Shape related constraints aware generation of Mechanical Designs through Deep Convolutional GAN

Abstract

Mechanical product engineering often must comply with manufacturing or geometric constraints related to the shaping process. Mechanical design hence should rely on robust and fast tools to explore complex shapes, typically for design for additive manufacturing (DfAM). Topology optimization is such a powerful tool, yet integrating geometric constraints (shape-related) into it is hard. In this work, we leverage machine learning capability to handle complex geometric and spatial correlations to integrate into the mechanical design process geometry-related constraints at the conceptual level. More precisely, we explore the generative capabilities of recent Deep Learning architectures to enhance mechanical designs, typically for additive manufacturing. In this work, we build a generative Deep-Learning-based approach of topology optimization integrating mechanical conditions in addition to one typical manufacturing condition (the complexity of a design i.e. a geometrical condition). The approach is a dual-discriminator GAN: a generator that takes as input the mechanical and geometrical conditions and outputs a 2D structure and two discriminators, one to ensure that the generated structure follows the mechanical constraints and the other to assess the geometrical constraint. We also explore the generation of designs with a non-uniform material distribution and show promising results. Finally, We evaluate the generated designs with an objective evaluation of all wanted aspects: the mechanical as well as the geometrical constraints.

arXiv comment"submitted to the Engineering Applications of Artificial Intelligence journal"

References in corpus (7)

• Conditional Generative Adversarial Nets
• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• Road Extraction by Deep Residual U-Net
• Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversarial Modeling
• Progressive Growing of GANs for Improved Quality, Stability, and Variation
• Topology optimization of 2D structures with nonlinearities using deep learning
• A Novel Topology Optimization Approach using Conditional Deep Learning