Closed-loop process control
This is still under development. The ALBA being installed today measures and records the entire process. The automatic correction described below is the line of work we are currently pursuing, not a feature of the machine we are delivering. We mention this because it reflects our focus and because in this sector, solutions are often announced too readily.
What is closed-loop control in LPBF
Closed-loop control means the machine measures what's happening, makes decisions, and corrects as it prints. Most LPBF machines on the market operate in open loop: the slicing software writes a file, the electronics execute it, and for the next twenty hours, no one checks if the process has deviated.
In an ALBA machine, half the work is already done: a 2D build chamber takes an image of the powder bed after each coating, and a battery of sensors reads the machine's status layer by layer. What's missing—and what we're working on—is the final step: getting that reading back into the process and correcting it automatically, within the machine itself, without cloud interference and without operator intervention.
Three things that are constantly confused
Monitoring is recording data for later review. It's useful for traceability and quality control, but it doesn't change anything during printing. Many machines advertised as having closed-loop control do just that—and that's where we stand today, plain and simple.
Closed-loop control between layers is what we're building: measuring at the end of one layer and correcting in the next. The latency would be on the order of the layer duration, a few seconds.
In-layer control, at kilohertz frequencies, over the fusion pool geometry is the goal discussed in academic literature. Currently, no commercial LPBF machine offers this capability, neither ours nor the one we are developing. When someone announces it, the right question is: what geometries have they validated it with, and what data?
What does the machine measure today?
This is indeed operational and appears in the manufacturing report for each build:
- Image of the powder bed after each coating, using a 2D build chamber
- Powder level in the hopper, using millimeter-wave radar
- Filter clogging, by differential pressure
- Oxygen in the chamber, with redundant measurement
- Pressure and temperature of the build chamber
- Temperature of the laser system and optical path
- Temperature of the overall cooling system
The four faults on which it will act
These are the four failure modes we've chosen to address first, because they're the ones that truly ruin a structure. The column on the right shows what we're developing, not what the machine is doing today.
| Failure | How it manifests | Fix in development |
|---|---|---|
| Lack of powder | A bald spot on the bed after dispensing | Increase the application time |
| Debris | Dirt on the print area; the gas knife is no longer sweeping properly | Increase the power of the recirculation pump, cross-referencing the data with the filter's differential pressure sensor |
| Streaks | Torn dispenser rubber; a precursor to brush blockage | Cancel the affected part to save the rest of the build, or notify the operator |
| Overlays | Silver areas above the powder bed | Extract their outline with a Sobel filter and subtract them from the path of the next layer |
The re-enlargement, in detail
This is the most frequent and most expensive failure, which is why it's the first one we addressed. It occurs when an area receives more energy than it can dissipate, usually on overhangs and edges: the material solidifies above the layer plane and protrudes. If this isn't corrected, the dispenser blade collides with it, the part shifts, and the entire print is lost.
The system we are developing detects this in the bed image, extracts the contour paths enclosing the affected area using a Sobel filter, and subtracts these contours from the path of the next layer. Where there was excess material, no energy is added. The overgrowth stops, and build continues. Once compensated, the software returns to normal operation.
What he won't do, already said
- It will not operate within a single layer, only from one layer to the next.
- It will operate on the four described failure modes, not on every conceivable defect.
- It will not close the loop on the molten pool geometry at kilohertz frequencies.
- It will not replace computed tomography inspection when the part requires it.
We say it this way, and with the actual situation in mind, because in this sector automatic correction is promised much more than it is demonstrated, and because a quality engineer detects an exaggeration at the first question.
If this is crucial for your project
Tell us when you talk to us and we'll tell you where it stands and what can be seen working on the test bench. And if what you need to check is how your geometry behaves with overhangs —which is where the ridges appear— bring it and we'll print it in front of you.
—
The defects this control targets, their origins, and how they are measured are explained in porosity and density. Why the part moves even when the process is running smoothly is explained in deformation and stress relief. And the difference between monitoring and correcting, explained from scratch, is explained in what is closed-loop control.