The fourth session of the workshop included presentations on accelerating product and process qualification and certification in additive manufacturing (AM). Paolo Gennaro (GF Precicast Additive SA), Adhish Majmudar on behalf of Michel Delanaye (GeonX), Vincent Paquit (Oak Ridge National Laboratory), Jens Telgkamp (Airbus Operations GmbH), David Teter (Los Alamos National Laboratory), and Richard Ricker (National Institute of Standards and Technology [NIST]) each discussed research, challenges, and future directions relating to the following questions:
- How can each part be built to be identical and conformant, within standard tolerances and without individual inspections?
- What new standards, methods, or techniques need to be developed to certify a part built with AM?
Paolo Gennaro, GF Precicast Additive SA
Gennaro introduced GF Precicast Additive SA, including its three large divisions: GF Piping Systems, GF Casting Solutions, and GF Machining Solutions. GF Precicast Additive SA was founded in November 2016 and focuses on electron beam melting AM methods for titanium aluminide and titanium Ti-6Al-4V; direct metal laser sintering for nickel, cobalt,
or other superalloys; and cladding for industrial materials (which is still in development). GF Precicast Additive SA has a fully certified supply chain, including the AM build, heat treatment and hot isostatic pressing, and finishing quality inspection. Gennaro discussed important steps for system qualification, process qualification, and part validation, as highlighted in Table 5.1.
One of the main advantages for improved qualification and validation would be the reduction of cost, he explained. He speculated that powders might cost less if there were only one stock for one material. An audience member noted that having a single stock might be a good short-term goal, but specific applications might need a larger suite of materials in the future. Another participant added that it depends on the company as well, since each company has a different philosophy on how it designs pieces. Gennaro responded by saying that since the final products will be similar, the standards should also be somewhat similar.
The Asset Management Standards from the International Organization for Standardization (AMS-ISO) could help establish customer standards
TABLE 5.1 Key Milestones, Standards, and Advantages for System Qualification, Process Qualification, and Part Validation
|Task||Milestones||Customer standards referring to AMS-ISO||Advantages|
||Related to AMS-ISO||A single machine qualification is valid for all customers|
||Acceptance criteria (X-ray, fluorescent penetrant inspection, microstructure) referring to AMS-ISO||No discussion on quality escapes|
NOTE: AMS, Asset Management Standards; ISO, International Organization for Standardization.
SOURCE: Paolo Gennaro, GF Precicast Additive SA, presentation to the workshop, October 25, 2018.
for machine qualifications. Although the milestones described in Table 5.1 are difficult to implement, doing so could help deliver a cost-effective AM process qualification and technological validation. In response to a question on in-situ monitoring, Gennaro emphasized that the goal is to complete products correctly the first time. In-situ monitoring reduces the time to fix problems that can affect the quality of a product. Makinde also mentioned that modeling and sensing could help with calibration.
Adhish Majmudar, GeonX (presenting on behalf of Michel Delanaye)
Majmudar began by referring back to Ade Makinde’s presentation describing GE Additive. All of the departments focus on a vision of creating a part correctly the first time. He gave an example of a part that had problems with manufacturing, including collision, shrinkage lines, and surface defects such as small cracks. Time is lost if a part fails, and it ended up taking 24 hours to make the part from the powder. Simulations of powder-bed fusion AM are needed to address these problems, but modeling challenges remain.
Majmudar discussed how GE goes from micron-scale to part-level simulations to help design a part. GE provides a workflow to its clients where they start from a particle-bed or single-track simulation in order to look at the melt pool. Then, that information is fed into a model at a track level, which feeds into the macro-level simulation in order to predict any distortions and residual stress.
Majmudar showed a demonstration of NIST’s AM Bench Challenge. Different process parameters were changed in three cases, resulting in different shapes of the melt pool. He also showed a demonstration of a mesoscale model of a track-level simulation. Nonlinear thermomechanical feeds can be determined by inputting data on the laser power, laser efficiency, laser speed, stripe angle, stripe angle increment, hatch distance, and powder material. This model matched well with experimental temperatures. Majmudar explained that thermal simulations help identify these defects or potential problems in parts. He also showed solidification models where thermomechanical properties can be used to estimate scan-level outputs, which helps to predict dendrite shapes and segregation.
Majmudar emphasized that more material properties are needed for modeling, but these can be expensive and time-consuming to obtain. Another challenge is failure prediction, specifically estimating cracking during a build. Failure during the development of a part results in significant delays in the development process, and it is difficult to understand
whether a new alloy is buildable or will crack. There are also ongoing challenges related to predicting the microstructures of AM parts.
An audience member asked how to couple length scales from microscale to mesoscale. Majmudar responded that the mesoscale simulations track the laser as it moves through the melt pool. This can also be characterized by using a microscale model. Another participant asked if Majmudar performed any validation of the microstructure model. He confirmed that his team regularly does quantitative validations for thermal and mechanical distortions of the parts and discussed how a spatial grain structure model compared to experimental results. Another audience member commented on the distortion model and the quantitative validation: if the support structure were too complex to model, finding trends in the model data could help avoid the problem. In response to a question about how long simulations for the industrial part took, Majmudar explained that the turnaround time was about 24 hours, including 3 to 4 hours for the melt-pool simulation, less than 1 hour for the thermal simulation, and several hours for the full mechanical simulation.
Following the presentations, Paquit, Telgkamp, Majumdar, Teter, and Ricker participated in a panel discussion led by Gennaro. An audience member noted that the other side of verification is making sure the machine is reliable and cannot be corrupted. She asked the panelists whether any of them are also considering approaches toward improving assurance, trust, and security. Teter mentioned that Los Alamos National Laboratory is considering these approaches in its work. While assurance typically means that all of the parts meet the requirements to be certified, not enough may be known about the process, structure, and property requirements to reach this goal. Gennaro discussed doing a risk assessment to help with mass production of parts. Ricker mentioned that there was a workshop (see Williams, 2015) at NIST about cybersecurity for print digital manufacturing in which a speaker had students build sample parts while he hacked their codes and put in defects without their knowledge. Ricker stated that machine hacking is a vulnerability for AM, as it is in all types of cyber-physical systems. He suggested that one could use a separate monitoring system that is independent from the computer and facilities that are doing the build to prevent both systems from being affected by the same hack.
John Turner (Oak Ridge National Laboratory) noted that conventional manufacturing has vulnerabilities as well, particularly when only one domestic supplier exists. He asked the panelists whether there are opportunities for AM to increase overall trust in the supply chain. Telgkamp
stated that a long-term vision is to go from a classical supply chain with specialized suppliers to a system where one of many possible suppliers could be identified to produce a part using AM. Makinde mentioned that GE’s software includes a blockchain feature to ensure the stability of a frozen process. Paquit commented that blockchain is not going to address cybersecurity challenges inside the machine, but sensors may be able to help with that. Telgkamp said that blockchain could be helpful in the future to attach the digital proof of quality to an individual part.
An audience member mentioned the lack of standards for safety of parameters, especially for powders, and asked whether qualification standards exist. Telgkamp said that his group has safety documents in place for mandatory requirements from suppliers. Gennaro stated that the suppliers need to provide a safety data sheet since they best understand the powder and how to use it safely. Teter mentioned that standards for testing the flammability of powders exist, but it can be difficult to find the facilities and resources to perform the tests. Another audience member asked about the possibility of reusing powder. Telgkamp replied that there needs to be a verified process and systematic investigation in place. Ricker and Teter added that water vapor, nitrogen levels, and corrosion are important considerations for powder reuse.
A participant asked whether standards exist to address defects that are rare but catastrophic. Paquit answered that sensing may be a short-term solution to avoid issues that result from defects. Teter mentioned that he thinks about the critical flaw size and location of common defects since some areas are more sensitive to defects than other areas within the part.
Teter asked about the use of model validation for instances when researchers can predict a result such as a mechanical property but cannot change any parameters. Majmudar stated that this question leads to discussions of variability in the process, which could also help researchers better understand measurement errors and common causes for variability. Ricker added that many tests are currently required to assess variability for qualification, and models can help understand variability and build trust in the systems. Teter emphasized the importance of representing the underlying physics of the materials and mechanical properties in machine learning models to increase the meaningfulness of possible predictions.
The panelists elaborated on the use of experimental data and modeling for calibration and qualification in response to a question from the audience. Gennaro stated that his team uses experimental data to help with calibrations. Teter emphasized that modeling is helpful in the qualification process, particularly with understanding which parameters are most sensitive to part quality. Modeling can help guide and focus the experimental efforts.
Gennaro asked how long it would take to certify the in-situ monitoring approaches to generate one part of production if computerized tomography (CT) scans were involved. An audience member noted that CT measurements are not used on their own because they are not traceable, unlike other measurements used in certification.
An audience member noted that more data are not always better for certification. Ricker agreed, particularly for data collected early in the process that may not be as relevant to the final part. Teter stated that data from a part with a known defect could be compared to data for other parts to help understand the impact of the defect, but this comparison depends on how well the sensors measure important parameters. Paquit added that his team stores a large amount of data to help address future questions and that it is important to have diversity in the data. Another participant commented that it is important to learn how to use these data to support decision making; in the future, hopefully all data will be usable for production.
Williams, C.B. 2015. “An Analysis of Cyber Physical Vulnerabilities in Additive Manufacturing,” Pp. 21–50 in Proceedings of the Cybersecurity for Direct Digital Manufacturing (DDM) Symposium (C. Paulsen, ed.). NISTIR 8041, National Institute of Standards and Technology, Gaithersburg, Md.