Article
Mechanics
Samuel S. Schoenholz, Ekin D. Cubuk
Summary: JAX MD is a software package for differentiable physics simulations, focusing on molecular dynamics. It includes physics simulation environments, interaction potentials, and neural networks. The simulations are differentiable, allowing for meta-optimization.
JOURNAL OF STATISTICAL MECHANICS-THEORY AND EXPERIMENT
(2021)
Review
Computer Science, Information Systems
Marcelo Guerra Hahn, Silvia Margarita Baldiris Navarro, Luis De la Fuente Valentin, Daniel Burgos
Summary: Automatic scoring and feedback tools are crucial in online learning, offering benefits such as improving student experience, eliminating scoring biases, but also potentially hindering student creativity and promoting surface-level learning. These drawbacks provide opportunities to enhance technologies for a better learning experience.
Article
Computer Science, Interdisciplinary Applications
Aaditya Chandrasekhar, Krishnan Suresh
Summary: In this study, a new topology optimization method is proposed using neural networks to represent and optimize the density field, resulting in sharp and differentiable boundaries. The research demonstrates that the method is simple to implement and illustrates its application through 2D and 3D examples. Some unresolved challenges with the proposed framework are also summarized.
STRUCTURAL AND MULTIDISCIPLINARY OPTIMIZATION
(2021)
Article
Computer Science, Interdisciplinary Applications
Yi Xing, Liyong Tong
Summary: In this work, a machine learning-assisted structural optimization (MLaSO) scheme is proposed to accelerate the computational speed of structural optimization. A new machine learning model is used to predict the update of the optimization quantity during the optimization process, eliminating the need for finite element analysis and sensitivity analysis. The MLaSO scheme can be easily integrated into different structural optimization methods and does not require additional training datasets.
STRUCTURAL AND MULTIDISCIPLINARY OPTIMIZATION
(2022)
Article
Computer Science, Artificial Intelligence
Aleksandar Kovacevic, Jelena Slivka, Dragan Vidakovic, Katarina-Glorija Grujic, Nikola Luburic, Simona Prokic, Goran Sladic
Summary: This paper compares the performance of machine learning-based and metric-based code smell detection methods, and evaluates the effectiveness of different source code representations. The study also explores the transferability of knowledge mined from code understanding models to code smell detection and provides a systematic evaluation of code smell detection approaches.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Engineering, Multidisciplinary
Jorge Lopez, Cosmin Anitescu, Timon Rabczuk
Summary: This study introduces a method for isogeometric structural shape optimization using a multilevel approach and automated sensitivity analysis. The use of automatic differentiation toolbox facilitates accurate computation of gradients, improving efficiency. The benefits and limitations of using automated gradients are discussed through analytical, numerical, and automatic sensitivity analyses.
APPLIED MATHEMATICAL MODELLING
(2021)
Article
Engineering, Multidisciplinary
Hao Deng, Praveen S. Vulimiri, Albert C. To
Summary: This paper presents an efficient and compact MATLAB code for three-dimensional stress-based sensitivity analysis, including finite element analysis and p-norm stress sensitivity analysis. The effectiveness of the sensitivity analysis code is demonstrated through verification of the correctness of the analytical sensitivity.
OPTIMIZATION AND ENGINEERING
(2022)
Article
Computer Science, Artificial Intelligence
Kadierdan Kaheman, Steven L. Brunton, J. Nathan Kutz
Summary: The Sparse Identification of Nonlinear Dynamics (SINDy) is a regression framework used for discovering dynamic models and equations from time-series data. This study presents an improved SINDy algorithm that combines automatic differentiation and constraint techniques to denoise the data, learn the noise probability distribution, and identify the underlying dynamic system. The algorithm achieves higher accuracy and robustness in handling noise compared to existing methods.
MACHINE LEARNING-SCIENCE AND TECHNOLOGY
(2022)
Review
Computer Science, Artificial Intelligence
Nan Li, Lianbo Ma, Tiejun Xing, Guo Yu, Chen Wang, Yingyou Wen, Shi Cheng, Shangce Gao
Summary: Machine learning (ML), the most promising paradigm for discovering deep knowledge from data, has been widely applied in practical applications such as recommender systems, virtual reality, and semantic segmentation. However, building high-quality ML systems for specific tasks is challenging due to the need for expert knowledge and high computation costs. This paper provides a comprehensive review of evolutionary machine learning (EML) methods, discussing concepts, taxonomy criteria, research problems, and limitations. The automatic design of ML using evolutionary computation is an increasingly popular research trend that can address the challenges of developing ML in large-scale practical applications.
APPLIED SOFT COMPUTING
(2023)
Article
Materials Science, Multidisciplinary
Fanping Sui, Ruiqi Guo, Zhizhou Zhang, Grace X. Gu, Liwei Lin
Summary: This paper introduces the concept of digital materials and their application in composite material design. Through a deep reinforcement learning scheme, an automated process for digital material design is achieved, resulting in improved design quality and significant computational advantages.
ACS MATERIALS LETTERS
(2021)
Article
Computer Science, Interdisciplinary Applications
Tianju Xue, Shuheng Liao, Zhengtao Gan, Chanwook Park, Xiaoyu Xie, Wing Kam Liu, Jian Cao
Summary: This paper introduces JAX-FEM, an open-source differentiable finite element method (FEM) library that is implemented with pure Python and scalable for solving moderate to large problems. It achieves significant acceleration compared to commercial FEM codes and enables automatic solving of inverse problems. JAX-FEM also serves as an integrated platform for machine learning-aided computational mechanics.
COMPUTER PHYSICS COMMUNICATIONS
(2023)
Article
Computer Science, Interdisciplinary Applications
Daniel Yago, Juan Cante, Oriol Lloberas-Valls, Javier Oliver
Summary: This paper introduces an efficient and comprehensive MATLAB code for solving two-dimensional structural topology optimization problems, including a variety of different problems such as compliant mechanism synthesis and multi-load compliance. The code includes various improvements and extensions, making it suitable for students and newcomers in the field of topology optimization.
STRUCTURAL AND MULTIDISCIPLINARY OPTIMIZATION
(2021)
Article
Computer Science, Artificial Intelligence
Xu Chen, Brett Wujek
Summary: In this paper, we propose a novel unified framework called AutoDAL for automated distributed active learning to address multiple challenging problems in active learning. The framework is able to handle limited labeled data, imbalanced datasets, automatic hyperparameter selection, and scalability to big data. Experimental results show that the proposed AutoDAL algorithm achieves significantly better performance compared to several state-of-the-art AutoML approaches and active learning algorithms.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Engineering, Geological
Hongwei Huang, Jiaqi Chang, Dongming Zhang, Jie Zhang, Huiming Wu, Gang Li
Summary: This study proposes an improved method to control the construction quality of tunnels by using machine learning algorithms to predict the posture of shield machines. The parameters for shield tunneling are extracted and three machine learning algorithms, MLP, SVM, and GBR, are improved using genetic algorithm and principal component analysis. The test results show that the averaged R-2 of MLP, SVM, and GBR based models are 0.942, 0.935, and 0.6, respectively. An application example is provided to illustrate the automatic control of the shield tunnel posture using the proposed models.
JOURNAL OF ROCK MECHANICS AND GEOTECHNICAL ENGINEERING
(2022)
Article
Multidisciplinary Sciences
Chunyan Zhang, Junchao Wang, Qinglei Zhou, Ting Xu, Ke Tang, Hairen Gui, Fudong Liu
Summary: Source code summarization is the natural language description of the function of source code, which helps developers understand the semantics of the code. Currently, automatic generation of code summaries is an efficient way to overcome the inefficiency of manual summarization.
Article
Computer Science, Interdisciplinary Applications
Akshay Desai, Mihir Mogra, Saketh Sridhara, Kiran Kumar, Gundavarapu Sesha, G. K. Ananthasuresh
Summary: In this study, topological derivatives are used to design structures with non-periodic continuous fibers optimally arranged in specific patterns, achieving maximum stiffness by determining the distribution of anisotropic fiber material within isotropic matrix material. By adjusting the level-set plane in the topological sensitivity field, a Pareto surface of stiffness and two volume fractions is generated. The use of topological derivatives allows for interchanging different materials during iterative optimization, with the optimal orientation of fibers determined at each point to align with principal stress directions.
STRUCTURAL AND MULTIDISCIPLINARY OPTIMIZATION
(2021)
Article
Computer Science, Software Engineering
Aaditya Chandrasekhar, Krishnan Suresh
Summary: This paper presents a neural network-based method for multi-material topology optimization, which can achieve the distribution of multiple materials within the topology and solve some challenges encountered when using mesh-based methods.
COMPUTER-AIDED DESIGN
(2021)
Article
Engineering, Multidisciplinary
Tej Kumar, Saketh Sridhara, Bhagyashree Prabhune, Krishnan Suresh
Summary: Multi-scale topology optimization (MTO) is widely used in design, but computationally expensive. Graded MTO is proposed to reduce computational cost, but faces positive-definiteness issues. A spectral decomposition-based approach is proposed in this paper to guarantee positive-definite elasticity matrices.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2021)
Article
Computer Science, Interdisciplinary Applications
Alireza H. Taheri, Krishnan Suresh
Summary: The authors extend the concept of Generalized NURBS (GNURBS) to bivariate parametric surfaces by decoupling the weights along different physical coordinates, providing additional flexibility and control. This concept effectively improves the performance of NURBS and enhances the approximation of certain surfaces. Through the development of a comprehensive MATLAB toolbox, the authors demonstrate the advantages of GNURBS surfaces compared to classic NURBS surfaces.
ENGINEERING WITH COMPUTERS
(2022)
Article
Engineering, Multidisciplinary
Bhagyashree Prabhune, Saketh Sridhara, Krishnan Suresh
Summary: This article introduces a tangled finite element method (TFEM) for handling concave elements in four-node quadrilateral meshes. TFEM resolves the ambiguity of the field in the tangled region through careful definition and imposes an equality condition on the field at re-entrant nodes of the concave elements. The method has been shown to achieve accurate results and optimal convergence even over severely tangled meshes.
INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
(2022)
Article
Computer Science, Software Engineering
Aaditya Chandrasekhar, Krishnan Suresh
Summary: In this paper, an approximate length scale filter strategy for topology optimization (TO) is proposed by extending a density-based TO formulation using neural networks (TOuNN). The proposed method enhances TOuNN with a Fourier space projection to approximately control the minimum and/or maximum length scales. The method does not involve additional constraints and automates sensitivity computations using the neural net's library.
COMPUTER-AIDED DESIGN
(2022)
Article
Computer Science, Interdisciplinary Applications
Aaditya Chandrasekhar, Saketh Sridhara, Krishnan Suresh
Summary: In this paper, the authors propose a method for simultaneous optimization using variational autoencoders. The discrete material database is projected onto a continuous and differentiable latent space using a data-driven VAE. This is combined with a fully-connected neural network embedded with a finite-element solver to optimize the material and geometry simultaneously.
ENGINEERING WITH COMPUTERS
(2022)
Article
Engineering, Manufacturing
Subodh C. Subedi, Ahmad Shahba, Mythili Thevamaran, Dan J. Thoma, Krishnan Suresh
Summary: In this study, a novel method for the design of support structures in laser powder bed fusion (LPBF)-based metal additive manufacturing is proposed. By extracting the equivalent static load (ESL) from transient simulation results, the size of support structures is optimized to minimize material and time consumption. Numerical experiments and sample fabrication validate the effectiveness of this method.
ADDITIVE MANUFACTURING
(2022)
Article
Computer Science, Interdisciplinary Applications
Aaditya Chandrasekhar, Saketh Sridhara, Krishnan Suresh
Summary: We propose a novel graded multiscale topology optimization framework using the classification capacity of neural networks. This framework has several key features: (1) the number of design variables is weakly dependent on the pre-selected microstructures, (2) it ensures partition of unity and discourages microstructure mixing, (3) it supports automatic differentiation for eliminating manual sensitivity analysis, and (4) it enables high-resolution re-sampling for smoother variation of microstructure topologies. The proposed framework is demonstrated through multiple examples.
ADVANCES IN ENGINEERING SOFTWARE
(2023)
Article
Engineering, Multidisciplinary
Bhagyashree Prabhune, Krishnan Suresh
Summary: The finite element method requires tangle-free elements and convex quadrilateral and hexahedral elements. However, generating high-quality tangle-free meshes, especially 3D hexahedral meshes, is challenging. The tangled finite element method (TFEM) was proposed to handle concave quadrilateral elements but is computationally expensive and complex. Here, they present a computationally efficient isoparametric-TFEM (i-TFEM) framework for inverted quadrilateral and hexahedral elements, which is simple, efficient, and accurate with optimal convergence rate.
COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING
(2023)
Article
Computer Science, Software Engineering
Aaditya Chandrasekhar, Amir Mirzendehdel, Morad Behandish, Krishnan Suresh
Summary: In this paper, a topology optimization framework is proposed to simultaneously optimize the matrix topology and fiber distribution of functionally graded continuous fiber-reinforced composites. By using a mesh-independent representation based on a neural network, the accurate extraction of functionally graded continuous fibers can be achieved, and automatic differentiation can be employed for sensitivity analysis to improve computational efficiency. The effectiveness and computational efficiency of the proposed method are demonstrated through various numerical examples, and it is shown that the optimized continuous fiber-reinforced composites can be directly fabricated at high resolution using additive manufacturing.
COMPUTER-AIDED DESIGN
(2023)
Article
Computer Science, Interdisciplinary Applications
Rahul Kumar Padhy, Aaditya Chandrasekhar, Krishnan Suresh
Summary: Fluid-flow devices with low dissipation and large contact area are important in many applications. Multi-scale topology optimization (MTO) is a well-known strategy for designing such devices, but it is computationally expensive. This study proposes a graded multiscale topology optimization method to minimize dissipation in fluid-flow devices and achieve a desired contact area.
ENGINEERING WITH COMPUTERS
(2023)
Article
Computer Science, Interdisciplinary Applications
Bhagyashree Prabhune, Krishnan Suresh
Summary: This paper presents a method called isoparametric tangled finite element method (i-TFEM) to handle tangled meshes efficiently, making it applicable for real-world scenarios.
ENGINEERING WITH COMPUTERS
(2023)
