AI-based spatially resolved parameter prediction in laser metal deposition for increased process stability

Abstract

In the additive manufacturing of components using powder-based laser metal deposition (LMD), the heating of the volume during buildup is a decisive factor for process stability and contour accuracy. If the process parameters remain constant, this intrinsic heating leads to deviations in the deposited layer thickness during the process because of changes in the melt pool volume. This leads to contour deviations and potentially causes the process to fail if the process parameters are no longer within the suitable range. Particularly in the case of complex geometries, this previously required time-consuming process development for adapted process parameters and buildup strategies. This paper examines the potential of data-driven approaches to enhance the stability and precision of LMD processes. To this end, a machine learning (ML) model is employed to optimize laser power settings. The objective of the study is to reduce the thermally induced geometric deviations that often occur during the LMD process by utilizing experimental data. The methodology employs the use of the alloy Inconel 718, renowned for its high strength and temperature resistance, in conjunction with the utilization of computer numerical control machines equipped with laser and imaging systems. The ML model is trained to predict the optimal laser power required to obtain consistent melt pool properties. The results demonstrate that the ML approach is an effective means of reducing geometric deviations.