CAD Feature Detection

A minimal app that converts STEP files to Three.js via pythonOCC and detects the manufacturing features with UV-Net. The model was trained on the MFCAD dataset, so the classification is a bit wonky. But it relays the concept.

This project also makes use of the occwl wrapper around PythonOCC to handle the mapping from BREP to graph representation.

A live demo can be found here

Tech Stack Info:

• Frontend

  • React
  • TypeScript
  • React-Three-Fiber
  • Tailwind CSS

• Backend

  • Python
  • FastAPI
  • PythonOCC
  • PyTorch

• Infra

  • CDK
  • Lambda
  • SageMaker
  • Docker

How to use

Frontend

• Run the frontend locally: cd frontend && npm run start

• Create a production build: cd frontend && npm run build

• Config:

  • Create an .env file in the frontend folder with the following fields:

    #to set the paths properly the assets folder
    PUBLIC_URL="https://yoururl.com or localhost:3000"
    
    # URL for your backend
    REACT_APP_API_URL="https://api.yoururl.com or localhost:8080"
    

Backend

• Run the backend locally: cd backend && uvicorn app.main:app --reload

• Config:

  • Create an .env in the backend/app folder with the following fields:

    ENDPOINT_NAME="name-of-sagemaker-endpoint"
    

SageMaker

• A trained UVNet model is included in the repo. To deploy the model to sagemaker, install the requirements.txt file in the feature_detector folder and then run the notebook. You must also create an .env file in the feature_detector folder with the following fields populated:

  SAGEMAKER_EXECUTION_ROLE="sagemaker execution role"
  SAGEMAKER_S3_BUCKET="sagemaker s3 bucket to store model.tar.gz file"
  

Infra

• Create an .env file in the infra folder for CDK:

  DOMAIN="yourdomain.com"
  APP_NAME="NameForYourAppInCDK"
  API_SUBDOMAIN="sub.domain.for.api"
  FE_SUBDOMAIN="fe.subdomain"
  FE_BUCKET_NAME="name-for-s3-bucket-for-fe"