Accelerating Process Development for 3D Printing of New Metal Alloys

David Guirguis,Conrad Tucker,Jack Beuth

2023-12-30

Abstract:Addressing the uncertainty and variability in the quality of 3D printed metals can further the wide spread use of this technology. Process mapping for new alloys is crucial for determining optimal process parameters that consistently produce acceptable printing quality. Process mapping is typically performed by conventional methods and is used for the design of experiments and ex situ characterization of printed parts. On the other hand, in situ approaches are limited because their observable features are limited and they require complex high-cost setups to obtain temperature measurements to boost accuracy. Our method relaxes these limitations by incorporating the temporal features of molten metal dynamics during laser-metal interactions using video vision transformers and high-speed imaging. Our approach can be used in existing commercial machines and can provide in situ process maps for efficient defect and variability quantification. The generalizability of the approach is demonstrated by performing cross-dataset evaluations on alloys with different compositions and intrinsic thermofluid properties.

Materials Science,Computer Vision and Pattern Recognition,Machine Learning

What problem does this paper attempt to address?

This paper aims to address the issues of quality instability and defects in the process of 3D printing metal alloys. Currently, 3D printing technology is becoming increasingly important in multiple industries, but the quality of metal 3D printing is influenced by various factors including alloy type, printing parameters, and machine settings, which limits its widespread application. Researchers propose a new in-situ method that utilizes video vision transformers and high-speed imaging to monitor the temporal characteristics of the molten pool dynamics during the interaction between laser and metal in real time, in order to improve the efficiency of defect and variability quantification.

Traditional methods rely on ex-situ experiments and static feature analysis, while in-situ methods are limited by the limited observable features and the need for expensive high-temperature measuring equipment. This study eliminates the saturation phenomenon of high-speed camera frames and utilizes advanced deep learning algorithms - video vision transformers instead of traditional convolutional neural networks, to classify different defect types in the molten pool process, such as poor bonding, spheroidization, and porosity formation. This method can be implemented on existing commercial machines, has the generalization ability to different alloys, and can generate variability process maps of molten pool properties to guide the optimization of printing parameters, reduce defects, and improve printing consistency.

The experimental results show that the accuracy of this method in different alloy datasets reaches up to 98%, and it can generate process maps that are consistent with ex-situ characterization. In addition, mapping the variability of molten pool shape changes helps in selecting process parameters that lead to higher stability of the molten pool, thereby improving the consistency of 3D printing results. Despite challenges such as data transmission rates and camera limitations, this research provides a foundation for real-time monitoring and process control, and is expected to accelerate the development of 3D printing processes for new metal alloys.