Home // ADAPTIVE 2022, The Fourteenth International Conference on Adaptive and Self-Adaptive Systems and Applications // View article

Usage of Machine Learning for Subtopology Detection in Wire and Arc Additive Manufacturing

Authors:
Dimitri Bratzel
Stefan Wittek
Andreas Rausch
Kai Treutler
Tobias Gehrling
Volker Wesling

Keywords: WAAM, Welding, GMAW, Machine Learning, Long Short-Term Memory, topology detection

Abstract:
In additive manufacturing, knowledge of the geometry of the weld seam is crucial for the quality of the component. This is especially true for Wire and Arc Additive Manufacturing (WAAM) based on Gas Metal Arc Welding (GMAW). The length of the free wire electrode ("stickout") should be almost constant during the entire manufacturing process. In additive manufacturing, it is also important to recognize height differences that occur during the process and to compensate for them by adjusting the process parameters in order to achieve a uniform build rate across the component cross-section, as geometric irregularities tend to be amplified by multiple layers. Furthermore, process disturbances can lead to locally altered seam properties. To counteract these problems, the presented investigations show to what extent such geometric irregularities can be detected in-situ from the existing process variables welding current and voltage. This makes it possible to dispense with the use of additional measurement technology. In our experiments, we simulated these height differences during multilayer welding by means of defined unevenness on the substrate plate. With the help of a Long Short Memory Neuronal Network (LSTM), the height information is determined indirectly during the process only via welding current and voltage. It is shown that this approach could be used to control the process. Furthermore, it is shown that this approach can reliably detect geometry errors and determine the height information with high accuracy, even if the process parameters are changed between training and validation.

Pages: 32 to 37

Copyright: Copyright (c) IARIA, 2022

Publication date: April 24, 2022

Published in: conference

ISSN: 2308-4146

ISBN: 978-1-61208-951-5

Location: Barcelona, Spain

Dates: from April 24, 2022 to April 28, 2022