Transfer learning of neural network based process models in Direct Metal Deposition
Daniel Knüttel, Stefano Baraldo, Anna Valente, Konrad Wegener, Emanuele Carpanzano
Abstract
Direct Metal Deposition is an effective additive manufacturing technology to build parts from scratch, add complex features on existing raw parts or even repair highly specialized products. The combinations of high performance materials and free shape complexity achievable by such technology offers great advantages for industrial applications such as the aeronautic industry, where lightweight parts are crucial. However, the complexity of the process is difficult to control, resulting in more scrap and rework. To reduce these drawbacks, predictive models are developed to improve the process understanding and compute expected part properties before actual manufacturing, but the generation of such models is often time expensive and requires many experiments. The present work provides a machine learning based method to facilitate the model generation process and demonstrates the transferability to different production machines. Such a methodology is of relevance to support the effective and efficient modelling of different machine tools based on the same manufacturing processes.