Vibrations have a significant influence on quality and costs in metal
cutting processes. Existing methods for predicting vibrations in machine tools enable an informed choice of process settings, however they rely on costly equipment and specialised staff. Therefore, this contribution proposes to reduce the modelling effort required by using machine learning based on data gathered during production. The approach relies on two sub-models, representing the machine structure and machining process respectively. A method is proposed for initialising and updating the models in production.