Deep Generative Model for Mechanical System Configuration Design

Yasaman Etesam1,2 Hyunmin Cheong1 Mohammadmehdi Ataei1 Pradeep Kumar Jayaraman1
1Autodesk Research, Toronto, Ontario, Canada
2School of Computing Science, Simon Fraser University, Burnaby, BC, Canada

Generative AI has made remarkable progress in addressing various design challenges. One prominent area where generative AI could bring significant value is in engineering design. In particular, selecting an optimal set of components and their interfaces to create a mechanical system that meets design requirements is one of the most challenging and time-consuming tasks for engineers.

This configuration design task is inherently challenging due to its categorical nature, multiple design requirements a solution must satisfy, and the reliance on physics simulations for evaluating potential solutions. These characteristics entail solving a combinatorial optimization problem with multiple constraints involving black-box functions.

To address this challenge, we propose a deep generative model to predict the optimal combination of components and interfaces for a given design problem. To demonstrate our approach, we solve a gear train synthesis problem by first creating a synthetic dataset using a grammar, a parts catalogue, and a physics simulator.

Listen to Podcast Explanation

GearFormer

A showcase of our approach to mechanical system configuration design

To demonstrate our approach's effectiveness, we applied our technique to gear train design synthesis. Below are examples of gear configurations automatically generated by our AI model based on specific design requirements, showcasing the model's ability to create diverse, physically valid mechanical systems.

GearFormer Architecture

A Transformer-based model for mechanical system configuration design

Model Architecture

Deep Learning Core

Transformer architecture optimized for mechanical design with adaptive loss weighting between sequence prediction and physical constraints

Physics-Aware Design

Incorporates mechanical constraints and physical requirements through specialized loss functions and Dymos-based simulation

Domain Specific Language

Incorporates a formal grammar and lexicon for mechanical systems, enabling structured generation of physically valid gear configurations

Hybrid Transformer + MCTS

Combining the power of Monte Carlo Tree Search with GearFormer

MCTS Integration

Strategic Exploration

MCTS explores the vast design space by focusing on the most promising initial component choices

AI-Powered Completion

GearFormer leverages learned patterns to complete configurations with high success rates

Superior Performance

The hybrid approach achieves better results than either method alone, with significantly fewer iterations

Demo - Generation

Interactive UI for rapid mechanical system design exploration

Intuitive interfaces can be designed on top of such systems to specify requirements and explore diverse gear configurations. The AI-powered system generates physically valid designs in real-time, allowing quick iteration and exploration of the design space. This dramatically accelerates the mechanical system design process while ensuring all generated solutions meet the specified constraints.

Demo - Copilot

AI-powered design assistance with GearFormer

GearFormer's transformer model powers an intelligent copilot that assists designers during the configuration process. The system provides smart autocompletion suggestions, predicting optimal component combinations that satisfy both the partial design and physical constraints. This creates a fluid, collaborative design experience between human expertise and AI assistance.