Artificial intelligence (AI) in process monitoring
In-situ monitoring, computer vision and the digital twin for zero-defect production.
1. Introduction: Blind printing is over
Metal 3D printing (particularly Laser Powder Bed Fusion - PBF) has long been a "black box" process. After hundreds of hours of printing, it was often only after depowdering or a CT scan that one knew whether the component was free of internal defects. Given the high hourly machine rates and powder costs, this was a massive economic risk.
Today, artificial intelligence (AI) paired with in-situ monitoring (process monitoring in real time) is revolutionizing quality assurance and dramatically reducing waste.
What does in-situ monitoring mean?
Sensors, high-resolution cameras and photodiodes capture thousands of images per second of the melt pool and the applied powder layer throughout the entire printing process. Every tiny laser flash is analyzed optically and thermally.
2. The flood of data: A case for neural networks
An industrial 3D printer with multiple lasers produces terabytes of sensor data per build job. It is impossible for a human to sift through this amount of data. This is where Machine Learning (ML) comes into play.
- Computer vision (camera data): Cameras monitor the powder bed after each doctoring process. An AI that has been trained with tens of thousands of images immediately detects missing powder, scratches from the coater or raised edges (warping). It stops the machine automatically before a crash occurs.
- Melt pool analysis: Photodiodes analyze the intensity and temperature of the melt pool with microsecond precision. The AI detects anomalies such as "keyholing" (melting too deep, leaving pores) or "lack of fusion" (melting too cold, insufficient connection).
3. From error detection to automatic correction
The AI in modern machines can do more than just sound an alarm (monitoring). It enables “closed-loop control”.
If the AI model detects that the component is overheating at a certain edge (because there is not enough surrounding powder there to dissipate the heat), it sends a signal to the controller in real time. The laser reduces its power for a fraction of a second at exactly this point to prevent overheating. This "feed-forward" process proactively eliminates defects before they even occur.
4. The digital twin and certification
Especially for highly regulated industries such as aviation or medical technology, AI-supported in-situ monitoring offers an invaluable advantage: the digital twin.
Instead of checking each component for cavities in a computer tomography (CT) machine after printing, which costs thousands of euros, the AI generates a 3D voxel model of the finished component from the collected melt pool data. A virtual CT scan is created (“Virtual CT”). If this model does not show any red (faulty) zones, the component can be released much more quickly.
5. Conclusion: AI as an enabler for series production
The integration of artificial intelligence transforms metal 3D printing from an error-prone art form into a highly deterministic, predictable industrial process. Machine learning models reduce the time for parameter development, prevent construction cancellations and are the key to reducing costs in additive large-scale production.