Development of Machine Learning prediction models for their integration in a Digital Twin for a tapered roller bearing production line

Abstract

The aim of this work is to develop the prediction models that are integrated in a digital twin for a tapered roller bearing multi-stage production line. The manufacturing process consists of rings machining and component assembly processes, including intense quality control. This work proposes the use of machine learning techniques for a 4-step strategy which consists of: a data analysis, the development of one prediction model for the manufacture of the double outer ring, one model for the two inners rings, and finally their integration in the digital twin. The strategy is validated with real data. Several regression techniques are tested and the selected model is the exponential regression due to its better performance when compared with other algorithms. Once incorporated in the digital twin, the developed models can predict the process behaviour under potential changes so determining the optimum operating conditions can be fairly facilitated; as well as predict the final bearing setting under different machining conditions.