4

Dimensional Accuracy, Part Quality, and Process Stability in Additive Manufacturing

The second panel was moderated by planning committee member Ade Makinde, VulcanForms, Inc.

IN-PROCESS QUALITY IMPROVEMENT

Jianjun Shi, Georgia Institute of Technology, spoke about using three-dimensional (3D) point cloud data modeling, analysis, and control for the quality improvement and qualification of additively manufactured parts. The quality and functionality of a finished part depend on both the printing process and the properties of the materials being used, Shi said, and there are generally many different sensing capabilities that provide information about both process variables and the quality and functionality of the finished product. A key way to monitor and improve part quality is through in-process quality improvement (IPQI), or by using real-time data about the additive manufacturing (AM) processes and the finished parts, combined with root cause diagnosis, automatic compensation, and control.

To illustrate how this works, Shi offered three examples. Functional qualification via physical and digital twins uses a simulation of a part (the part’s digital twin) to digitally evaluate its performance along with physical testing of the part to functionally evaluate the part’s performance. The audit of the two results can be used for functional qualification. A related approach is 3D profile evolution modeling and control used to achieve high-precision AM via dynamic modeling of 3D profiles of each layer and

compensation control. The third example was the modeling, optimization, and control of dynamic 3D shape-morphing behavior in 4D printing, where the fourth dimension refers to stimuli-responsive metamaterials.

ADDITIVELY MANUFACTURED METAL PARTS FOR EXTREME ENVIRONMENTS

Anthony Rollett, Carnegie Mellon University, described his work using AM to build components to withstand high-temperature, high-pressure environments, such as solar-thermal receivers or sodium-to-molten salt heat exchangers for use in nuclear reactors. This work, he said, has shown that AM can be used to create components that will operate in temperatures up to 850°C and pressures of up to 200 bars.

When using laser powder bed fusion (LPBF) to create high-performance parts, Rollett said, one must avoid defects in the parts as much as possible, which in turn requires operating inside the appropriate process window. His group developed a process map for the alloy Haynes 282, which depended not just on the material but also on the machine that is processing it. Rollett’s team studied the various mechanical properties of Haynes 282, such as its high-temperature strength, as well as the relationship between the manufacturing process and the part properties. Rollett showed how the defect content of the material controlled its fatigue behavior.

He closed with guidance: Use knowledge of process wherever appropriate to bound the range of properties based on the physics. Identify critical process inputs and variables in terms of their effect on key process outcomes. Apply aleatoric variability/uncertainty only when needed to quantify variability of the outcomes. And build through-process models such as numerical digital twins to integrate all sources of uncertainty.

A PERSPECTIVE ON ADDITIVE MANUFACTURING

Jesse Boyer, Pratt & Whitney, discussed recent AM activities at that company. Recent devices made there with AM include an engine with a 99 percent reduction in part count and an engine oil system that took 27 days from concept to delivery. RTX, which owns Pratt & Whitney, is focusing on LPBF as well as directed energy deposition, Boyer said. RTX works with a wide range of alloys, although Pratt & Whitney typically focuses on high-strength nickel alloys and titanium alloys.

Pratt & Whitney has been struggling with variation in the material properties of some of its additively manufactured parts, Boyer said, which happens when there is not a standardized process for controlling the relevant variables. He notes that the additive process has many sources of variability, from the machine, the process variation, the feedstock, and

so on. To combat this issue, Pratt & Whitney is using what it calls the Q³⁺ framework, which involves three types of qualification: qualifying the installation of the machine; qualifying the operation of the machine and also the material’s performance; and qualifying the interaction of the material, process, and part(s).

DISCUSSION

In the discussion, Shi was asked how the physical twin differs from the object of interest. In the audit study he discussed, the physical twin is designed with a similar structure but is much simpler so that it can capture properties of interest without all the complexity of the object of interest. Machine learning (ML) models are used in conjunction with the physical twins to improve the analysis.

In response to a question about the sort of training that Pratt & Whitney is using to better train its workforce for AM technologies, Boyer said that across RTX there is a “four-legged curriculum” consisting of operator training, materials training, design training, and management training. They focus on core competencies and will offer hands-on training on various AM materials, from plastics to polymers and on to metals.

Rollett said that it is crucial to pay close attention to the feedstocks used in AM because slight variations in the compositions of the feedstock can lead to different properties in the AM product, and he offered two specific examples of when AM products made with the same process but different batches of powder ended up with different properties.

On the topic of degradation of the properties of materials and products made with AM, Rollett noted that it would be useful to have a model that could predict microstructure, but there is not yet much known about degradation over time. Boyer emphasized the role of post-processing to homogenize microstructures so that they had more consistent properties over time. More generally, he added, a great deal of work remains to be done on the relative contributions of the AM process and post-processing to the finished properties of a product.

On the topic of the best way to understand defects in AM products, Rollett emphasized the importance of physics-based modeling. Shi agreed but said that it is important to combine physics-based models with data-driven models, which can provide different and often complementary insights. Rollett added that because there are so many variables in a model of an AM process, it is crucial to identify the key variables for each particular process so that model-based computations can be made in a reasonable amount of time and for reasonable cost.