.png){kind=icon}
.png){kind=icon}
Return
A force neural network framework for structural optimization
.png){kind=icon}
PRE
AI
DOI：10.1016/j.engappai.2024.109991.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In this paper, an efficient Force Neural Network (FNN) is developed to reformulate the size optimization of truss structures as an operator learning problem. A Deep Neural Network (DNN) is designed to directly map the connectivity information of truss members to the corresponding design variables. Therein, the entire unlabeled training data contains only the connectivity information of members, without any structural responses, weights, or cross-sectional areas. By integrating Force Method (FM), our framework embeds the optimal design problem represented by the objective and constraint functions in the loss function to guide the training process. And it guarantees that the generated solution is consistent with the underlying physical principles. In addition to enhance efficiency in finding the optimum structural weight, Bayesian Optimization (BO) is applied for automatic hyper-parameters tuning instead of the trial and error method. As soon as the training phase ends, the optimal weight of truss structures is found without using any other numerical methods. Several numerical examples are investigated to demonstrate the effectiveness and applicability of the FNN for the optimization of truss structures. The obtained results indicate that it not only be simple to perform but also overcomes the local optimal problem and reduces the computational cost in high-dimensional problems.
Keywords:
Force neural network
Deep neural network
Structural optimization
Auto-tuning hyper-parameters
Bayesian optimization
Journal
IF:
8
Papers:
5.4K
Citations:
3.5W
Organization
View more organizations