Welding Defect Dataset

Dataset Overview

Problem Statement

• Predict welding defects using ML models.
• Develop algorithms based on the provided parameters.
• Goal: assist Godrej Aerospace in producing defect‑free products.

Dataset Provided

• Parameters include: ambient temperature, welding operation temperature, humidity, voltage & current, welding speed, protective‑gas flow rate, metal composition.
• Machine data includes welder details.

Dataset Analysis

1. Parameter Correlation:

   • Significant correlation observed between current and voltage, temperature and humidity.
   • Current‑voltage relationship aligns with Ohm’s law (V = IR).
   • Positive correlation between temperature and humidity is a physical phenomenon.

2. Impact of Welding Operation Temperature:

   • High welding temperatures may cause tungsten inclusions.
   • Recommended welding temperature range: 30‑60 °C.

3. Effects of Current and Voltage:

   • High current generates more heat, increasing the likelihood of tungsten inclusions.
   • Voltage may also affect tungsten inclusions.

4. Temperature, Humidity, and Flow Rate on Porosity:

   • Temperature and humidity have limited impact on porosity.
   • Porosity flow rate is higher than that of defect‑free and tungsten‑inclusion samples.

5. Imbalanced Data Handling:

   • Applied oversampling, undersampling, and SMOTE techniques.
   • Combined the three methods to create a robust dataset.

Models Used

• Experimented with SVM, AdaBoost, Decision Tree, Random Forest, and Gradient Boosting.
• Final selection: XGBoost, achieving 96 % accuracy and strong F1 score.

Model Deployment

• Deployed with Tailwind CSS and a Flask server on AWS EC2.
• Model runs efficiently, using only 30 % of a 1 GB RAM instance; model size is 35 MB.