Speaker
Description
Metal Additive Manufacturing (MAM) via Laser Powder Bed Fusion (L-PBF) has become an established production technology, ranging from rapid prototyping to the serial production of complex components. However, the increasing diversity of machines, materials, and process configurations creates significant challenges for quality assurance and process stability.
To achieve First-Time-Right or Zero-Defect Manufacturing, reliable in-situ monitoring is required to detect anomalies before defects propagate through the build. This work presents a multimodal, machine-independent AI-based monitoring system that combines external sensing, machine data, and knowledge from materials science and manufacturing processes.
The system utilizes layer-wise information, including optical images and machine data, to detect and classify process anomalies in real time. Once identified, defects can be reported immediately, enabling corrective actions that may prevent the loss of parts or entire print jobs.
The approach was developed and validated using aluminium and titanium alloys. To ensure multimodality and higher prediction accuracy, the framework relies on external camera systems and standardized process data acquisition. Training data were generated progressively, starting with simple test cubes containing intentionally induced defects, followed by benchmark artifacts and finally complex drone components. The AI architecture processes stage 1 (powder before melting) and stage 2 (after melting) images of each layer in parallel, fusing their representations before classification rather than treating them as independent predictions. The resulting system can detect and classify multiple defect categories, including porosity, layer misalignment, and other powder-bed anomalies. Trained across two material classes, the framework demonstrates transferability to new machines and materials with only limited retraining requirements.
The presented work represents a step toward industrial AI for additive manufacturing, combining advanced AI methods with process expertise to create robust, transferable, and easily deployable quality-monitoring solutions for agile manufacturing environments.
| Speaker Country | Austria |
|---|---|
| Would you like to publish your paper in the special issue of BHM "Berg- und Hüttenmännische Monatshefte" | No |