A Deep Learning Surrogate Model for Topology Optimization

Barmada, Sami; Fontana, Nunzia; Formisano, A.; Thomopulos, Dimitri; Tucci, Mauro

doi:10.1109/tmag.2021.3063470

Cited by 25 publications

(33 citation statements)

References 3 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In previous works, authors have introduced various deep learning-based optimization algorithms [7,9,10], and their main characteristics are summarized in Table 1. In particular, in [10], a standard autoencoder was used to represent the input geometries of a special device as bitmaps, and the electromagnetic problem was optimized in the latent space using an evolutionary optimization algorithm.…”

Section: Related Workmentioning

confidence: 99%

“…In particular, in [10], a standard autoencoder was used to represent the input geometries of a special device as bitmaps, and the electromagnetic problem was optimized in the latent space using an evolutionary optimization algorithm. In [7,9], deep learning models were developed to represent both the geometry and the relationship between the geometry and the resulting electromagnetic fields, and the optimization was carried out using particle swarm optimization (PSO) [9] and a genetic optimization algorithm (GA) [7]. In the aforementioned works, particularly [7,10], the problem of obtaining a surrogate model to properly solve a given electromagnetic problem for different geometries was solved using two main DL models: namely a model to represent the input image in a reduced dimensional space, i.e., the latent space of the autoencoder; and a model to predict the electromagnetic variables (to be optimized) as a function of geometry, that we shall denote as a surrogate model (SM).…”

Section: Related Workmentioning

confidence: 99%

“…Recently, many research areas have benefited from the potential offered by deep learning (DL) tools, such as convolutional neural networks (CNNs) [1,2]. In fact, recent works on the DL-assisted analysis of electromagnetic (EM) field computation problems showed the promising potential of CNN applications [3][4][5][6][7][8][9][10][11][12][13][14][15].…”

Section: Introductionmentioning

confidence: 99%

“…The main application of most of the mentioned works is the optimization of the geometry of rotating electrical machines, in particular permanent magnet machines, in order to optimize an electromagnetic parameter such as maximizing the motor torque and/or minimizing the cogging torque. Some works have faced additional theoretical problems such as obtaining reduced models [9], or treating the optimization of special electromagnetic devices [7,10].…”

Section: Introductionmentioning

confidence: 99%

“…Among the DL models and architectures, autoencoders enjoy particular attention in applications where the high dimensional input space needs to be mapped and interpreted in a lower dimensional space, as in the case of complex geometries of electromagnetic devices [7,10]. Autoencoders can use fully connected and convolutional layers and perform nonlinear compression through the encoder network of the input space to a lower dimensional, called the latent space, and the compressed data can be decoded back (reconstructed) to the original space through the decoder network.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

A Regularized Procedure to Generate a Deep Learning Model for Topology Optimization of Electromagnetic Devices

et al. 2021

Self Cite

View full text Add to dashboard Cite

The use of behavioral models based on deep learning (DL) to accelerate electromagnetic field computations has recently been proposed to solve complex electromagnetic problems. Such problems usually require time-consuming numerical analysis, while DL allows achieving the topologically optimized design of electromagnetic devices using desktop class computers and reasonable computation times. An unparametrized bitmap representation of the geometries to be optimized, which is a highly desirable feature needed to discover completely new solutions, is perfectly managed by DL models. On the other hand, optimization algorithms do not easily cope with high dimensional input data, particularly because it is difficult to enforce the searched solutions as feasible and make them belong to expected manifolds. In this work, we propose the use of a variational autoencoder as a data regularization/augmentation tool in the context of topology optimization. The optimization was carried out using a gradient descent algorithm, and the DL neural network was used as a surrogate model to accelerate the resolution of single trial cases in the due course of optimization. The variational autoencoder and the surrogate model were simultaneously trained in a multi-model custom training loop that minimizes total loss—which is the combination of the two models’ losses. In this paper, using the TEAM 25 problem (a benchmark problem for the assessment of electromagnetic numerical field analysis) as a test bench, we will provide a comparison between the computational times and design quality for a “classical” approach and the DL-based approach. Preliminary results show that the variational autoencoder manages regularizing the resolution process and transforms a constrained optimization into an unconstrained one, improving both the quality of the final solution and the performance of the resolution process.

show abstract

Section: Related Workmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

A Regularized Procedure to Generate a Deep Learning Model for Topology Optimization of Electromagnetic Devices

et al. 2021

Self Cite

View full text Add to dashboard Cite

show abstract

Accelerating Analysis for Structure Design via Deep Learning Surrogate Models

Shao

Chen

Wang

et al. 2023

Advanced Intelligent Systems

View full text Add to dashboard Cite

Using computer simulation tools such as finite element analysis (FEA) to perform material stress analysis is a common design method in engineering practice. In order to model more realistic real‐world systems, simulation models have become more complex, and calculation becomes more expensive as a result. The rise of artificial intelligence technologies has made it possible to integrate deep learning methods and material stress analysis. Herein, FEA software is employed to obtain a large number of analysis cases as training samples and uses a fully connected neural network and long‐short‐term memory neural network as surrogate models, which can predict the stress distribution and stress sequence in the process of the bullet impacting target plates with different materials. These models can give results similar to FEA with 92.19% and 92.41% accuracy, respectively. The experimental results show that the deep learning surrogate models have great potential in material stress analysis.

show abstract

AuTO: a framework for Automatic differentiation in Topology Optimization

Chandrasekhar

Sridhara

Suresh

2021

Struct Multidisc Optim

View full text Add to dashboard Cite

A critical step in topology optimization (TO) is finding sensitivities. Manual derivation and implementation of the sensitivities can be quite laborious and error-prone, especially for non-trivial objectives, constraints and material models. An alternate approach is to utilize automatic differentiation (AD). While AD has been around for decades, and has also been applied in TO, wider adoption has largely been absent. In this educational paper, we aim to reintroduce AD for TO, and make it easily accessible through illustrative codes. In particular, we employ JAX, a high-performance Python library for automatically computing sensitivities from a user defined TO problem. The resulting framework, referred to here as AuTO, is illustrated through several examples in compliance minimization, compliant mechanism design and microstructural design.

show abstract

A Deep Learning Surrogate Model for Topology Optimization

Cited by 25 publications

References 3 publications

A Regularized Procedure to Generate a Deep Learning Model for Topology Optimization of Electromagnetic Devices

A Regularized Procedure to Generate a Deep Learning Model for Topology Optimization of Electromagnetic Devices

Accelerating Analysis for Structure Design via Deep Learning Surrogate Models

AuTO: a framework for Automatic differentiation in Topology Optimization

Contact Info

Product

Resources

About