🛡️ WeldCorp AI Inspector

Automated Quality Assurance for Welding using Computer Vision. A student project for Scaling WeldCorp.

📝 Project Overview

WeldCorp AI Inspector is a web-based tool designed to assist technicians in identifying welding defects automatically. By leveraging a Neural Network trained on the Orange Data Mining platform, this application allows users to upload photos of welding seams and receive an instant "Good" or "Bad" classification along with a confidence score.

This project addresses the scaling challenges of WeldCorp by moving from manual inspection to AI-assisted Quality Assurance.

✨ Features

• Instant Analysis: Upload an image (JPG/PNG) and get a result in seconds.
• Computer Vision: Uses Inception v3 embeddings to "see" metallurgical details.
• Defect Detection: Classifies welds into Good Weld or Bad Weld (including defects like porosity, cracks, etc.).
• Confidence Score: Displays the model's certainty percentage.
• Web Interface: Built with Streamlit for easy deployment on any device.

⚙️ Technical Architecture

The system follows a hybrid pipeline:

1. Training (Orange ML):

   • Data: Welding Defect Object Detection Dataset (Labelled).
   • Embedding: Inception v3 (Google).
   • Classifier: Neural Network (optimized for AUC).
   • Output: A pickled model file (group-project-final.pkcls).

2. Inference (Streamlit App):

   • The Python script converts the uploaded image into a 2048-dimension vector using orange3-imageanalytics.
   • The vector is passed to the loaded .pkcls model.
   • The prediction is displayed to the user.

🚀 Installation (Run Locally)

To run this app on your own machine, follow these steps:

1. Clone the repository

git clone [https://github.com/AzNobody/WeldCorp.git](https://github.com/AzNobody/WeldCorp.git)
cd WeldCorp

2. Install dependencies

Make sure you have Python installed (Python 3.10 or 3.11 recommended).

pip install -r requirements.txt

3. Run the app

python -m streamlit run app.py

The app should open automatically in your browser at http://localhost:8501.

☁️ Deployment (Streamlit Cloud)

This app is optimized for deployment on Streamlit Community Cloud.

Configuration Files

The repository includes specific configuration files for the cloud environment:

• requirements.txt: Python libraries (Streamlit, Orange3, etc.).
• packages.txt: System dependencies (specifically libgl1 for image processing).

How to deploy

1. Push this code to GitHub.
2. Go to share.streamlit.io.
3. Select the repository and the main file app.py.
4. Click Deploy.

Note: The first deployment may take 3-5 minutes as it downloads the Inception v3 model.

📂 Project Structure

weldcorp/
├── app.py                          # Main application script
├── group-project-final.pkcls       # The trained Orange model (Neural Network)
├── requirements.txt                # Python dependencies
├── packages.txt                    # Linux system dependencies (for Streamlit Cloud)
└── README.md                       # Project documentation

👥 Authors

• Andrew Hong Wei Lee
• Axel Lindvall
• Théodor Jagler
• Arthur Foutrel
• Hyeyoon Park

---

Developed for the AI for Business course at Umeå University.