Deep Autoencoder for Off-Line Design-Space Dimensionality Reduction in Shape Optimization

D’Agostino, Danny; Serani, Andrea; Campana, Emilio F.; Diez, Matteo

doi:10.2514/6.2018-1648

Cited by 22 publications

(13 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The effectiveness and efficiency of the ADR method are affected, in general, by the nonlinearities involved in the process. In order to quantify their effects, ongoing research focuses on nonlinear extensions of the current methodology [28][29][30].…”

Section: Discussionmentioning

confidence: 99%

Stochastic Shape Optimization via Design-Space Augmented Dimensionality Reduction and RANS Computations

Serani

Diez

Wackers

et al. 2019

AIAA Scitech 2019 Forum

Self Cite

View full text Add to dashboard Cite

The paper presents how to efficiently and effectively solve stochastic shape optimization problems by combing Reynolds-averaged Navier-Stokes (RANS) equation solvers with designspace augmented dimensionality reduction (ADR). This study has been conducted within the NATO Science and Technology Organization, Applied Vehicle Technology, Task Group AVT-252 "Stochastic Design Optimization for Naval and Aero Military Vehicles." The application pertains to the robust and the reliability-based robust design optimization of a destroyer hullform for resistance in calm water and waves and seakeeping performance, under stochastic environmental and operating conditions (speed, sea state, heading). The current work extends previous research by the authors, presented at earlier AIAA conferences [1-3], where only potential flow solvers were used. In the present work, the expected value of the total resistance is reduced respectively by 4.4 and 3% in calm water and waves. An 8% improvement of the seakeeping performance is also achieved. Design-space assessment by ADR is demonstrated to be a viable option in solving the curse of dimensionionality in shape optimization, especially when high-fidelity CPU-expensive solvers are used.

show abstract

Section: Discussionmentioning

confidence: 99%

Stochastic Shape Optimization via Design-Space Augmented Dimensionality Reduction and RANS Computations

Serani

Diez

Wackers

et al. 2019

AIAA Scitech 2019 Forum

Self Cite

View full text Add to dashboard Cite

show abstract

“…Thus, researchers also apply non-linear methods to reduce the dimensionality of design spaces. This non-linearity can be achieved by (1) applying linear reduction techniques locally to construct a non-linear global manifold [14,15,16,17,12]; (2) using kernel methods with linear reduction techniques (i.e., using linear methods in a Reproducing Kernel Hilbert Space that then induces non-linearity in the original design space) [1,12]; (3) latent variable models like Gaussian process latent variable model (GPLVM) and generative topographic mapping (GTM) [18]; and 4) neural networks based approaches such as self-organizing maps [19] and autoencoders [20,1,12,21].…”

Section: Design Space Dimensionality Reductionmentioning

confidence: 99%

“…Usually dimensionality reduction techniques allow inverse transformations from the latent space back to the de-sign space, thus can synthesize new designs from latent variables [11,1,12,21]. For example, under the PCA model, the latent variables define a linear combination of principal components to synthesize a new design [11]; for local manifold based approaches, a new design can be synthesized via interpolation between neighboring points on the local manifold [17]; and under the autoencoder model, the trained decoder maps any given point in the latent space to a new design [20,21]. Researchers have also employed generative models such as kernel density estimation [23], Boltzmann machines [24], variational autoencoders (VAEs) [25], and generative adversarial nets (GANs) [26,27] to learn the distribution of samples in the design space, and synthesize new designs by drawing samples from the learned distribution.…”

Section: Data-driven Design Synthesismentioning

confidence: 99%

Synthesizing Designs With Inter-Part Dependencies Using Hierarchical Generative Adversarial Networks

Chen

Jeyaseelan

Fuge

2018

Volume 2A: 44th Design Automation Conference

View full text Add to dashboard Cite

Real-world designs usually consist of parts with hierarchical dependencies, i.e., the geometry of one component (a child shape) is dependent on another (a parent shape). We propose a method for synthesizing this type of design. It decomposes the problem of synthesizing the whole design into synthesizing each component separately but keeping the inter-component dependencies satisfied. This method constructs a two-level generative adversarial network to train two generative models for parent and child shapes, respectively. We then use the trained generative models to synthesize or explore parent and child shapes separately via a parent latent representation and infinite child latent representations, each conditioned on a parent shape. We evaluate and discuss the disentanglement and consistency of latent representations obtained by this method. We show that shapes change consistently along any direction in the latent space. This property is desirable for design exploration over the latent space.

show abstract

“…In a design context, these techniques are structured on the assumption that the geometric variability in design space is not the same in all directions and there are only a few inherent feature directions which materialise most improvement in the design and these inherent features can form the basis of a new lower-dimensional latent subspace [10]. In literature, such feature extraction has been achieved with (1) linear techniques to locally construct a nonlinear global manifold such as Principal Component Analysis (PCA) [9,11], (2) kernel function with linear reduction techniques such as Kernel PCA [5,12], or (3) with neural network-based approaches such as auto-encoders [13].…”

Section: Introductionmentioning

confidence: 99%

Physics-Informed Feature-to-Feature Learning for Design-Space Dimensionality Reduction in Shape Optimisation

Khan

Serani

Diez

et al. 2021

AIAA Scitech 2021 Forum

Self Cite

View full text Add to dashboard Cite

High-dimensional parametric design problems cause optimisers and physics simulations to suffer from the curse-of-dimensionality, resulting in high computational cost. In this work, to release this computational burden, we adopted a two-step feature-to-feature learning methodology to discover a lower-dimensional latent space, based on the combination of geometry-and physics-informed principal component analysis and the active subspace method. At the first step, statistical dependencies implicit in the design parameters encode important geometric features of the underline shape. During the second step, functional features of designs are extracted in term of previously learned geometric features. Afterwards, both geometric and functional features are augmented together to create a functionally-active subspace, whose basis not only captures the geometric variance of designs but also induces the variability in the designs' physics. As the new subspace accumulates both the functional and geometric variance, therefore, it can be exploited for efficient design exploration and the construction of improved surrogate models for designs' physics prediction. The validation and experimental studies presented in this work show the beneficial effects of the current approach in comparison to a conventional single-step feature learning.

show abstract

Deep Autoencoder for Off-Line Design-Space Dimensionality Reduction in Shape Optimization

Cited by 22 publications

References 24 publications

Stochastic Shape Optimization via Design-Space Augmented Dimensionality Reduction and RANS Computations

Stochastic Shape Optimization via Design-Space Augmented Dimensionality Reduction and RANS Computations

Synthesizing Designs With Inter-Part Dependencies Using Hierarchical Generative Adversarial Networks

Physics-Informed Feature-to-Feature Learning for Design-Space Dimensionality Reduction in Shape Optimisation

Contact Info

Product

Resources

About