Challenges | Solutions | Results

Tool failure detection with convolutional neural network


Bladder micro leakage causes pinholes on the inside
Because of immense heat, tire must be rested  at least 1 hour before manual check
This results in 5+ scrap tire production
Computer vision is not possible due to black homogeneous material of rubber
Moisture sensor is reliable, but cannot be used, because its replacement is costly


Binary classification model to identify leakage cycles:

  • Data preparation: numerous statistics which can describe well the given curves of a cycle
  • Image transformation: create images from tabular data in a way, where similar rows will get similar images
  • CNN: use convolutional neural network learning for classifying the images


The model is installed, currently we are testing in production.
Scrap production reduction.