1 5 A D V A N C E D M A T E R I A L S & P R O C E S S E S | A P R I L 2 0 2 3 (applications that are most mature today) are AM process optimization and certain aspects of process management and control. Once an ML model is trained, AM users can interrogate the model with various questions, such as “what process parameter set should be used if we require ultimate tensile strength of Y?” or “what is the tradeoff between the different process parameter inputs?” or “which process parameter inputs have a large influence on part density?” This allows the AM user to rapidly develop optimized process parameters. Additionally, by monitoring the AM manufacturing process through testing coupon samples at regular intervals, the model can be further trained to detect process drift as a function of time or other factors (e.g., room temperature, humidity, personnel). More sophisticated but far less mature applications for process management and control could come from applying ML to real-time sensory inputs or measurements, such as using computer vision to monitor in-situ sensors, feature engineering of time-series based measurements or training a model to identify features in visual input (e.g., CT scans, microstructure images). In the medium to long term, ML could be used to develop methods for allowable calculations and assist in qualification, requalification, or delta qualification. While the concept of ML allowables is novel, it could be useful today as a “gate check” to help AM users decide whether or not a certain process is stable enough to warrant the time and resources needed to develop conventional allowables using the Metallic Materials Properties Development and Standardization (MMPDS) or Composite Materials Handbook-17 (CMH-17) methods. In other words, ML allowables today could be an estimate of conventional allowables yet to be developed. The ML approach could also be used to demonstrate equivalency, which would greatly aid with requalification or delta qualification. The ML approach allows the issue of equivalency to be easily inverted. Instead of fixing the process parameters and expecting future AM machines to achieve the same requirements with a frozen set of process parameters, the AM user can fix the requirements and ask the ML model to determine what process parameter window on the new AMmachine would allow the AM user to achieve the desired requirements. ML MATERIALS PROPERTY ALLOWABLES DEVELOPMENT Introduction. Senvol recently completed two programs that focused on demonstrating an ML-enabled approach to support materials allowables development. The first project was an Army program [funded via the Advanced Manufacturing, Materials, and Processes (AMMP) consortium] focused on stainless steel 17-4PH. Project members included Senvol, Lockheed Martin Missiles & Fire Control, EWI, Pilgrim Consulting, and Battelle. The second program was an America Makes program focused on a flame retardant polymer, where the project members were Senvol, WSU-NIAR, Northrop Grumman, Stratasys Direct Manufacturing, and Pilgrim Consulting. Both projects completed a sideby-side comparison that evaluated an ML-enabled approach to allowables development. Results showed that an ML-based approach can be more flexible, cost-effective, time-effective, and equivalent to the conventional (e.g., MMPDS in the case of metals, and CMH17 in the case of polymers) approach to materials allowables calculation. Despite the potential that AM offers, the rate of AM adoption is very slow due in part to the high cost and time associated with material allowables development. Furthermore, AM is an advanced manufacturing technique that is process-intensive by definition; the creation of thematerials and the part occurs in the same process. As such: • Conventional materials allowables development binds the user to a limited set of machines and build parameters. • The current allowables paradigm freezes the technology and user in time. • Deviations or multiple allowables require generation of large amounts of additional data. This results in an AM process that is not only costly and time-consuming to implement the first time, but equally costly and time-consuming to maintain in the long run when there are inevitably changes to the AM process. There were two primary objectives of these two projects. 1. Develop and demonstrate a new approach to calculate materials allowables that is not a fixed-point solution. • The projects developed an approach to AM allowables that leverages the digital nature of AM and leverages machine learning (ML). 2. Demonstrate an ML-enabled approach to statistically substantiating materials property predictions across an entire parameter range. • An ML approach is extremely flexible and is able to handle any change to the AM process, thus providing materials property predictions even when deviating from the point at which an allowable was developed. These two projects demonstrated that: 1. An ML approach enables a user to do parameter development and materials allowables development in parallel using the exact same empirical dataset. 2. An ML approach enables a user to make statistically substantiated predictions about performance and scatter everywhere in a given parameter range. This is particularly useful if a user needs to make parts using different parameters (e.g., one parameter set for performance reasons, and a different parameter set for efficiency/cost reasons). 3. ML allowables predicted materials behavior consistently with the
RkJQdWJsaXNoZXIy MTYyMzk3NQ==