Modeling spatial variations in co-axial melt pool monitoring signals in laser powder bed fusion

The transition of additive manufacturing from rapid prototyping to large-scale manufacturing is being impeded by the limited repeatability of part quality within and across independent builds. The property variations are a result of process fluctuations, which can be random or systemic, like the location of the part on the build plate. In-situ sensors are employed to measure these process variations to monitor and control the part properties. A priori estimates of the systemic variations in the in-situ signals can enhance the efficiency of such techniques. In this work, custom builds for IN718, along with multiphysics simulations, are employed to probe systemic trends in the signals from the co-axial melt pool monitoring system on laser powder bed fusion based GE Concept Laser M2. Significant spatial variations are observed in the in-situ melt pool area, and intensity signals, and the subsequent analysis suggests that the interplay of the angled laser beam and the shield gas flow is likely a major source of the spatial trends. Machine learning is employed to develop data-driven surrogate models that make accurate predictions for this systemic variation in the melt pool area and intensity, generalizing across inde-pendent builds with significantly diverse print parameters. These models are anticipated to inform control al-gorithms and tune optimal print parameters with the objective of enhancing the repeatability of the part quality for laser powder bed fusion.

Automated summary

The limited repeatability of part quality in additive manufacturing hinders its transition to large-scale manufacturing. Process fluctuations, both random and systemic, contribute to the variations in part properties. In-situ sensors are employed to measure these fluctuations and improve efficiency by estimating systemic variations in the signals.