DATA ANALYTICS EXAMPLES

Challenges | Solutions | Results

Welding quality forecasting

Challenges

Special machine for fast welding plates connecting battery cells
Testing the welding quality via tensile test, and the results of scrap parts
Cannot test all the parts, as there would be no parts kept

Solution

  • Analysis of sensor data: force, current, tension and displacement
  • Testing Machine Learning algorithms to predict tensile strength
  • Automatic detection of scrap quality welding

Results

Trained ML model ready for test phase in production

  • 6-11% mean absolute error between the value of real and predicted tensile forces
  • 96-99% accuracy of predicting the welding quality