Speakers
Description
Artificial intelligence (AI) can enable adaptive, in-process quality assurance in arc-based manufacturing and wire-arc directed energy deposition (wire-arc DED). Industrial deployment lags behind. Machine environments are heterogeneous, sensor systems are proprietary, and transferable concepts for feeding process data into AI-based monitoring are missing. AI already works for isolated welding and deposition tasks; what is lacking are frameworks that integrate sensors independently of the specific hardware and scale across plants. This work presents a modular framework for AI-enabled process monitoring. Robotic gas metal arc welding (GMAW) serves as the demonstrator, a process that directly underlies wire-arc DED. The framework combines internal machine signals (electrical signals such as welding current and voltage, wire feed speed, shielding-gas flow rate) with external sensor data such as airborne arc sound (microphone-based). Machine-internal signals are read via a standardized OPC UA interface, while the high-frequency external signals are captured by a dedicated data-acquisition system and aligned through a common time base in the unified data architecture. A spatiotemporal representation maps all signals onto common spatial and temporal coordinates, so process states from different sources can be linked and interpreted consistently and further sensors can be added later without redesigning the architecture. Industrial requirements and implementation barriers were identified in structured interviews with welding experts at an automotive manufacturer, covering the operational, engineering, planning, and strategic-management levels. The framework was validated in a robotic GMAW cell using synchronized electrical and acoustic measurements. The signals were transformed into image representations, which a 2D convolutional neural network classified by nozzle-to-workpiece distance. The classifier reached 96.0% accuracy on electrical-signal images and 93.2% on acoustic images. Nozzle-to-workpiece distance affects arc stability and heat input; in wire-arc DED it influences layer-height stability. Fusing both signal types gives a more robust basis for process assessment. The framework is sensor-agnostic and designed to transfer to other cells and arc-based processes, showing how standardized data architectures and spatiotemporal process representation make AI-based quality assurance scalable across welding and metal additive manufacturing.
| Speaker Country | Deutschland |
|---|---|
| Would you like to publish your paper in the special issue of BHM "Berg- und Hüttenmännische Monatshefte" | No |