| 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 |
Benchmark Machine Learning Models in Design Optimization of Tow-Steered Composite Structures |
| Author(s) |
Bangde Liu, Xin Liu |
| On-Site Speaker (Planned) |
Bangde Liu |
| Abstract Scope |
Tow-steered composites are new material systems that can be tailored to achieve curvilinear fiber orientations for optimal load paths to enhance structural performance. One challenge for developing tow-steered composite structures is the computational cost of optimizing fiber paths, considering the vast design space. Although many machine learning (ML) models have been developed for tow-steered composites, the performance of various ML models with the increasing design complexity in tow-steered composites has not been evaluated. In general, advanced ML, such as the neural network (NN) model, requires a more extensive training dataset to achieve the desired accuracy. This may not result in optimal efficiency when handling a small number of design variables. This work will test the accuracy and efficiency of several ML models, such as the support vector regression, Gaussian process regression, and NN model, with an increasing number of design variables and structural complexity. A plate structure and a shell structure with tow-steered composite designs will be employed to generate simulation data. The corresponding costs for generating simulation data will also be considered in the performance of ML models. The trained models will also be used to perform design optimization of the two structures to maximize the critical buckling load. The accuracy and efficiency of the optimization analysis will be utilized as the additional metrics for evaluating the ML models. |
| Proceedings Inclusion? |
Definite: Post-meeting proceedings |