Camera-Based Coaxial Melt Pool Monitoring Data Registration for Laser Powder Bed Fusion Additive Manufacturing

The quality of powder bed fusion (PBF) built parts is highly correlated to the melt pool characteristics. Camera-based coaxial melt pool monitoring (MPM) is widely applied today because it provides high resolution monitoring on the time and length scales necessary for deep PBF process understanding, in-process defect detection, and real time control. The dataset resulting from the MPM system enables advanced data analytics, including deep learning to identify or validate relationships between the quality of built parts and melt pool characteristics. Different from typical machine learning applications, the problems span the measurement data on various scales and various resolutions in both time and length. Hence, a meaningful correlation or fusion for advanced data analytics requires the correct data registration of melt pool monitoring data process both to assign every image pixel of a single measurement to the right location in the part if the monitoring is camera-based and further to assign every measured melt pool characteristics to the right location of the partially finished part.

This paper presents methods for camera-based coaxial melt pool monitoring (MPM) data registration using the build volume coordinate system defined in ISO/ASTM52921. This paper has two scenarios based on both the environment of collecting data and some issues of registering data. In the first scenario, Open architecture AM systems are considered which provides synchronized scanning time stamps for the melt pool frames with using the same clock for laser beam spot position defined in the build volume coordinate system, but there is an inconsistency between scanner command and real scanner position because of the control issues of AM machine; in the second scenario, close AM systems with 3rd part MPM additions are considered which provides a camera-based coaxial melt pool monitoring data installed on commercial systems, whose Data Acquisition (DAQ) is independent of the motion and laser control of the machine. The associated laser beam spot position has to be estimated from the melt pool image characteristics.

The proposed methods utilize data sychronization and calibration techniques for solving the issue presented in Open archiecture system, likewise for solving the issue presented in the closed system, image preprocessing techniques, and machine learning to identify a melt pool in an image frame and to predict the position of the laser beam spot where the melt pool was created. Uncertainties are evaluated for the proposed methods and case studies are provided to demonstrate the effectiveness of the methods.