| 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 |
Unsupervised Learning of Damage Modes During Fatigue of Composites |
| Author(s) |
Partha Pratim Das, Monjur Morshed Rabby, Vamsee Vadlamudi, Rassel Raihan, Muthu Ram Prabhu Elenchezhian |
| On-Site Speaker (Planned) |
Vamsee Vadlamudi |
| Abstract Scope |
Fiber reinforced polymer (FRP) composites are used in abundance in load bearing structures due to their exceptional durability while exhibiting resistance to corrosion and being lightweight. However, repetitive loading, such as fatigue, can cause progressive damage accumulation in the composite parts. If the damage evolution is not monitored for timely maintenance, the part can experience catastrophic failure.
Previous studies have shown that Broadband Dielectric Spectroscopy (BbDS)/Impedance Spectroscopy (IS) can effectively track changes in dielectric permittivity during damage progression in FRP composites. By utilizing this data, supervised artificial neural networks (ANN) can predict the life (durability) and remaining useful life (RUL) of these composites under fatigue loading. However, supervised learning methods, which rely on historical time domain data from run-to-failure tests, may be limited in identifying various damage modes (such as initiation, accumulation, and interaction) in composite parts lacking such historical data. In this work, we present an unsupervised machine learning (uML) model, that can detect different damage modes during fatigue loading of FRP composites from in-situ time series dielectric state variable data. The uML model eliminates the need of having any historical knowledge about the relationship between the damage progression and dielectric evolution, yet provides an accurate understanding and identification of present damage modes. |
| Proceedings Inclusion? |
Definite: Post-meeting proceedings |