be achieved. Our proposed topology formation approach, based on deep reinforcement learning and graph embedding, offers a less complex and more scalable solution with significant performance gains compared to baseline approaches.
IEEE COMMUNICATIONS LETTERS
(2021)
Review
Computer Science, Software Engineering
Xiaoqun Dai, Yan Hong
Summary: The primary objective of this research is to enhance the understanding of fabric mechanical behaviors, measurement techniques, and parameters essential for cloth simulation. The findings and information presented herein can be effectively utilized to enhance the precision and fidelity of apparel CAD systems, thereby facilitating advancements in virtual garment design and production.
COMPUTER-AIDED DESIGN
(2024)
Article
Computer Science, Software Engineering
Zhen-Pei Wang, Brian N. Cox, Shemuel Joash Kuehsamy, Mark Hyunpong Jhon, Olivier Sudre, N. Sridhar, Gareth J. Conduit
Summary: Three-dimensional non-periodic woven composite preforms have great design flexibility, but the design space is too large. This paper proposes a Background Vector Method (BVM) for generating candidate designs that can adapt to local architecture and global design goals while ensuring fabricability. Examples are provided to illustrate the design scope and speed of the BVM, as well as pathways for incorporating it into optimization algorithms.
COMPUTER-AIDED DESIGN
(2024)
Article
Computer Science, Software Engineering
Mohammad Mahdi Behzadi, Jiangce Chen, Horea T. Ilies
Summary: This paper proposes an approach to enhance the topological accuracy of machine learning-based topology optimization methods. The approach utilizes a predicted dual connectivity graph to improve the connectivity of the predicted designs. Experimental results show that the proposed method significantly improves the connectivity of the final predicted structures.
COMPUTER-AIDED DESIGN
(2024)
Article
Computer Science, Software Engineering
Jiaze Li, Shengfa Wang, Eric Paquette
Summary: In this study, a texture-driven adaptive mesh refinement method is proposed to generate high-quality 3D reliefs. By conducting feature-preserving adaptive sampling of the texture contours and using constraint-driven and feature-adaptive mesh subdivision, the method is able to accurately follow the texture contours and maintain good polygon quality.
COMPUTER-AIDED DESIGN
(2024)
Article
Computer Science, Software Engineering
Xi Zou, Sui Bun Lo, Ruben Sevilla, Oubay Hassan, Kenneth Morgan
Summary: This work presents a new method for generating triangular surface meshes in three dimensions for the NURBS-enhanced finite element method. The method allows for triangular elements that span across multiple NURBS surfaces, while maintaining the exact representation of the CAD geometry. This eliminates the need for de-featuring complex watertight CAD models and ensures compliance with user-specified spacing function requirements.
COMPUTER-AIDED DESIGN
(2024)
Article
Computer Science, Software Engineering
Ulderico Fugacci, Chiara Romanengo, Bianca Falcidieno, Silvia Biasotti
Summary: This paper proposes a method for suitably resampling a 3D point cloud while preserving the feature curves to which some points belong. The method enriches the cloud by approximating curvilinear profiles and allows for point removal or insertion without affecting the approximated profiles. The effectiveness of the method is evaluated through experiments and comparisons.
COMPUTER-AIDED DESIGN
(2024)
Article
Computer Science, Software Engineering
J. Hinz, O. Chanon, A. Arrigoni, A. Buffa
Summary: The objective of this study is to address the difficulty of simplifying a geometric model while maintaining the accuracy of the solution. A goal-oriented adaptive strategy is proposed to reintroduce geometric features in regions with significant impact on the quantity of interest. This approach enables faster and more efficient simulations.
COMPUTER-AIDED DESIGN
(2024)
Article
Computer Science, Software Engineering
Hao Qiu, Yixiong Feng, Yicong Gao, Zhaoxi Hong, Jianrong Tan
Summary: Sandwich panels with excellent mechanical properties are widely used, and kirigami-inspired structural designs are receiving increasing attention. In this study, novel graded self-locking kirigami panels based on a tucked-interleaved pattern are developed and analyzed. The experimental and simulation results demonstrate that the proposed kirigami panels have outstanding load-to-weight ratios and can generate graded stiffness and superior specific energy absorption.
COMPUTER-AIDED DESIGN
(2024)
Article
Computer Science, Software Engineering
Zheng Zhan, Wenping Wang, Falai Chen
Summary: This article proposes a learning based method using a deep neural network to simultaneously parameterize the boundary and interior of a computational domain. The method achieves robust parameterization by optimizing a loss function and fitting a tensor-product B-spline function. Experimental results demonstrate that the proposed approach yields parameterization results with lower distortion and higher bijectivity ratio.
COMPUTER-AIDED DESIGN
(2024)