About this Abstract |
Meeting |
2024 ASC Technical Conference, US-Japan Joint Symposium, D30 Meeting
|
Symposium
|
2024 ASC Technical Conference, US-Japan Joint Symposium, D30 Meeting
|
Presentation Title |
Development of Machine Learning Based Method for Prediction Complete Delamination Morphology |
Author(s) |
Paul Davidson |
On-Site Speaker (Planned) |
Paul Davidson |
Abstract Scope |
Current ultrasound technology when applied to quantify Low velocity impact (LVI) can reveal the presence of some ply delaminations; however, the extent and morphology of damage in the hidden region—the lower portion of damage occluded during single-sided normal-incidence inspection by preceding delaminations —cannot be determined, therefore only providing an incomplete damage mapping. Such information, though important, is insufficient for risk evaluation, for example, the risk of fatigue damage growth or knockdown in compression after impact.
In this talk, we present machine learning methods to predict complete ply-by-ply delamination pattern of a BVID composite plate given partial information from UT Scan and information of about the lamination stack. Approach used is to train Machine Learning (ML) based in-filling techniques using a combination of sparse “real data” and “simulated data”, to create a hybrid dataset for training. UNet based segmentation and Generative Adversarial Networks (GAN) will be used as ML models. Initial results have shown that the approach is capable to distinguish different delamination patterns from a noisy UT scan data. |
Proceedings Inclusion? |
Definite: Post-meeting proceedings |