Machine learning tools in production engineering

Abstract Machine learning methods have shown potential for the optimization of production processes. Due to complex relationships often inherent in those processes, success such is uncertain and unreliable. Therefore, understanding (algorithmic) behavior results machine crucial improve prediction Here, mathematical tools may help. This paper shows how efficient algorithms training neural networks their retraining framework transfer are expressed a discrete as well time-continuous formulation. The latter can be analyzed investigated using techniques from kinetic gas dynamics. obtained provide first step towards explainable artificial intelligence. Based on description, an adapted ensemble method proposed compared with backpropagation algorithms. process common task therefore demonstrated two very different one involves specific cutting forces second particle properties plasma spraying coating process. For both use cases, presented applied performance evaluated giving thereby indication mathematically inspired classical tasks