Lawrence Livermore National Laboratory

Although the powder bed fusion process is conceptually simple, the underlying physics is complex and covers a broad range of time and length scales. Laser beams and powder layer thicknesses are ~10s of µm and laser speeds are ~1 m/s. On the other hand, parts are many cubic centimeters in dimension and build times can be hours, days, even weeks. Further, the process involves around 130 parameters that could affect the quality of the final part. (Yadroitsev: 2009) Parameters such as the laser power, speed, and beam size control the length, width, and depth of the melt pool. The geometry of the melt pool is important as its width and depth can affect part density and length can affect the microstructure through the cooling rate.

Generally speaking, it is desirable to maintain a constant or controlled melt pool geometry during a build. However, because the thermal boundary conditions change as a function of the part geometry, the parameters required to achieve desired melt pool characteristics will also be a function of geometry. In current powder bed fusion systems, geometry-specific parameters can be entered for geometries such as the core, skin, and downward-facing surfaces. But, achieving controlled melt pool characteristics throughout a part requires voxel-by-voxel control of the parameters. In situ sensors and feedback schemes aid such control. (Kruth et al.: 2008; Craeghs et al.: 2010; Craeghs et al.: 2012) Feedback works best when the parameters are close to the optimal for the given geometry. This is particularly the case for the high laser speeds involved in metal powder bed fusion where the time constant for the response of the melt pool to changes in power or speed can be relatively slow.

Achieving optimized input parameters is referred to as a priori (Clijsters et al.: 2012) or "intelligent feed forward" (Bourell et al.: 2009; Frazier: 2014) control. One system manufacturer is implementing a geometry-dependent scanning (or exposure) strategy. (Realizer: 2014) Modeling and simulation combined with high-performance computing optimization (solving the inverse problem) have the potential to provide the next step in such voxel-by-voxel control of the process.


Challenges addressed:

  • Accelerates process optimization to achieve desired properties and performance
  • Reduces or eliminates the need to "lock down" the process for the duration of a production cycle (establishing machine equivalence)
  • Achieving "right every time" production
  • Establishing a "digital thread" to accommodate and leverage the large amounts of data (including in situ sensor data) that come with the AM process
Contact: Ibo Matthews

Click for Intelligent feed forward

Subscribe to ACAMM mailing list: