| About this Abstract |
| Meeting |
2022 TMS Annual Meeting & Exhibition
|
| Symposium
|
Additive Manufacturing Fatigue and Fracture: Developing Predictive Capabilities
|
| Presentation Title |
J-7: Monitoring Additive Manufacturing Process Stability with Bayesian Changepoint Detection on High-Throughput Tensile Test Data |
| Author(s) |
Stefan L. Colton, Brad L. Boyce, Aaron Stebner |
| On-Site Speaker (Planned) |
Stefan L. Colton |
| Abstract Scope |
Additive Manufacturing (AM) allows one machine to produce diverse geometries, yet this creates a challenge for Statistical Process Control. Here, we apply ML techniques to optimally predict process changepoints solely from the tensile tests of miniaturized coupons. We analyze a dataset of over one year of high-throughput testing conducted by Sandia National Lab. Features are derived from the stress-strain curves using Principal Component Analysis (PCA) and common tensile properties; careful resampling is shown to be critical for PCA. Bayesian Changepoint Detection (BCD) is applied to determine changepoint probabilities. We find that using a subset of PCA features or tensile properties, both known changepoints can be identified with no false positives. Previously additional data sources were required for this result; thus, we demonstrate the potential of BCD to economically detect process changepoints in AM processes, and of PCA features as an alternative to considering a diverse set of tensile properties. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Titanium |