Constraint Active Search in Process Window Optimization for Powder Feed Directed Energy Deposition

Abstract Optimizing process parameters for directed energy deposition is crucial to achieve high-quality printed parts. However, this optimization process often entails significant time and cost investments. An initial investigation into the process window can be conducted through the examination of single tracks. In this work, we investigate the utility of constraint active search (CAS) to efficiently identify process window that yield 4340 low-alloy steel single tracks with desired geometrical features. The effectiveness of the CAS method was assessed through experiments with physical and interpolated measurement. Fifty single tracks from randomly sampled process parameter combinations with different power, scan velocity, and laser spot size and ten single tracks from CAS-generated parameters were produced and analyzed. The results demonstrate that our search method outperforms random search, with 80% of parameter sets identified as desirable compared to only 4% in the case of random search. Moreover, an interpolated ground truth in input spaces of various dimensionalities was built in order to assess repeatability without excessive experimental cost. The results indicate that the CAS achieves higher precision compared to grid search and random search, especially in higher-dimensional process parameter spaces.