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 |
A Machine Learning Based Damage Constitutive Model for Short Fiber
Reinforced Polymer (SFRP) Composite |
Author(s) |
Subrat Kumar Maharana, Ganesh Soni, Mira Mitra |
On-Site Speaker (Planned) |
Subrat Kumar Maharana |
Abstract Scope |
Short-fiber reinforced polymer composites (SFRPs) have gained significant interest due to their high stiffness-to-density ratio, durability, and cost-effective manufacturing of intricate parts. This study proposes a surrogate damage constitutive model employing an artificial neural network (ANN) to enhance the understanding and prediction of SFRP behavior. Finite element (FE)-based RVE simulations are used to simulate the behavior of the SFRPs. An SFRP model consisting of elastic glass fibers within a polyamide matrix, employing the Drucker-Prager damage model, is adopted. Cohesive zone modeling (CZM) is introduced to capture damage initiation and propagation at the fiber-matrix interface. Traditional damage models like the Drucker-Prager and Jonson-Cook models are helpful but limited by their phenomenological nature and inability to encompass all critical parameters accurately. Hence, a two-step machine learning approach to develop a surrogate constitutive model is adopted to address this limitation. Firstly, an ANN-based damage model is trained using RVE simulation data, where input parameters varied during FE analysis serve as features and RVE responses act as target values. In the second step, the response of the ANN model is compared with experimental data obtained under the same input conditions. A constrained optimization is implemented to minimize the difference between the two responses, where the weights are updated for improved accuracy. Furthermore, the robustness of the model is compared with additional experimental results in the literature under diverse scenarios. This comprehensive methodology facilitates the development of an accurate surrogate model for predicting SFRP behavior, encompassing critical parameters and experimental observations effectively. |
Proceedings Inclusion? |
Definite: Post-meeting proceedings |