skip to Main Content

ML models for quality control

Nowadays, companies collect data in large quantities in order to be able to analyze events retrospectively. However, in the time it takes to derive improvements from the findings and implement them, the efficiency of the company suffers. For this reason, fast analysis methods are needed, where predictions of results can be made using current data, which run through specially developed models. These advanced analytics using machine learning (ML) are the cornerstone for better and faster decisions as well as higher efficiency.

Use cases

Quality control of welding electrodes in real time using ML

  • High complexity of the production process with many influencing factors
  • A wide variety of quality problems with the electrodes
  • High reject and thus waste rate
  • Development of an ML model for electrode quality control.
  • Development of a software tool for the implementation of the model
  • Implementation of the prototype in the industry
  • Computer as well as sensors and cameras (Jetson Nano, HR grayscale camera, B&R sensor technology)
  • Software tools and libraries (Python IDE, Tensorflow/Keras, OpenCV, Mapp Vision)
  • Building, programming and testing of the prototype, which detects the errors

ML models for quality control in the milling and sawing process

  • Need to detect defective surfaces early in the production process
  • Read in sensor and control data of the machine
  • Development of a ML model for predicting the surface finish of sawn metal components.
  • Development of a ML model for the quality classification of milled components
  • Behringer band saw LPS60-T, EMCO milling machine Concept Mill 105
  • MarSurf PS 10 C2 mobile surface tester

Approach to make ML models interpretable

  • Increasing number of applications of ML in the industrial context
  • Lack of transparency and interpretability of ML models as well as their predictions.
  • Approach to the interpretation of ML models for image classification
  • Using the approach to interpret optical quality control using ML in an industrial context.
  • Local Interpretable Model-agnostic Explanations (LIME) method.
  • Shapley Additive Explanations (SHAP) Method
  • TensorFlow for Convolutional Neural Networking modeling.
Back To Top
Close search