Metal parts manufactured using Laser Powder Bed Fusion or “L-PBF” find application in aerospace and medical industries. L-PBF is used to produce complex geometries directly from a CAD model. However, L-PBF is known to produce parts with porosity issues. To meet stringent safety and certification requirements, researchers at Flanders Make, a manufacturing science center in Lommel, Belgium, used the Mikrotron EoSens 3CL three-megapixel camera to gain a better understanding of melt-pool behavior and stability during processing.
In L-PBF, metal powder is usually spread over layers in thicknesses of 0.1 mm on a build platform and melted by the laser in a repetitive process until the part is built. Essentially, L-PBF is the combination of many individually stacked weld lines, with the overlap and penetration depth determining final part density and quality. Any defects can lead to unwanted variations in mechanical properties, reducing the part to scrap.
To detect problems, time- and cost-ineffective post production quality control techniques such as X-ray Computed Tomography are used to ensure that the final component meets the required specification. Researchers at Flanders Make sought to create a feasible, economical alternative to these costly methods to estimate the melt-pool depth and width.
The main components of the Flanders Make imaging system are the Mikrotron camera and a large area silicon photodiode sensor. Sensors are band-pass filtered in a range of 800–950 nm to prevent any stray ambient and laser light from interfering with measurements. This system is built using a custom FPGA-based frame grabber linked to a fiber coupled Network-Attached Storage. The NAS is equipped with a RAID 0 array of four 1TB Solid State Disks resulting in a sustainable data rate up to 1 GB/s and a total capacity of 4 TB. The Mikrotron EoSens 3CL camera interface is “Full” configuration Camera Link and is capable of data rates up to 850 MB/s. Balancing the availability of light, resolution and frame rate, the acquisition speed for the camera was set at 20,000 frames-per-second by the researchers. Captured images are 8-bit gray-scale images, measuring 120 x 120 pixels in size. The calibrated pixel size is 11.8 μm, resulting in a field of view of 1416 x 1416 μm. As a frame can contain more than one object due to the occurrence of spatter or break-up of the melt-pool tail, the first step in image processing was to filter out the melt-pool outline. This is done using a combination of an edge detection using a Sobel filter, and the relative location of the laser in the frame.
In system testing, researchers found that the measured melt-pool width and predicted depth corresponded well with metallographic measurements over a range of processing parameters. This ability to determine the melt-pool dimensions without the need for destructive testing is useful in determining a suitable processing parameter window. Time required to process all the data and generate process maps was less than five minutes on a desktop computer equipped with a NVIDEA QUADRO M4000. Of these five minutes about two minutes were spent on image processing, which amounts to an average of about 0.5 ms per frame. Processing times are negligible compared to the time-consuming approach for generating processing maps using standard techniques. Those techniques usually involve the manufacturing of many samples with different processing parameters, after which those samples are removed from the baseplate, cross-sectioned, polished, and then etched for microstructural analysis.
The Flanders Make system offers a solution to the challenging task of meeting stringent quality and repeatability standards typical for high-end industries using L-PBF produced parts. With further refinement researchers believe that their methodology can be applied to other additive and more general materials processing techniques, such as electron beam, as well as laser and electron beam welding/cutting.