| About this Abstract |
| Meeting |
MS&T24: Materials Science & Technology
|
| Symposium
|
Computational Materials for Qualification and Certification
|
| Presentation Title |
Using Unsupervised Learning to Cluster Fatigue Life Based on Ti64 Fatigue Fracture Surface Characteristics |
| Author(s) |
Katelyn Jones, Paul A. Shade, Reji John, Patrick Golden, Elizabeth A. Holm, Anthony D. Rollett |
| On-Site Speaker (Planned) |
Katelyn Jones |
| Abstract Scope |
This work collects scanning electron microscope (SEM) images of Ti-6Al-4V fatigue fracture surfaces and applies pre-trained convolutional neural networks (CNNs) to build a dataset of fatigue fracture surface images and make a connection between them and fatigue life. Round bar specimens at various stress levels and fatigue lives, but corresponding load ratios have been imaged in regular intervals with an SEM. The images capture the crack initiation site, short and long crack, and fast fracture regions at multiple magnifications to determine which length scale allows the machine learning algorithms to infer physically meaningful information. The images are used to fine-tune pre-trained CNNs and compared with unsupervised machine learning methods to determine if fracture surface alone can be linked to the fatigue lifetime. The images taken, the algorithms used, identified fatigue properties, and fracture characteristics will be presented. |