1 6 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 conventionally developed allowables (i.e., ML allowables were just as accurate as conventionally developed ones). These projects also included a validation portion, which included a performance assessment of the ML allowables against the conventional allowables, as well as a cost-benefit assessment. While the word “allowable” is used in this article, the authors wish to highlight an important caveat, which is that no true allowables were generated in either of the two projects discussed. First, the term “ML allowable” is used for convenience, however it should be noted that an ML-based approach is not an approved methodology for allowable development. Second, due to cost and programmatic constraints, several simplifying decisions needed to be made in generating the conventional allowables based on MMPDS or CMH-17 guidelines (e.g., only one lot of powder was used in each project). Project Steps. The project steps for the two projects are summarized in Table 2. Results and Discussion. The ML approach is vastly different from the conventional MMPDS or CMH-17 approaches to materials allowables. To illustrate, imagine a parameter space, such as the three-dimensional parameter space in Fig. 2, consisting of parameters A, B, and C. The ML approach can be applied to n-dimensions and is particularly suited for high- TABLE 2 — ADDITIVE MANUFACTURING ML ALLOWABLES PROJECT SUMMARIES U.S. Army funded (AMMP Consortium) project developing ML-allowables on stainless steel 17-4 PH America Makes project developing ML-allowables on fire retardant polymer Machine and material Machine: EOS M290 Material: Stainless steel 17-4 PH Machine: 3D Systems sPro60 Material: Nylon 11 flame retardant (FR-106) Step 1: Build and collect training data to develop ML model ML software was used for the DOE (design of experiments). 293 vertical coupons over 3 builds. All coupons were built on a single AM machine. Each coupon was a different parameter set. ML software was used for the DOE (design of experiments). 6 builds of 50 coupons each (i.e., 300 coupons total). Half of the builds were on machine 1, half were on machine 2. Each coupon was a different parameter set. Step 2: Select two optimized parameters based on two different engineering requirements ML model was the basis from which two optimized parameters were selected. Parameter set A is optimal to achieve requirement A. Parameter set B is optimal to achieve requirement B. ML model was the basis from which two optimized parameters were selected. Parameter set A is optimal to achieve requirement A. Parameter set B is optimal to achieve requirement B. Step 3: Calculate ML allowables ML model was used to calculate ML allowable A at parameter set A and ML allowable B at parameter set B. ML model was used to calculate ML allowable A at parameter set A and ML allowable B at parameter set B. Step 4: Develop conventional allowables Followed MMPDS S-basis guidelines: Parameter set A: Three builds with 10 coupons per build (30 coupons total) Parameter set B: Three builds with 10 coupons per build (30 coupons total). Followed CMH-17 B-basis robust sampling guidelines: Parameter set A: 10 builds of five coupons each (50 coupons total) Parameter set B: 10 builds of three coupons each (30 coupons total). Step 5: Calculate conventional allowables Based on MMPDs guidelines, calculate S-basis allowable A and S-basis allowable B. Based on CMH-17 guidelines, calculate B-basis allowable A and B-basis allowable B. Step 6: Build validation build (i.e., previously unseen data) Four representative parts total. 60 witness coupons: 30 built with ML-selected parameter set A, 30 built with ML-selected parameter set B. Eight representative parts total. 38 witness coupons total: 15 built with ML-selected parameter set A, 15 built with ML-selected parameter set B. Step 7: Analysis and comparison Accuracy and usability of ML allowables A and B was compared against those of MMPDS S-basis allowables A and B using data from the validation build. Accuracy and usability of ML allowables A and B was compared against those of CMH-17 B-basis allowables A and B using data from the validation build.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==