Stay current on the latest from America Makes.*
*Submission is for announcements only and does not include the bi-weekly members-only newsletter, AM Digest.
Project reduces engineering and qualification cost for 3D printing aerospace legacy parts made from Ti64
Established causal, physics-based models that accurately predict the performance of additively manufactured (AM) parts should drastically reduce non-recurring engineering costs. Without these models, the development and qualification of AM processes and parts will continue to be done on a point basis, with each variant being treated independently. There is a high demand for more general approaches because of the significant cost associated with qualification/certification ($>1M per part).
Ultrasonic additive manufacturing (UAM) uses sound to merge layers of metal drawn from foil stock. The process produces true metallurgical bonds with full density in a variety of metals. Understanding the mechanical performance of UAM, however, is limited in comparison to more mainstream additive processes. Some performance data exists, but most of the data has been funded through commercial sources and is considered proprietary to the entities who have funded the testing. This lack of public data has slowed the adoption of this unique solid-state metal 3D printing process.
The purpose of this program was to develop a process, thermal, structure property (PTSP) model for single-pass Ti64 wall built with laser blown-powder directed energy deposition (DED-LB) that predicts alpha lath size, yield strength, and elongation as a function of process parameters. The goal was for the model to be explainable and include uncertainty quantification (UQ). The explain ability of the model was based on its ability to flow through a predictive network based on physical changes occurring through the build and post-processes. Namely, predictions flowed from build parameters to thermal features to structural properties to strength properties. Another goal was for the model to be assessed based on its accuracy in predicting the experimental measurements and by comparing its performance against an exclusively data-driven approach.
The program followed a proven data science approach. First, data from the Featured-Based Qualification (FBQ) ingested, assessed, and used as a basis for the draft data management plan. Second, a PTSP model was architected in collaboration with EWI’s material science and structural integrity experts. Finally, the model was applied to the data, assessed, and deliverables generated. EWI was the sole participant in this project. The data acquired from the FBQ program was parsed and converted into a pandas data frame. Next, the data of interest associated with single pass walls was isolated, including process parameters, alpha lath size, and
tensile testing results. Structural properties were tied to strength properties based on specimen ID and extraction location. Finally, basic data investigation was performed, including correlation analysis, pair plots, and data completeness checks. EWI followed its standard procedure for developing hierarchical hybrid machine learning models. The relationships in these models developed by EWI’s AM, Material Science, and Structural Integrity Leads, as well as by observed correlations within the dataset. The process parameters and thermal history relationship leveraged EWI’s multi-source Rosenthal model for time-efficient thermal
predictions. Two methods were used to assess the model. First, standard leave-one-out cross-validation was used to assess the accuracy of the hierarchical hybrid machine learning model. Second, the model was compared against a baseline. The baseline was a single-level, exclusively data-driven regression model which links process parameters (e.g., laser power, layer thickness, contour speed) directly with tensile results (e.g., yield and tensile strength). Both modeling approaches were r^2.
The first highlight of this project was using data gathered from a previous America Makes Project 3011 FBQ program that studied the tensile properties of Ti64 DED-LD for different canonical feature types with different build strategies and process parameters. Because the project team was able to leverage
previous data, this project accomplished all its tasks within a short time schedule. EWI proved that using a physics-based PTSP model versus the traditional process property model development can accurately predict the outcome of a printed Ti64 wall structure. The EWI Windows GUI software program can be used to predict the yield and tensile strength with high confidence based on certain process parameters.