PaDGAN: Learning to Generate High-Quality Novel Designs

Abstract

Deep generative models are proven to be a useful tool for automatic design synthesis and design space exploration. When applied in engineering design, existing generative models face three challenges: (1) generated designs lack diversity and do not cover all areas of the design space, (2) it is difficult to explicitly improve the overall performance or quality of generated designs, and (3) existing models generally do not generate novel designs, outside the domain of the training data. In this article, we simultaneously address these challenges by proposing a new determinantal point process-based loss function for probabilistic modeling of diversity and quality. With this new loss function, we develop a variant of the generative adversarial network, named “performance augmented diverse generative adversarial network” (PaDGAN), which can generate novel high-quality designs with good coverage of the design space. By using three synthetic examples and one real-world airfoil design example, we demonstrate that PaDGAN can generate diverse and high-quality designs. In comparison to a vanilla generative adversarial network, on average, it generates samples with a 28% higher mean quality score with larger diversity and without the mode collapse issue. Unlike typical generative models that usually generate new designs by interpolating within the boundary of training data, we show that PaDGAN expands the design space boundary outside the training data towards high-quality regions. The proposed method is broadly applicable to many tasks including design space exploration, design optimization, and creative solution recommendation.

PcDGAN

“…As two items and become more similar, ( ) ( ) increases and the probabilities of sets containing both and decrease. The previous model, PaDGAN, introduces a DPP loss to maximize the DPP probability [7]. This simultaneously encourages a larger coverage of the data space and high-quality sample generation.…”

“…Further, for the quality term in Eq. 10 in the airfoil example, we use a realistic conditioning quality [7] which means that the conditioning quality will be multiplied by the discriminator output, i.e. (x) = (x) ′ (x), so as to promote GAN stability.…”

Data-Driven Design-by-Analogy: State of the Art and Future Directions

“…As a comparison, various large-scale datasets collected and represented for previous DbA subprocesses also provide opportunities for scholars to explore the deep learning-based data synthesis and generative methods, which we term as a new subprocess of data-driven DbA, analogy-based design synthesis. Unlike traditional design synthesis strategies, such as shape grammars and constraint programming, data-driven design synthesis methods do not necessarily require expert knowledge and can automatically learn to generate plausible new designs from datasets [124,125,126,127,128,129,130].…”

“…GAN [111] and Variational autoencoder (VAE) [131] are the most commonly used deep generative models to assist engineering design. Specifically, variations of GAN have been used in designing airfoils [124,125], car wheels [126], bicycles [127], airplane [129] and social robots [130], and VAE has been used in designing material microstructures [128]. Recently, the pre-trained GPT [115] and DALL•E [116] models released by Open.AI have exhibited record-breaking performances in creating novel texts and images.…”

PcDGAN: A Continuous Conditional Diverse Generative Adversarial Network For Inverse Design

“…As two items 𝑖 and 𝑗 become more similar, 𝜙 (𝑖) 𝑇 𝜙 ( 𝑗) increases and the probabilities of sets containing both 𝑖 and 𝑗 decrease. The previous model, PaDGAN, introduces a DPP loss to maximize the DPP probability [7]. This simultaneously encourages a larger coverage of the data space and high-quality sample generation.…”