Accelerating Automotive Design Iteration with Integrated Generative Design and Predictive Performance

Votes: 0
Views: 749

Introduction:

Our innovative approach leverages machine learning to transform automotive 3D design, focusing on accelerating the product design cycle. By integrating deep learning into traditional design iterations, we provide a unified solution that significantly reduces costs and enhances efficiency.

Problem Statement:

The traditional automotive 3D design process is plagued by high costs and long lead times. Engineers face challenges in inheriting past design experiences, conducting extensive simulations, and generating high-maturity designs rapidly.

Design Overview:

Our solution integrates three key machine learning technologies to address these challenges holistically, using a suspension bracket as an example:

Enriching 3D Product Database with Deep Learning:

We use deep learning to convert pictures of existing, in-market suspension bracket designs, which are high in maturity into 3D point cloud models, enriching the 3D product database for reference.

Cost and Time Savings: Traditional 3D scans and CAD rebuilt cost $500-$1500 and take days. Deep learning NeRF reduces costs to reusable GPUs and cuts model creation time by 70%.

Generative Design with Deep Learning Diffusion:

Diffusion-based models learns learn past successful product design concept from 3D database, and generate new 3D point clouds for reference, allowing engineers to start with high-maturity designs.

Time Savings: Traditional design start from scratch requires 10-20 iterations over weeks. Our model starts with 60% design maturity, reducing the design cycle from months to weeks.

Performance Prediction Using Neural Networks:

Machine learning models trained on existing simulation from 3D product database to predict the performance of new designs (outside of the database), minimizing the need for extensive CAE simulations.

Cost and Time Savings: Full CAE simulations on a bracket cost $500-$2000 and take hours to days. Our model predicts performance in minutes, reducing time by 90%.

Unified Production Feasibility:

Our solution seamlessly integrates these machine learning technologies with existing 3D design space, using point cloud as a medium. Models are trained on available product datasets and get improved with new design in production. Cloud-based solutions ensure scalability and cost-effectiveness.

Applications:

  • Conceptualization: Rapid generation and visualization of new design ideas.
  • Performance Evaluation: Efficient prediction of design performance.
  • Target markets include automotive suppliers, designers, and engineering firms.

Benefits:

  • Cost Reduction: Enriching 3D product databases with 3D models from 2D images reduces costs of 3D model rebuilds.
  • Accelerated Design Cycle: Innovative design ideas generated by machine learning reduce development time.
  • Resource Savings: Rapid performance predictions minimize the need for extensive CAE simulations.

Conclusion:

Our machine learning-driven approach offers a unified solution to automotive 3D design challenges. By enriching the 3D product database with high-maturity models, predicting performance in minutes, and generating new design ideas, we provide a comprehensive toolset that enhances productivity, reduces costs, and accelerates the design cycle. 

Video

Like this entry?

Voting is closed!

  • About the Entrant

  • Name:
    Zongyue Liu
  • Type of entry:
    team
    Team members:
    • Chloe Xu
    • Guoliang He
    • Ziyan Xu
    • Zongyue Liu
  • Software used for this entry:
    Python, Blender, MeshLab
  • Patent status:
    none