Machine Learning Based Quality Assessment Of The Feedstock Kneading Process By Computer Vision

Kneading processes are relevant in many areas of daily life and technical fields. They are the basis of the MIM process in which powders and binders are combined to the injectable feedstock. The quality of the mixing directly influences the following process steps. Inhomogeneities in the form of binder or powder nests can cause local defects like pores or induce locally differing shrinkage and thus warpage, or make the feedstock behaviour instable. The assessment of the feedstock state during mixing is a promising way to assess quality and optimize process control. We approached this goal with a first step that made use of images, generated by video recordings of the feedstock mass during mixing. An artificial intelligence network was constructed, trained, and tested to simulate the quality assessment of a human expert in two categories, “ok” and “not ok”. All images and mixing states were assessed with 100 % accuracy.