5

Dimensional Accuracy, Part Quality, and Process Stability in Post-Additive Processes

The workshop’s third panel, moderated by planning committee member Teresa Clement, Raytheon, dealt with post-additive processes, with a specific focus on dimensional accuracy, part quality, and process stability.

QUALITY CONTROL IN ADDITIVE MANUFACTURING

Zhimin Xi, Rutgers University, spoke about modeling uncertainty in additive manufacturing (AM) and its relationship with validation of AM processes and parts. Noting that the quality and reliability of a finished AM part are influenced by more than 100 parameters, he said that there is inevitably some randomness added to the process, which makes control of the final product’s quality very difficult.

To characterize uncertainty, one can use statistical methods such as maximum likelihood estimate–based methods or Bayesian-based methods to produce distributions of various factors, such as wall thickness, and then compare those distributions with the actual data to determine their agreement. A second important tool in studying uncertainty is the use of simulation models.

Once such a model has been validated against experimental data, one has a digital twin—that is, a valid digital representation of the experiment. This in turn makes it possible to calibrate the uncertainty from the machine, powder, and process, as well as to identify optimal control actions in real time.

Xi offered three key takeaways: (1) AM quality control relies on a digital twin model to reduce the influence of uncontrollable process noise; (2) the development of the digital twin model depends on simulation models, machine learning (ML) models, statistical methods, and experimental data; and (3) optimal resource allocation for developing the digital twin model is critical to saving costs and time.

INTEGRATED COMPUTATIONAL MATERIALS ENGINEERING

Tanner Kirk, QuesTek Innovations, spoke about how integrated computational materials engineering (ICME) can accelerate and inform the qualification of materials and processes used in AM. He defined ICME as a method of developing products that uses evaluation, simulation, prediction, and optimization of performance and links material models across multiple domains, such as multiple length scales or time scales.

Kirk highlighted various techniques used to predict the properties of AM materials, including computational use of CALculation of PHAse Diagrams (CALPHAD) microstructure predictions, various physics-based models, grain structure modeling, and AM process simulation. To use these techniques to accelerate and inform the qualification process, the company employs three basic methods: specification propagation, sensitivity analysis, and accelerated insertion of materials (AIM) analysis. Specification propagation relies on part composition and process (and post-processing) information to predict such things as the expected distribution of properties of the finished part or the distribution of intermediate variables. Sensitivity analysis is used to answer the question, “Where should a specification be refined?” AIM qualification is used to reduce the number of experiments needed for qualification confidence.

He closed by emphasizing the importance of standardizing AM data.

METROLOGY TECHNIQUES FOR ADDITIVE MANUFACTURING

The next speaker, Ping Guo, Northwestern University, discussed three metrology techniques that can potentially be used to make measurements during both the processing and post-processing stages of AM: hyperspatial imaging, photometric stereo, and digital fringe projection. These techniques are applicable not just to laser powder bed fusion (LPBF) processing but also to such things as directed energy deposition, wire-arc, and additive friction stir deposition. At this point, they are still under development, but Guo said he believes that they will provide better information about the physics of AM processes than other techniques currently in use, and better measurements should lead to better validation and simpler control algorithms.

The current version of hyperspatial imaging uses a low-cost hyper-spectral imaging camera designed for a drone that provides images at 24 different frequencies ranging from 665 to 960 nanometers. It could be particularly useful, he suggested, in measuring melt pool temperatures, as it would, theoretically at least, be more accurate than the infrared cameras currently in use.

The other two types of measurements—photometric stereo and digital fringe projection—can take 3D snapshots of materials as they are being processed. Photometric stereo uses information from light reflected off an object to reconstruct that object in three dimensions. Initial results indicate that the images could be processed very quickly, providing information on, for example, surface texture in near real time. And digital fringe projection, currently under development, could provide information on interlayer defects with micron-level accuracy.

ADDITIVELY MANUFACTURED PARTS AS PREFORMS FOR THE POST-ADDITIVE PROCESS

Vlastimil Kunc, Oak Ridge National Laboratory, began by asking what “quality” means. “In my opinion,” he said, “it is an answer to a simple question: Does this additively manufactured part meet my requirements?” In his experience, he continued, the answer is usually no, because most parts require post-processing techniques to get them to meet the requirements. The parts are almost never used as printed.

In that case, he continued, the additively manufactured part is actually just a preform for post-additive processes, of which there may be many. “What comes after additive manufacturing is, in my experience, actually more costly, more time consuming than the additive process itself.”

In his work in polymer additives, Kunc said that the particular type of AM that he is most likely to use is extrusion deposition of thermoplastic polymers. The process uses relatively little energy, is useful for printing parts of all sizes, and prints parts relatively quickly. Post-processing techniques are used to make modifications, such as adjusting the shape, smoothing the surface, or removing voids.

After going into details on the manufacturing and post-processing of specific items, Kunc closed by offering three conclusions: (1) additively manufactured parts can serve as preforms for the post-additive process; (2) the design of the additive part must then account for the post-additive process; and (3) effective integration of the additive process into manufacturing workflow is key to productivity.

DISCUSSION

The panel was asked how one can deal with the “curse of dimensionality”—that is, the way that the amount of data explodes exponentially as one takes on data with more and more parameters. Xi said that one can use a technique called dimensional reduction, which involves paying attention to the dimensions that one cares about and ignoring the others. For instance, if one cares about processes, the idea is to pay attention to the key parameters that influence the processes of interest. In response to a follow-up question focusing specifically on the need to be able to make predictions in near real time as a process is occurring, Xi said that one can use a validated digital twin instead, although one still must make trade-offs between the uncertainties in the digital twin and the time and cost involved in validating that digital twin.

In response to a related but more specific question, Xi spoke about controlling melt pool parameters in real time. In effect, what he does is monitor one such parameter, the melt pool width, and then use a validated model to predict the other parameters of interest, such as the melt pool depth, based on the processing conditions. That allows him to, in effect, monitor and control multiple parameters of interest much more quickly and economically by employing the digital twin.

Kirk added that his group uses a model to predict melt pool parameters as well, but it is a very simplified model because many of the specifics are not particularly important in what they do.

In response to a question about the post-processing of polymer parts, Kunc said that they see a great deal of variability in the feedstock that they receive. Even from a single supplier, there is variation from lot to lot and sometimes even from the top to the bottom of a container. Often, he said, the quality of the material depends simply on how well the last worker to handle the feedstock at the supplier did their job in screening the material. Fortunately, he added, his team now has the tools to examine feedstock when it arrives and determine whether it is of the proper chemical and physical composition to be used.

Concerning the recycling of materials in an ICME model, Kirk said that with the ICME model, it is generally possible to recycle materials, providing one does a good job of characterizing the materials being recycled so that users know the properties of the recycled materials.

Kunc added that recycling is particularly important with polymers, as there tend to be significant amounts of leftover materials after post-processing. Returning the processed materials to a form that can be used as feedstock is generally possible, although the molecular weight is typically decreased. But if the recycled material is properly characterized, it can be used again in AM processes. It does add uncertainty to the overall process, however.

In response to a question about how one might take post-processing capabilities into account when designing an AM process, Kunc said that it is generally a good idea to begin one’s design from inspection. Organizations that do not take post-processing into account in their design are not likely to be very successful in the long run, he said. However, he added, it can be difficult to do so because most design tools are intended for a single process, not a workflow, and carrying data from one simulation to the next is not always easy.

Kirk said that his group has developed an entire workflow analysis that does indeed make it possible to take into account post-processing steps in combination with the AM processes and that such a total analysis has made it possible to improve the overall quality of the finished components, such as a turbine blade.

Xi emphasized that such a combined approach can make it possible to design the AM processes in order to minimize the necessary post-processing.

In response to a question about what the goals of post-processing should be, Kirk said that performance is the ultimate goal, and what exactly is being optimized about the processing to achieve that goal can change greatly from material to material and part to part.

Another questioner commented that in the early days of AM, one of the goals was to be able to eliminate post-processing: Is that still a goal, or should people adjust their expectations to accept that some post-processing is always likely to be needed? Kirk said that in some metals systems, it should be possible to reach that goal—that the proper microstructures can be achieved without post-processing. But for the majority of metals systems, some post-processing will likely always be necessary to get the desired properties. Xi added that whether one needs post-processing is partly dependent on the application. In surgical implants, for instance, it is important that the materials be able to function effectively in the body for as long as possible, so post-processing is necessary to achieve the optimal material characteristics, but there are other applications where it is not so important to have a specific microstructure and one can use AM parts without post-processing.

Guo said that the goal should still be to reach a point where post-processing is not necessary. While there may be some sensitive applications that will always require some post-processing to achieve exactly the right structure and properties, if additive manufacturing is to become a practical alternative to traditional manufacturing, it will be important to remove the post-processing step. Planning committee member Melissa Orme responded by saying that there will always be post-processing if we want the part to be useful. Kunc added that the ultimate goal is to make manufacturing faster and cheaper and that if the overall AM process,

including post-processing, is faster and cheaper than the traditional process for making a particular part, then it should not matter how much post-processing is still necessary. It is the big picture that is important, not the individual steps.

Speaking about bottlenecks in the AM process and post-processing, Kunc identified two. The first is all of the work done with the very first part of its kind to optimize the design and the various processes. This can take a long time, he said, but the second part is faster, and by the third part, things are moving quickly. A second bottleneck that is applicable to polymer parts is the time it takes an AM part to cool down and cure before it can go into post-processing.