FedML meets WAAM: XR assistance system makes setting up additive welding processes easier

In the ITEA project FAMILIAR, the Fraunhofer Institute for Additive Production Technologies IAPT, together with consider it, Pumacy Technologies, and NXRT, presented a digital assistance system for setting up Wire Arc Additive Manufacturing (WAAM). The core of the solution is an architecture for federated machine learning that evaluates sensor data in a decentralized manner and feeds it directly back into the setup process via mixed-reality headsets. The target group is companies that want to use WAAM without having already built deep process knowledge.

Technically, the system is based on a FedML topology between edge devices in the 3D printing cell. An in-process stereo camera provides image and depth information that are used to further train the models without collecting raw data centrally. As a result, operational parameters such as wire feed rate, current, or arc length remain under the users’ control. Via XR headsets, operating personnel receive context-specific instructions, for example for correctly clamping the substrate plate or calibrating the robot path, including validation steps and plausibility checks.

Within the directed energy deposition processes, WAAM is among the productive metal AM approaches but often suffers from labor-intensive setups on the shop floor. Errors in preparation lead to distortion, rework, and scrap. The FAMILIAR subproject addresses precisely this phase: Fraunhofer IAPT was responsible for the system architecture in the cell, sensor coupling as well as data flows and interfaces, and conducted evaluations under production conditions. According to the project, the assistance solution reduces redundant work steps and halves the preparation effort; at the same time, process stability increases because setup parameters are proposed and verified based on data.

In practice, this means shorter downtimes between build jobs and traceable documentation of setup decisions. Because the learning is collaborative and decentralized, the approach is also suitable for industries with heightened confidentiality requirements. FAMILIAR thus shows how XR, edge analytics, and additive manufacturing can be combined into an integrated workflow that makes WAAM systems productive more quickly and reduces quality risks before the first layer.