A data-driven model for predicting the mixed-mode stress intensity factors of a crack in composites

•Apply Artificial Neural Networks (ANN) for predicting stress intensity factors in heterogeneous composites.•Investigate the significance of parameters through feature engineering techniques.•Implement K-fold cross-validation and Bayesian optimization to enhance model performance.•Leverage an active...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Engineering fracture mechanics Ročník 288; s. 109385
Hlavní autori: Zhang, Xinyu, Zhao, Tingting, Liu, Yifan, Chen, Qingqing, Wang, Zhiyong, Wang, Zhihua
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: Elsevier Ltd 04.08.2023
Predmet:
Active learning
Bayesian optimization algorithm
Composites
Data driven
Stress intensity factors
Data driven
Composites
Stress intensity factors
Active learning
Bayesian optimization algorithm
ISSN:0013-7944, 1873-7315
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:•Apply Artificial Neural Networks (ANN) for predicting stress intensity factors in heterogeneous composites.•Investigate the significance of parameters through feature engineering techniques.•Implement K-fold cross-validation and Bayesian optimization to enhance model performance.•Leverage an active learning framework to reduce data dependency and increase predictive accuracy. A data-driven model is trained to predict mixed-mode stress intensity factors (SIFs) of composites through an artificial neural network (ANN) method. The model is based on a database generated by combining the interaction integral and the extended finite element method (XFEM). To reduce the input dimensionality and improve predictive performance, feature engineering is performed on the input data, and principal component analysis is conducted. Hyperparameters of the model are adjusted by using K-fold cross-validation and Bayesian optimization algorithm (BOA) to enhance the adaptability and generalization of the model. To overcome the data scarcity issue, an active learning knowledge extraction framework is constructed, which allows for accurate knowledge extraction even with limited data. By utilizing data-driven models to solve mixed-mode SIFs, the computational cost and complexity are greatly reduced compared to numerical simulations, while computational stability and ability to deal with high-dimensional nonlinear problems are significantly improved.
ISSN:0013-7944
1873-7315
DOI:10.1016/j.engfracmech.2023.109385