Development of an ML model for the classification of surface quality in a milling process

Abstract

Digitalisation is increasingly finding its way into the production process of manufacturing companies. The paper deals with the question of how manufacturing data of a milling process can be analysed using machine learning (ML) methods to classify surface defects at an early stage of production. The paper develops an ML model that classifies image data based on the surface roughness of produced parts. For this purpose, sample parts were produced on a milling machine in the learning and research factory for Industry 4.0, the Smart Production Lab at FH Joanneum, Austria. This resulted in a data set of about 38,500 images. The developed based on a convolutional neural network model achieved an accuracy of 82% in predicting surface quality. The model divides produced sample parts into quality classes with a surface roughness between Ra 0.3 and Ra 2.3. Being used in industrial processes, the developed ML model enables reliable prediction of surface quality without manual measurement and evaluation.