Generative Adversarial Networks for Inverse Design Problems in Engineering: Methods to handle performance, constraints, and creativity requirements
Engineering design tasks often require synthesizing new designs that meet desired performance requirements. The conventional design process, which requires iterative optimization and performance evaluation, is slow and dependent on initial designs. Automating this process in its entirety, therefore...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Massachusetts Institute of Technology
2022
|
| Online Access: | https://hdl.handle.net/1721.1/144700 |
| _version_ | 1826198092933758976 |
|---|---|
| author | Heyrani Nobari, Amin |
| author2 | Ahmed, Faez |
| author_facet | Ahmed, Faez Heyrani Nobari, Amin |
| author_sort | Heyrani Nobari, Amin |
| collection | MIT |
| description | Engineering design tasks often require synthesizing new designs that meet desired performance requirements. The conventional design process, which requires iterative optimization and performance evaluation, is slow and dependent on initial designs. Automating this process in its entirety, therefore can speed up the design process significantly and mitigate the need for the conventional iterative and time-consuming design process. If successful in automating the engineering design process using data-driven approaches such as the ones proposed in this body of work, the impacts on humanity as a whole would be extraordinary. The engineering design process is an omnipresent aspect of our modern life, affecting our day-to-day lives in a major way. Accelerating this process through enabling inverse design and automation can reduce costs and increase productivity. In doing this, we develop data-driven approaches based on the generative adversarial networks~(GANs) to address some of the main challenges in data-driven inverse design. First, we propose a new model, named Performance Conditioned Diverse Generative Adversarial Network (PcDGAN), which introduces a singular vicinal loss combined with a Determinantal Point Processes (DPP) based loss function to enhance diversity. PcDGAN uses a new self-reinforcing score called the Lambert Log Exponential Transition Score (LLETS) for improved GAN performance on inverse design based on performance requirements. Experiments on synthetic problems and a real-world airfoil design problem demonstrate that PcDGAN outperforms state-of-the-art GAN models and improves the likelihood of meeting performance requirements by 69\% in an airfoil generation task and up to 78\% in synthetic conditional generation tasks and achieves greater design space coverage. The proposed method enables efficient design synthesis and design space exploration, however, the problem of handling constraints remains as only taking performance into account for inverse design leaves design constraints aside and therefore, we must also address constraints. To do this, we propose a conditional deep generative model, Range-GAN, to achieve automatic design synthesis subject to range inequality constraints. The proposed model also addresses the sparse conditioning issue in data-driven inverse design problems by introducing a label-aware self-augmentation approach. We also propose a new uniformity loss to ensure the generated designs evenly cover the given requirement range. This work is the first of its kind and outperforms conventional optimization-based approaches such as genetic algorithms, specifically we compare our method to NSGA-II. Through a real-world example of constrained 3D shape generation, we show that Range-GAN outperforms state of the art methods and furthermore the label-aware self-augmentation leads to an average improvement of 14\% on the constraint satisfaction for generated 3D shapes, and the uniformity loss leads to a 125\% average increase on the uniformity of generated shapes' attributes. This work laid the foundation for data-driven inverse design problems where we consider range constraints. Finally, we must turn our attention to another aspect of the design process that is crucial, and that is creativity and novelty. The previous models demonstrate the efficacy of GAN-based models in performing inverse design tasks. GAN models, however, are not capable of generating unique designs, a key to innovation and a major gap in AI-based design automation applications. to alleviate this we propose an automated method, CreativeGAN, for generating novel designs using GANs. It does so by identifying components that make a design unique and modifying a GAN model such that it becomes more likely to generate designs with identified unique components. The method combines state-of-art novelty detection, segmentation, novelty localization, rewriting, and generative models for creative design synthesis. Using a dataset of bicycle designs, we demonstrate that the method can create new bicycle designs with unique frames and handles, and generalize rare novelties to a broad set of designs. Our automated method requires no human intervention and demonstrates a way to rethink creative design synthesis and exploration. By addressing these important challenges in data-driven inverse design we hope to enable a complete model that combines all three approaches in the future to establish an ultimate generalizable model for automating inverse design based on data. |
| first_indexed | 2024-09-23T10:58:44Z |
| format | Thesis |
| id | mit-1721.1/144700 |
| institution | Massachusetts Institute of Technology |
| last_indexed | 2024-09-23T10:58:44Z |
| publishDate | 2022 |
| publisher | Massachusetts Institute of Technology |
| record_format | dspace |
| spelling | mit-1721.1/1447002022-08-30T03:45:20Z Generative Adversarial Networks for Inverse Design Problems in Engineering: Methods to handle performance, constraints, and creativity requirements Heyrani Nobari, Amin Ahmed, Faez Massachusetts Institute of Technology. Department of Mechanical Engineering Engineering design tasks often require synthesizing new designs that meet desired performance requirements. The conventional design process, which requires iterative optimization and performance evaluation, is slow and dependent on initial designs. Automating this process in its entirety, therefore can speed up the design process significantly and mitigate the need for the conventional iterative and time-consuming design process. If successful in automating the engineering design process using data-driven approaches such as the ones proposed in this body of work, the impacts on humanity as a whole would be extraordinary. The engineering design process is an omnipresent aspect of our modern life, affecting our day-to-day lives in a major way. Accelerating this process through enabling inverse design and automation can reduce costs and increase productivity. In doing this, we develop data-driven approaches based on the generative adversarial networks~(GANs) to address some of the main challenges in data-driven inverse design. First, we propose a new model, named Performance Conditioned Diverse Generative Adversarial Network (PcDGAN), which introduces a singular vicinal loss combined with a Determinantal Point Processes (DPP) based loss function to enhance diversity. PcDGAN uses a new self-reinforcing score called the Lambert Log Exponential Transition Score (LLETS) for improved GAN performance on inverse design based on performance requirements. Experiments on synthetic problems and a real-world airfoil design problem demonstrate that PcDGAN outperforms state-of-the-art GAN models and improves the likelihood of meeting performance requirements by 69\% in an airfoil generation task and up to 78\% in synthetic conditional generation tasks and achieves greater design space coverage. The proposed method enables efficient design synthesis and design space exploration, however, the problem of handling constraints remains as only taking performance into account for inverse design leaves design constraints aside and therefore, we must also address constraints. To do this, we propose a conditional deep generative model, Range-GAN, to achieve automatic design synthesis subject to range inequality constraints. The proposed model also addresses the sparse conditioning issue in data-driven inverse design problems by introducing a label-aware self-augmentation approach. We also propose a new uniformity loss to ensure the generated designs evenly cover the given requirement range. This work is the first of its kind and outperforms conventional optimization-based approaches such as genetic algorithms, specifically we compare our method to NSGA-II. Through a real-world example of constrained 3D shape generation, we show that Range-GAN outperforms state of the art methods and furthermore the label-aware self-augmentation leads to an average improvement of 14\% on the constraint satisfaction for generated 3D shapes, and the uniformity loss leads to a 125\% average increase on the uniformity of generated shapes' attributes. This work laid the foundation for data-driven inverse design problems where we consider range constraints. Finally, we must turn our attention to another aspect of the design process that is crucial, and that is creativity and novelty. The previous models demonstrate the efficacy of GAN-based models in performing inverse design tasks. GAN models, however, are not capable of generating unique designs, a key to innovation and a major gap in AI-based design automation applications. to alleviate this we propose an automated method, CreativeGAN, for generating novel designs using GANs. It does so by identifying components that make a design unique and modifying a GAN model such that it becomes more likely to generate designs with identified unique components. The method combines state-of-art novelty detection, segmentation, novelty localization, rewriting, and generative models for creative design synthesis. Using a dataset of bicycle designs, we demonstrate that the method can create new bicycle designs with unique frames and handles, and generalize rare novelties to a broad set of designs. Our automated method requires no human intervention and demonstrates a way to rethink creative design synthesis and exploration. By addressing these important challenges in data-driven inverse design we hope to enable a complete model that combines all three approaches in the future to establish an ultimate generalizable model for automating inverse design based on data. S.M. 2022-08-29T16:05:40Z 2022-08-29T16:05:40Z 2022-05 2022-06-23T14:10:09.283Z Thesis https://hdl.handle.net/1721.1/144700 In Copyright - Educational Use Permitted Copyright MIT http://rightsstatements.org/page/InC-EDU/1.0/ application/pdf Massachusetts Institute of Technology |
| spellingShingle | Heyrani Nobari, Amin Generative Adversarial Networks for Inverse Design Problems in Engineering: Methods to handle performance, constraints, and creativity requirements |
| title | Generative Adversarial Networks for Inverse Design Problems in Engineering: Methods to handle performance, constraints, and creativity requirements |
| title_full | Generative Adversarial Networks for Inverse Design Problems in Engineering: Methods to handle performance, constraints, and creativity requirements |
| title_fullStr | Generative Adversarial Networks for Inverse Design Problems in Engineering: Methods to handle performance, constraints, and creativity requirements |
| title_full_unstemmed | Generative Adversarial Networks for Inverse Design Problems in Engineering: Methods to handle performance, constraints, and creativity requirements |
| title_short | Generative Adversarial Networks for Inverse Design Problems in Engineering: Methods to handle performance, constraints, and creativity requirements |
| title_sort | generative adversarial networks for inverse design problems in engineering methods to handle performance constraints and creativity requirements |
| url | https://hdl.handle.net/1721.1/144700 |
| work_keys_str_mv | AT heyraninobariamin generativeadversarialnetworksforinversedesignproblemsinengineeringmethodstohandleperformanceconstraintsandcreativityrequirements |