October AMP_Digital

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 | O C T O B E R 2 0 1 9 1 9 The multi-tiered platform inte- grates a set of cutting-edge technolo- gies aimed at efficient, scalable storage and linking all pertinent data, as well as at the management and deployment of complex analytics and models against this linked data. The top tier of the plat- form offers user interfaces that enable additive domain experts to interactive- ly explore, visualize, and analyze data, while absolving them of the responsi- bility to understand any of the under- lying data storage infrastructure. The bottom tier is the data storage layer, which provides federated access to di- verse, heterogeneous, multimodal data (including image data such as optical micrographs and numerical data such as process parameters and quantified defects) captured across multiple dis- parate data stores for storage and re- trieval efficiency. To enable domain experts and data scientists to request “just the data” of interest from multi- ple such sources, the platform’s middle tier leverages ontologies [17] (a formal, human-readable and computer-inter- pretable representation of knowledge within a domain). Ontologies enable connecting disaggregated datasets (i.e., material chemical composition, process parameters, defects, etc.) into one unified AM knowledge graph, en- abling persons not expert in software to explore the data and seamlessly feed it to several model-based approaches and optimization routines. The platform enables direct de- ployment of data-driven models for process parameter optimization and alloy design, such as GEBHM and GE- IDACE. Upon configuring an analyt- ics’ input and output variables via the platform’s user interface, invoking the model build pipeline is just a button click away. USE CASES Build rate optimization for LPBFAM . The development of AM machines is an important focus area as the technolo- gy is adapted to produce larger compo- nents and deliver higher throughputs. The application of LPBFAM is still rel- atively confined due to its generally low build productivity compared with higher throughput additive modalities. Increasing layer thickness beyond typi- cal 20 to 50 µm thicknesses and subse- quent increases in build rates is a key enabler to fabricate larger components via LPBFAM. Also, increases in layer thickness can enable the use of broad- er particle size distributions (PSDs) be- yond the generally accepted 15 to 45 µm diameter range. Until recently, the use of larger layer thicknesses (which also has po- tential to use larger PSDs), has not yet been systematically researched. In the current work, larger layer thicknesses (70 and 100 µm) and larger PSD rang- es (nominally 15 to 75 µm) are explored for a CoCrMo alloy, in which the LPBFAM process parameters for bulk sections (i.e., laser power, laser scan speed, hatch spacing, and laser beam spot size) are optimized toward part density as a first-order screen of LPBFAM com- patibility. It is important to note that layer thickness is defined as the incre- mental shift of the build plate during LPBFAM, whereas the effective powder layer thickness is determined by the powder packing efficiency. In coupons similar to those in Fig. 3a, both layer thickness and build parameters were modified at incre- mental cylinder heights. As-built de- fect quantification was completed via automated optical imaging on the full transverse cross section of each 0.5 in. diameter by 0.25 in. high coupon seg- ment and reported as area percent- age of defects, as described previously. GEBHM and GE-IDACE protocols were then used to actively guide each subse- quent set of experiments. The box plots shown in Fig. 4a illustrate defect distri- butions obtained at layer thicknesses ranging from 20 to 100 µm for two PSDs of 15 to 45 and 15 to 75 µm, wherein the process parameters tested (excluding layer thickness) were otherwise kept constant for comparison. This data sug- gests that as layer thickness increases from 20 to 100 µm, total defect content decreases for the broader PSD of 15 to 75 µm. Additionally, the 100 µm layer thickness (an effective layer thickness of ∼ 190 µm for 0.52 powder packing ef- ficiency) appears to accommodate all large particles in the powder bed. Fig- ure 4b shows a normalized plot of the experimental data collected at a 100 µm layer thickness during the study. Ultimately, build parameters were optimized such that build rate improve- ments up to 1.75 times were achieved while maintaining acceptable levels of as-built defects in a 400 W LPBFAM sys- tem. The improvement was achieved within two experimental iterations. Similar concepts of active learning us- ing both exploitation and exploration capability of the GE-IDACE methodol- ogy were used in other applications showing significant improvement in the parameter development-cycle time ef- ficiency in LPBFAM, and have been re- cently published [12,15] . Fig. 3 — Experimental sequence to characterize additive builds: (a) multi-segment LPBFAM pins are automatically segmented, (b) defects are quantified and categorized using shape filters, and (c) image analysis results are visualized via contour plots or response surfaces. (a) (b) (c)